Conventional gaze tracking systems are limited in cases where the user is wearing glasses because the glasses usually produce noise due to reflections caused by the gaze tracker's lights. This makes it difficult to locate the pupil and the specular reflections (SRs) from the cornea of the user's eye. These difficulties increase the likelihood of gaze detection errors because the gaze position is estimated based on the location of the pupil center and the positions of the corneal SRs.
In order to overcome these problems, we propose a new gaze tracking method that can be used by subjects who are wearing glasses. Our research is novel in the following four ways: first, we construct a new control device for the illuminator, which includes four illuminators that are positioned at the four corners of a monitor. Second, our system automatically determines whether a user is wearing glasses or not in the initial stage by counting the number of white pixels in an image that is captured using the low exposure setting on the camera. Third, if it is determined that the user is wearing glasses, the four illuminators are turned on and off sequentially in order to obtain an image that has a minimal amount of noise due to reflections from the glasses.
As a result, it is possible to avoid the reflections and accurately locate the pupil center and the positions of the four corneal SRs. Fourth, by turning off one of the four illuminators, only three corneal SRs exist in the captured image. Since the proposed gaze detection method requires four corneal SRs for calculating the gaze position, the unseen SR position is estimated based on the parallelogram shape that is defined by the three SR positions and the gaze position is calculated. Experimental results showed that the average gaze detection error with 20 persons was about 0.70° and the processing time is 63.72 ms per each frame.
Proposed Gaze Tracking System with the Device for Controlling Four Illuminators In our study, we propose a new gaze tracking system with a device that controls four illuminators. Our gaze tracking system is based on a wearable device that includes a lightweight eye capturing camera and is used in a desktop computer environment as shown in. A conventional web-camera with a zoom lens of fixed focal length and a universal serial bus (USB) interface is used for the eye capturing camera. The field of view of the eye capturing camera is −16.98°∼+16.98°. Proposed gaze tracking system.
Since the pupil area is usually distinctive in images that are captured by a near-infrared light (NIR) illuminator with a wavelength of 850 nm, the NIR cut filter (in the eye capturing camera) which passes the visible light is replaced with an NIR passing filter. Four NIR illuminators are attached at the four corners of the monitor as shown in. Each illuminator includes 32 NIR light emitting diodes (LEDs) with wavelengths of 850 nm. These four illuminators generate four corneal SRs and the quadrangle defined by these four SRs represents the monitor region. We also constructed a device for controlling the four illuminators as shown in. The device is constructed using a USB relay board and it can be turned on and off selectively turn by controlling the power supply to the illuminator. That is, our gaze tracking program in a desktop computer determines whether the illuminator should be on or off and sends the command to the USB relay board via the USB interface.
Overall procedure for the proposed gaze tracking method. In previous research, Wu et al., proposed a method for detecting glasses using Haar and Gabor features based on boosting methods. However, they used the images where the entire face area and glasses were included for training and testing. In addition, both the Haar and Gabor features selected in the first boosting stage were detected in the area between the pair of eyes, which show that the nosepiece of the frame of the glasses was the important feature for detecting the glasses. However, the nosepiece is not included in the image that is captured by our gaze tracking system, and the part of glasses frame may not be seen in the image, as shown in. Hence, the method in cannot be used for our study. Eye images that were captured at various exposure times ( a) Image of naked eye at the normal (unreduced) exposure time; ( b) Image of eye with glasses at the normal (unreduced) exposure time; ( c) Image of naked eye of (a) at the reduced exposure time.
Instead, the initial check that determines whether the user is wearing glasses or not is performed as follows. Firstly, the exposure time of camera is reduced and an image is acquired using the eye capturing camera in. In general, if a user wears glasses, many reflections occur from the surfaces of the glasses as shown in. Since the shapes and sizes of the reflections vary, it is difficult to discriminate these reflections from reflections that are caused by the skin.
In order to solve this problem, our system reduces the exposure time of the camera. Conventional cameras usually accumulate the light on the camera sensor during a 33.3 ms interval when the exposure time is set to 1/30 s. If the exposure time is reduced to 1/60 s, the time interval during which the camera accumulates the light is reduced to 16.7 ms (33.3/2). In general, the brightness of the reflections from the surfaces of the glasses is higher than the brightness of reflections from the skin because the reflection rate from the glasses is higher than that from the skin. As a result, the reflections from the skin cannot be seen when the exposure time of the camera is reduced as shown in. In this image, which was taken at a lower exposure time, the number of white pixels is counted within a predetermined area in the captured eye image (the red-colored box in ) because the eye is usually positioned in the restricted area by the device in.
If the number of white pixels exceeds a certain threshold (200), our system determines that the user is wearing glasses as shown in. If it is determined that the user is wearing glasses, our system increases the exposure time (like ) to the normal exposure time (like ), and turns of the 1st illuminator. If all four of the NIR illuminators are on, they frequently cause reflections from the surface of the lens, as shown in, and it is very difficult to detect the regions of the pupil and the four genuine corneal SRs. Examples of eye images where the reflections hide the pupil or the corneal SRs. As a result, our system turns the illuminators on and off sequentially.
As shown in, the four NIR illuminators are attached to the four corners of monitor and we designated the upper-left, upper-right, lower-left, and lower-right illuminators as the 1st, 2nd, 3rd, and 4th illuminators, respectively. Our system turns off the 1st illuminator and captures an eye image. If the number of white pixel exceeds the threshold (15,000) within the pre-determined area of the eye image (the red-colored box of ), our system determines that many reflection noises still exist in the image with the 2nd, 3rd, and 4th illuminators.
Thus, it selects a different illuminator to turn off. Accordingly, the 2nd illuminator is turned off and the other three illuminators (the 1st, 3rd, and 4th ones) are turned on. Another eye image is captured with these illuminators turned on and the number of white pixel is counted within the pre-determined area of the eye image (the red-colored box of ). If it exceeds the threshold (15,000), our system determines that many reflection noises still exist in the image and changes the illuminator that is turned off. The same procedures are repeated with the 3rd and 4th illuminators.
If the number of white pixels is less than the threshold (15,000) in one of the resulting images, our system determines that the number of reflections is low enough because the corresponding illuminator has been turned off. At this point, the additional procedures for detecting the pupil and the corneal SR positions are performed and the final gaze position is calculated as shown in. In order to cope with the worst case of an infinite loop ( i.e., the number of white pixels exceeds in threshold in all the cases), we include a stopping condition based on the number of trials as shown in. If the trial number is greater than the threshold, our system displays a message to the user that says, “Please, take off your glasses”, and the gaze tracking system restarts.
We set the threshold at 1. Examples of reflections as the illuminators are turned on and off in sequence ( a) All four of the illuminators are on; ( b) Only the upper-left illuminator (the 1st illuminator) is off; ( c) Only the upper-right illuminator (the 2nd illuminator) is off. In order to accurately measure the effect of the reflections on the pupil region or the corneal SR, the number of white pixels should be counted in the detected eye region. However, a conventional eye detection algorithm based on the Adaboost method does not give good performance for eye detection for images that include reflections as shown in. The green box in represents the eye detection region and in the top-right image in, there is no area that is detected by the Adaboost method. From these images, we can confirm that the Adaboost method cannot locate the eye region in images that include reflections inside the eye area. Thus, it is difficult to determine the actual eye region.
Examples of incorrect detections of the eye region while using the Adaboost eye detector for images that include reflections (We show the results for images in in the clockwise direction from the top-left image). In our research, we used the Adaboost algorithm already trained, which are provided from OpenCV library (Version 2.4.2) , and we did not perform the additional procedure of training for the Adaboost algorithm. If we perform the training of the Adaboost with the sets including reflections like, its performance of eye detection with the images including the reflections can be enhanced. However, the performance with the images of no reflection can be affected.
In order to solve this problem, the training of the Adaboost should be performed with a lot of images with and without the reflections. In our system, a user wears the gaze tracker device that is shown in. Thus, the eye position in the captured eye image can be restricted within the predetermined area that is shown in and (within the red-colored box).
Based on this restriction, our system can determine whether the reflections have been removed by counting the number of white pixels in the pre-determined area of the image. Shows examples of reflections as the illuminators are turned on and off in sequence. Our system can determine that the image in is best for detecting the pupil and the corneal SRs and calculating the gaze position by comparing the number of white pixels in the pre-determined area (the red-colored box) of images in Because the camera in the wearable eye capturing device acquires the eye image below the eye, as shown in, it is common for the eye region to be in the upper area of the glasses as shown in. In addition, based on the positions of the illuminators, the glasses, and the camera that are shown in the, it is more likely for the SRs on the glass surfaces from the 1st (upper-left) and 2nd (upper-right) illuminators to be close to the eye region than it is for the SRs from the 3rd (lower-left) and 4th (lower-right) illuminators to be close as shown in and. Thus, it is more likely to avoid the SRs by turning off the 1st or 2nd illuminators than it is to avoid the SRs by turning off the 3rd or 4th ones. Example of the calculated gaze positions based on the nine reference points (five trials of one person) (the “°” symbols signify the reference points and the “+” signs are the detected gaze points) ( a) User is not. The procedure of turning off the 1st (upper-left) ∼4th (lower-right) illuminators with image capturing is sequentially performed as shown in.
For each image, if the number of white pixels is less than the threshold, the system determines that by turning off the corresponding illuminator, the number of reflections has been reduced. That is, if the image of satisfies the threshold for the number of white pixels, the systems stops the process of turning off illuminators and capturing images (thereby, not acquiring the images in ). At this point, the procedures for detecting the pupil and corneal SR positions and calculating the gaze position are performed as shown in. Consequently, based on these methods, the system determines that is the best image for the gaze detection process. Examples from the pupil detection process. ( a) Original image; ( b) After erasure of the SR regions; ( c) Image resulting from morphological operations; ( d) Image resulting from histogram stretching; ( e) Pupil area that is detected by the CED method; (. In general, SR areas have high pixel values and sharp changes in the gray values when compared to neighboring non-SR areas.
This characteristic of sharp changes can cause errors in the pupil detection process. Thus, regions in the captured image that have bright pixels with gray levels that are higher than a threshold (200) are roughly estimated as SR regions. Then, these pixels are interpolated using their (left and right) neighboring pixels as shown in. As a result, the bright pixels have the characteristics of smooth changes in their gray values when compared to their neighboring ones. Then, the input image is processed using a morphological operation (the morphological opening is performed two times) in order to remove the reflections and group the regions with similar gray levels as shown in. In general, the pupil area is darker than other regions such as the iris, sclera, and skin. Thus, histogram stretching is performed as shown in in order to increase the differences in the pixel levels between the pupil and other regions.
Then, the CED method is used to locate the approximate position of the pupil in the image as shown in ,. However, the shape of the pupil is usually not perfectly circular. It is usually a more complicated shape.
As a result, it is usually not possible to use the CED method to obtain an accurate detection of the pupil center. Thus, the restricted area of the image of based on the detected pupil center and radius by the CED is binarized as shown in. Morphological erosion and dilation are performed on the binary image in order to remove the isolated reflections as shown in. Then, the image is processed using component labeling, canny edge detection, and the convex hull method as shown in.
Subsequently, the actual pupil area is detected using ellipse fitting and the center of ellipse is designated as the center of the pupil as shown in. The restricted region is binarized based on the detection of the pupil center. The regions whose sizes are smaller than the threshold (20) or bigger than the threshold (600) are removed by component labeling and size filtering processes.
Then, the maximum four regions remained are selected, and the centers of the four regions are determined by calculating the geometric center of each region. In our system, one of the four NIR illuminators is turned off when the user is wearing glasses in order to avoid reflections as shown in. Thus, only three corneal SRs exist in this case. Since the four NIR illuminators are attached at the four corners of monitor as shown in, the quadrangle that is defined by the four corneal SRs represents the monitor region and the positions of these four SRs are required in order to calculate the gaze position.
In order to solve this problem, the unseen SR position is estimated based on the parallelogram shape that is defined by the three existing SR positions that are shown in, which is novel in our research. Calculating the Gaze Position In order to calculate the gaze position in the monitor, we use a geometric transform method that is based on the locations of the center of the pupil and the centers of the four corneal SRs ,. Then, the angle kappa is compensated for by the user-dependent calibration (each user gazes at the monitor center once during the initial stage) ,. From that, the difference between the calculated gaze position and the monitor center is obtained, and it is compensated for calculating the final gaze position. The resolution of the monitor that is used for the calibration is 1280 × 1024 pixels.
Each user is instructed to gaze at the red (filled) circle. In order to induce the user's attention and increase the accuracy of the calibration accuracy, the diameter of the red circle is gradually reduced from 38 pixels to 30 pixels during the calibration process.
3. Experimental Results The proposed method was tested on a desktop computer with an Intel ® Core™ i7 3.5GHz processor (Intel Corporation, Santa Clara, CA, USA) equipped with 8 GB RAM. Our algorithm was implemented using Microsoft Foundation Class (MFC) based C programming, the DirectX 9.0 software development kit (SDK), the library for controllable illumination devices, and the OpenCV library (Version 2.4.2). A 19-inch monitor with a resolution of 1280 × 1024 pixels was used. In the first test, we measured the accuracy of our system determining whether the users were wearing glasses or not (“initial checking whether a user wears glasses” of ). The experiments were performed with 400 images, which were captured from 20 persons. Each person tried 20 times. Our system captured an image using the low exposure setting of the camera during each trial for each of the test subjects.
Out of the 20 participants, 10 wore glasses and the other 10 did not wear glasses. A total of 20 graduate students (whose ages were in the 20s to 30s range) volunteered to take part in the experiments without any payment. There were no restrictions during the selection of participants. Each the 10 persons brought their own glasses. Each pair of glasses that was worn by one of the 10 subjects was different from the others as shown in, and the number of glasses types is 10, consequently.
Shows the examples of captured images. Examples of the images for experiments ( a) Images of people not wearing glasses ( b) Images of people wearing glasses (The top-left image is from user 1, the top-right image is from user 2, and the bottom-right image is from user 10 from. Detailed information about the glasses is shown in and. In and, users 1∼10 correspond to the users in and, and. In and, the spherical strength of the glasses is shown as “S-XXX” where the number “XXX” represents the diopter of the lens. Information about astigmatism of the lenses is also shown in and.
The larger number in “C-XXX” represents the highest degree of astigmatism of the lenses. And we show the kind of lens and glasses frame. We also subjectively evaluated the quality of the glasses surfaces as high, medium, or low as shown in and. The gaze detection errors from the previous method for users who were wearing glasses (units: °). We define Type 1 and Type 2 errors for measuring the accuracy of the proposed method. A Type 1 error means that the test subject was wearing glasses, but it was incorrectly determined that they were not wearing glasses.
A Type 2 error signifies that the test subject was not wearing glasses, but it was incorrectly determined that they were wearing glasses. The experimental results showed that the rate of Type 1 and Type 2 errors was 0%. That is because the two distributions of wearing glasses and not wearing glasses do not have any overlapped area as shown in. Two histogram distributions for not wearing and wearing glasses in terms of the numbers of white pixels.
In the second test, we measure the error of our gaze tracking system. The distance between the monitor and the eyes of participants is about 85 cm. We tested with 20 users. Each person is requested to gaze at 9 reference points on the monitor as shown in. Among 20 persons, 10 people wore glasses, and the other 10 people did not wear glasses. These experiments were repeated five times per each person.
Thus, each person gazes at the 45 gaze positions (9 reference points × 5 times). Each person is instructed to look at the monitor center for the initial user calibration (see Section 2.4), and see the nine reference points (of ) five times. Except for these, no instruction was given. All the participants were allowed to move their head freely. The error of gaze detection is measured as the unit of ° based on the difference between the reference and the calculated gaze points. The difference means the angular disparity of two vectors (one is from the pupil center to the reference point, and the other is from the pupil center to the calculated point). And display the gaze detection errors that occurred for users who were not wearing glasses and for user who were wearing glasses, respectively.
As shown in and, the error for the former group (0.70°) is almost same to that of the latter group (0.70°). As a result, we conclude that our gaze tracking system works irrespective of whether the users are wearing glasses or not. The reason why the errors were greater for users 5 and 8 than for other users in was that these users failed to gaze at the exact center point of the monitor during the initial calibration of the angle kappa that is described in Section 2.4.
The errors in gaze detection for cases where the users were not wearing glasses (units: °). As shown in, the lowest gaze errors were obtained with users 2–4 and the highest gaze errors were for users 5 and 10.
RECOVERY SETTINGS: UMD MODE: Sony NP9660 -NO UMD- FAKE REGION: DISABLED FREE UMD REGION: ENABLED USE VSH MENU: ENABLED CPU SPEED IN XMB and UMD/ISO: 333 1.you must have CUSTOM FIRMWARE 3.52m33-4 or higher 2.you must have ISO in ISO/VIDEO folder 3.press SELECT (3.71m33 or higher) or HOME (3.52m33-4) in XMB (main menu of PSP) to get to the VSH MENU 4.set CPU CLOCK XMB and CPU CLOCK GAME for 333/166 5.set UMD VIDEO MODE to NORMAL 6.select ISO VIDEO MOUNT and select your ISO (with ) 7.Click EXIT or click select from VSH MENU, please wait. All star yakyuken battle psp update download. UMD VIDEO-MOVIE How to Mount a UMD VIDEO ISO? YOU MUST HAVE A UMD DISC inside your PSP (For example: DEMO UMD) if you don't have umd disc you don't play UMD MOVIES on your PSP-sorry.
Based on these results and the analyses of the characteristics of the glasses of and, we found that there is no relationship between the properties of the glasses and the accuracy of the gaze detection process. When the images in are compared, it is apparent that the images of people wearing the glasses include much larger SRs from the glass surfaces. Some of the SRs hide the pupils or the corneal SRs. Nevertheless, as shown in and, the average gaze detection errors for users who are not wearing the glasses is about 0.70° and it is the same for users who are wearing glasses. For the stochastic analysis of the experimental results, we compared the errors in to the errors in and used the t-test , to establish the confidence levels. For the two tailed t-test with the null-hypothesis (the total average error of is same to that of ), we obtained a p-value as 0.9862.
Because the p-values are greater than 0.01 ( i.e., confidence level of 99%), the null-hypothesis fails to be rejected and we can conclude that the total average error for the former case (users not wearing the glasses in ) is almost identical to the total average error for the latter case (users wearing glasses in ) with a confidence level of 99%. Thus, we can conclude that the proposed method solves the problem with the SRs from the surfaces of glasses hiding the pupil or the corneal SRs and that the proposed method obtains accurate gaze positions irrespective of whether the user is wearing glasses or not. And show the gaze detection errors for each of the nine reference points in. The upper-left, upper-center, upper-right, middle-left, middle-center, middle-right, lower-left, lower-center, and lower-right reference points in are the gaze positions for 1–9 in and, respectively. As shown in and, the gaze detection errors at the four positions (close to the four monitor corners) of 1, 3, 7 and 9 seem to be larger than others. The reason for this result is that each user gazes at the center of the monitor during the user-dependent calibration for the angle kappa that is described in Section 2.4.
This calibration process does not provide sufficient information about the angle kappa when the user gazes at positions that are close to the corners of the monitor. The errors of gaze detection along X and Y-coordinates in cases where the users were not wearing glasses (units: °). Shows examples of the calculated gaze positions based on the 9 reference points. The resolution of the monitor is 1,280 × 1,024 pixels and the point that each user was supposed to gaze at is shown as a black (filled) circle with a diameter of 30 pixels. In, the nine reference points are displayed as blue (blank) circles in order to enhance the distinctions between the calculated gaze points (red cross marks) and the reference points. During the experiment, black (filled) circles were actually used as the reference points. It is difficult to compare our method to previous methods because different hardware systems and different methods were used for detecting the pupil and the SRs.
Cornea Monitor Drivers Reviews
As a result, we have opted to compare the accuracy of proposed method with the accuracy of. In , the cross-ratio-based method was used in conjunction with vanishing points in order to calculate the gaze position. In order to construct a fair comparison, the same method was used for the initial calibration, the process for controlling illuminators, the process for detecting the pupil and the SRs. As shown in, and, the experiment confirms that the accuracy levels of the proposed method are higher than those from the previous method. Images for cases where the outer fluorescent light was on or off ( a) All of the NIR illuminators are on and the outer fluorescent light is on; ( b) Only the upper-left illuminator (the 1st illuminator) off and the outer fluorescent light is on; ( c) All. Shows the processing times from our gaze tracking system for each sub-module. We do not include the average processing time (about 84.1 ms) in for the initial check for determining whether the user is wearing glasses because it is performed once only during the initial stage, and it is not performed again after that.
The average processing time (about 0.83 s) for turning off the illuminator and checking for reflections (the procedures from the bottom-left (blue) box in ) is also not included in because it is only performed once. The processing times for sub-modules in the gaze tracking system (units: ms). As shown in, the processing time for the case where the user is wearing glasses is similar to that for the case where the user is not wearing glasses. When the user is wearing glasses, the step of turning off the illuminator and checking for reflections (the procedures of the bottom-left (blue) box of ) is also performed and as a result, the processing time is increased by as much as 0.83 s. However, because this step is only performed once, the processing time for wearing glasses is nearly identical to the processing time for the case where the user is not wearing glasses after this process is completed as shown in. Based on the average total processing time of 63.72 ms, we conclude that our system can be operated at the speed of about 15.7 frames/s (1000/63.72). In order to analyze the influences of the properties of the glasses on the results in a more systematic fashion, we included five additional participants (whose ages are in the 20s) with glasses that were different from those of users 1–10 in and, and.
The images of people wearing glasses are shown in. The characteristics of the glasses of the additional users (users 11–15) are shown in and the gaze detection accuracy levels for these users are shown in. Since the glasses of user 12 do not include the functionality of correcting nearsightedness (myopia), there is no information on the spherical strength of the glasses. The gaze detection errors for users who were wearing glasses (units: °). As shown in, the lowest gaze errors were obtained with users 2–4, and the highest gaze errors were for users 5 and 10. When comparing users 2–4 with users 5 and 10 in and, the (spherical) strength of the glasses of user 2 is similar to that of user 10. Astigmatism correction is not included in the glasses of user 3, while that is included in users 2 and 4.
The lens type of user 5 is a convex lens while that of user 10 is a concave one. The types of glasses frames of users 2–4 are different (plastic, aluminum, and wood ones, respectively).
As shown in, the lowest gaze error was obtained with user 15, and the highest gaze error was for user 14. When comparing user 14 with user 15 in, the lens types of users 14 and 15 are similar concave ones. The types of glasses frames of users 14 and 15 are also similar plastic ones. From this, we found that there is no relationship between the properties of the glasses and the level of the accuracy of the gaze detection process. We can think that the glasses surface of lower quality can usually produce more reflections and the consequent error of gaze detection increases. However, the qualities of glasses surface of users 2∼4 are low while those of users 5 and 10 are high, as shown in and.
From that, we found that there is also no relationship between the quality of the glasses' surface and the level of the accuracy of our gaze detection method. The accuracy levels and frame rates of commercial systems are typically very high.
The frame rates of these systems are usually very high due to the use of expensive, high speed cameras. As a result, the overall costs of these systems are very high. They also tend to be very bulky. For example, the size of Tobii TX300 Eye Tracker is 55 × 24 × 6 cm 3. However, the cost and size of the proposed system are very low because the system is based on a low-cost web-camera.
Although the accuracy of the commercial system was reported as 0.5°, the accuracy level for users with glasses was not reported. The average processing time for the proposed system is 63.72 ms as shown in, but most of the processing time is concerned with detecting the pupil in the 1600 × 1200 pixel image. In order to reduce the processing time, we sub-sampled the original image, obtained an image with 800 × 600 pixels, and performed the pupil detection using the sub-sampled image.
Experimental results with the data from users 11–15 in and showed that the average processing time was reduced by about 23.47 ms (42.6 Hz). The level of accuracy for gaze tracking with the revised method was almost 0.64° as shown in, which was similar to the level of accuracy in and. 4. Conclusions In this paper, we have proposed a new method for tracking the gaze of a user who is wearing glasses.
This method is based on a scheme for controlling the illuminator and estimating the unseen SR position based on the parallelogram shape. Through experiments with the data from 20 test subjects, we were able confirm that our system was effective regardless of whether the test subject was wearing glasses or not. In order to reach a higher level of accuracy during the gaze tracking process, a high resolution image of the eye should be acquired as shown in. A high resolution camera with a zoom lens is required in order to accomplish this.
The viewing angle of the camera in a gaze detection system with a zoom lens will be very small. As a result, a non-wearable (non-head-mounted) gaze tracking system should include functionality for panning and tilting in order to track the eye region based on the natural movements of the user's head.
However, this kind of functionality will cause the size and cost of the system to increase. Therefore, we use a head-mounted (wearable) gaze tracking system that is lightweight and inexpensive. Our system allows the user to move naturally because the camera in our system is attached to the user's head and moves with the user. The image of the eye that is captured by the camera in the head-mounted system is not distorted when the user moves because the camera moves with the user. The image of the eye from the non-wearable gaze tracking system, on the other hand, can be distorted by head movements, which can reduce the accuracy of the gaze detection process. Non-wearable gaze tracking systems are usually more convenient for the user than head-mounted systems, but the inconvenience of our system is reduced through the use of a lightweight frame and a lightweight web-camera. As a result, our system can be used in various applications that require a compact and inexpensive, yet accurate gaze tracking system.
It can be used in desktop computer environments for monitoring the web-surfing patterns of users, measuring the effects of advertisement during web-browsing, and also during driver training or pilot training. We plan to test our system in various environments, including outdoors, in a future study. We also plan to research methods for increasing the processing speed of our system.
Introduction The cornea is a clear layer that covers the front portion of the eye, and plays numerous roles essential for the maintenance of clear vision and ocular health, including contributing most of the eye's focusing power, and helping to shield the rest of the eye from bacteria, dust, and other harmful particles. The structure of the human cornea consists of five layers, from anterior to posterior: corneal epithelium, Bowman’s layer, corneal stroma, Descemet’s membrane, and corneal endothelium. Each of these layers has its own role for maintaining normal visual function. In the stroma, specifically, there has been much work examining collagen fibril assembly and the maintenance of corneal transparency. Here, we summarize the function of corneal stroma from a new perspective: a potential nutritional source for the cornea.
Growth factors and cytokines Communication between the epithelium and the stromal mesenchyme occurs during normal development, and this communication is sustained during adulthood to maintain homeostasis. Epithelial stratification is a complex and precise process that occurs during corneal development, and is highly dependent on this intercellular communication. Have shown that, in conditional knockout of β-catenin in corneal stroma via Kera-rtTA driver mice, growth factor bone morphogenetic protein 4 (Bmp4), released from the stroma, acts on corneal epithelium to trigger its stratification through activation of transcriptional factor p63. Before stratification, Bmp4 is suppressed by Wnt/β-catenin signaling. At the onset of epithelial stratification, Wnt/β-catenin signaling is dampened, leading to the loss of Bmp4 repression, and the subsequent initiation of epithelial stratification.
In addition to Bmp4, expression levels of several other growth factors/cytokines in the stroma were also altered after knockout of β-catenin, but the functions of these growth factors/cytokines have not been yet elucidated. Therefore, this result suggests that the corneal stroma might be a potential reservoir of growth factors/cytokines for corneal development and homeostasis.
The production of growth factors/cytokines in the stroma also contributes to wound healing. After wounding, hepatocyte growth factor (HGF) and keratinocyte growth factor (KGF) are markedly upregulated in stromal keratocytes, while expression of HGF receptor and KGF receptor are simultaneously upregulated in the corneal epithelium.
Stromal keratocyte cell cultures and wounded corneal organ cultures in vitro showed HGF and KGF affected the stratification and differentiation of the epithelium, with HGF delaying and KGF accelerating epithelial coverage of the wound. Further studies have demonstrated that wounded corneal epithelial cells release IL-1 alpha and IL-1 beta, signaling stromal keratocytes to upregulate HGF and KGF to modulate healing of the wounded cells by regulating proliferation, motility, and differentiation. Extracellular Matrix ECM produced by stromal keratocytes also contributes to corneal development and homeostasis. Collagen is one of the main components of ECM. In a study by Sun et al, the function of collagen V was examined by ablating Col5a1 in the stroma via the generation of conditional knockout mice, which were bred by the mating Col5a1 flox/flox mice with Kera-Cre driver mice.
The conditional knockout mice had abnormal corneas with decreased stromal thickness, decreased fibril concentration, increased fibril diameters, and disorganized lamellae, which resulted in corneal opacities. This study indicated that the central regulatory functions of collagen V play an important role in fibril and matrix assembly during tissue development. Moreover, collagen also contributes to corneal homeostasis. In alkali-burned or lacerated corneas, expression of collagen IV in the stroma of the injured corneas was increased, suggesting collagen IV may also contribute to the formation of basal lamina-like structures between the epithelium and the stroma. Lumican is a proteoglycan, located in the ECM, that is essential for corneal transparency. Previous studies have demonstrated that lumican maintains corneal transparency by modulating the synthesis of collagen fibrils, promoting corneal epithelial wound healing, regulating expression of multiple collagen genes, and maintaining corneal homeostasis (, ). Due to these functions, lumican-null mice exhibit delayed wound healing.
Additionally, anti-lumican antibodies can retard corneal epithelial wound healing in cultured mouse eyes. However, wound healing can be rescued by the addition of glycosylated lumican core proteins to the injured corneas. Further studies have revealed that the C-terminal domain of lumican binds the glycine-serine rich domain of transforming growth factor-β receptor 1 (ALK5) to promote epithelial wound healing by activation of pERK1/2 (, ). Keratocan is another protein located in the ECM that belongs to the small leucine-rich proteoglycan family. Corneal keratocan plays a pivotal role in matrix assembly, which is essential for corneal transparency. Keratocan-null mice have thinner corneal stromas with larger collagen fibril diameters and less organized fibril packing, compared to their wild-type counterparts.
In addition, the corneas have an altered shape with narrower corneairis angles. This reveals that keratocan plays a unique role in maintaining normal corneal shape and ensuring normal vision. Fibronectin is another prominent ECM component.
Fibronectin mediates the binding of epithelial cells to denuded corneal surfaces via integrin receptors. Additionally, after anterior keratectomy in rabbit models, fibronectin expression in the stroma is increased, which may contribute to wound repair (, ). Kinases IkB kinase β (IKKβ) is a kinase active in inflammatory response signaling, as well as the regulation of epithelial migration during corneal wound healing. A recent study demonstrated that IKKβ knockout in stromal keratocytes caused failed recovery in over half of the wounded corneas. The mice developed recurrent haze with increased stromal thickness, severe inflammation, and apoptosis. This study suggests that IKKβ in keratocytes is required for corneal wound healing via the repression of oxidative stress and attenuating fibrogenesis.
Comments are closed.
|
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |