EYE GAZE MOVEMENT BASED DETECTION AND CLASSIFICATION OF AUTISM

Project Title

EYE GAZE MOVEMENT BASED DETECTION AND CLASSIFICATION OF AUTISM

Project Area of Specialization Artificial IntelligenceProject Summary

This project will provide a benchmark for different algorithms for Eye Blink and eye gaze movement detection. The dataset consists of two modalities that can be combined for PoG definition:

(a) a set of videos recording the eye blink rate of specially abled human participants by doing different interactive activities on the computer screen.

(b) a sequence of eye gaze tracking also recorded using a webcam. The ground truth regarding the point of gaze provides the five regions (up, down, centre, right, left) which are known in advance since the participants are always looking at predefined targets on a monitor in a sequential pattern due to our conducted activities.

We try to give a practical explanation how data about eye blink rate and gaze movement patterns using computer vision domain, can be used to differentiate a normal and autistic person. The general consensus is that eye blink rate is directly proportional to user activity, our primary focus is to prove that data collected by us could be capable to serve a useful purpose. The procedure is assessed on a dataset gathered from different institutes for the disabled children having various  types of diseases(autism) by experiment containing different activities like playing games, talking , reading etc).

This project will provide a benchmark for different algorithms for Eye Blink and eye gaze movement detection. The dataset consists of two modalities that can be combined for PoG definition:

(a) a set of videos recording the eye blink rate of specially abled human participants by doing different interactive activities on the computer screen.

(b) a sequence of eye gaze tracking also recorded using a webcam. The ground truth regarding the point of gaze provides the five regions (up, down, centre, right, left) which are known in advance since the participants are always looking at predefined targets on a monitor in a sequential pattern due to our conducted activities.

We try to give a practical explanation how data about eye blink rate and gaze movement patterns using computer vision domain, can be used to differentiate a normal and autistic person. The general consensus is that eye blink rate is directly proportional to user activity, our primary focus is to prove that data collected by us could be capable to serve a useful purpose. The procedure is assessed on a dataset gathered from different institutes for the disabled children having various  types of diseases(autism) by experiment containing different activities like playing games, talking , reading etc).

Project Objectives

The project have following primary objectives:

i) To study the work related to this problem.

ii) To collect and develop a new dataset for training and classification.

iii) To classify a subject as Autisic or Normal.

Project Implementation Method 1  Detection of Eye Gaze & Blink

We have developed two different applications on python which are namely

1) Eye Blink Detection: This application detects the eye blinks of the participant and shows the timer and the blink counter on the screen. ( see figure 3.2)

2) Eye Gaze Movement Detection: This application detects the eye gaze movement of the participant on the screen. ( see figure 3.3)

 2 Eye Gaze & Blink Dataset

1) Eye Blink Data: We collect the data using the application in a txt file for autism and normal person respectively (see figure 2.2). After that we generate the graph of the eye blink data on the basis of blinks w.r.t time , where the peak value shows that the blink has occured at that particular point. ( see figure 3.5& 3.6 respectively for autistic and normal child graph)

2) Eye Gaze Movement Data: We use the approach as above and collected data for both the autistic and normal child respectively in a txt format ( see figure 2.3), which is further used to generate a heat map that shows the fixation time of the participants on the respective sides ( Centre, Up , Down , Left, Right). (see figure 3.5 & 3.6 respectively for autistic and normal child heat map)

3 Training the Collected dataset model and predictions:

We used linear regression to train our dataset and to generate a scatter plot to analyze and differentiate between an autistic and normal child on the basis of difference of Eye blink and Eye gaze patterns(average value). The training dataset model gives good enough accuracy on the test sets. For prediction we used the data of eye blink and the averaged data of the gaze movement to predict as accurately as possible if a person is autistic or normal.

Benefits of the Project

The main benefit on the success of this project will be that , that the disease such as like Autism could be detected at an earlier age with the help of eye blink and gaze movement which could inevitably help to diagnose it way earlier stage and possibly curing it at a later stage.

Technical Details of Final Deliverable

The project will have the following final outcomes:

Firstly the major goal of this proposal is to find out the accurate way to classify the child  into normal & abnormal(autistic) and to develop and train a new dataset model that will be publically available for more research in the future.

Final Deliverable of the Project Software SystemType of Industry Medical Technologies Artificial Intelligence(AI)Sustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	40000
LED Screen	Equipment	1	20000	20000
HD Webcam	Equipment	1	20000	20000