Descripció del projecte
The problem of diagnosing subjects with mental disorders like ADHD, ADD, Alzheimer, etc. is of a great importance in the current state of the healthcare worldwide. Furthermore, the current challenge is to identify diseases at their early stages in order to be able to introduce less harmful and effective treatment with no additional costs. One of the cheap and effective approaches to address this problem is the cognitive games paradigm which requires infra-red eye tracker, a conventional display and 10 minutes for the game duration to predict the likeability of a subject to have a certain disease. Such approach relies on signal processing of the data acquired from an eye-tracker and further machine learning techniques to perform classification. The clear advantage of the above-mentioned approach is the time required for behavior markers measurement as well as the size of the installation which makes it possible to be used in schools/offices/hospitals etc. Conceptually this solution eases the diagnosis procedure but introduces the problem of purchasing the relatively costly equipment needed for eye tracking. This PhD project aims at investigating the potential of the conventional web-cam captured frame sequences to be used in the machine learning applications for quantifying the eye vergence within subjects suffering from cognitive disorders like ADHD, ADD, Alzheimer etc. together with other non-medical applications like estimation of the add efficiency on websites.
The project aims at comparing 2 acquisition approaches: infra-red tracker vs web-cam for eye vergence with further model generation for predicting the diagnosis. In order to use the web-cam approach to capture the orientation of eye’s gaze one must take into account the head pose estimation problem. For this purpose, a data-base collection step must be introduced involving a variety of CMOS sensors and infra-red eye tracker capturing the subject’s movement simultaneously. This study model will provide us with the face orientation (roll, pitch, yaw) together with the gaze position captured by eye tracker allowing us to use both measurements on the web-cam frame-sequence. Consequently, this data will be used to train a model capable of estimating the location of the pupil within the image (captured by different CMOS sensors) which will serve as the rough estimation of the gaze location. Image feature extraction techniques combined with machine learning or deep learning are expected to serve as the base for the prediction model in this case. Later stage of the study will involve upgrading the model to measure the real eye vergence for more precise classification.
Braingaze collaborates with academic hospitals like, Vall d Hebron, HSJD, Hospital Clinico, Teknon, Sanitas, Hospital Mataro and Universities being UB, UPC, UAB, Goethe University, Kings College, New York University. As well as with the companies Validators and Facebook
