Gaze estimation system and method thereof

Abstract

A system to estimate audience parameters having the following features: receiving inputs from (a) a monocular camera ( 102 ) placed top-center/bottom-center of a target signage board ( 101 ); (b) a camera tilt calibration module ( 103 ); (c) camera parameters ( 104 ); (d) signage parameters ( 110 ); and (e) output from the ML systems ( 105 ) to process and analyse the monocular camera images to aid as input to the gaze estimation module ( 106 ) and produce output ( 107 ). The output gives the person gaze at the target and other additional parameter like gender, person height and depth from signage and provides data regarding user interest and engagement levels with the target signage board.

1 . A system for audience head gaze estimation  5  comprising:
 a gaze estimation module ( 106 ) provided with the following inputs:
 a. input from a monocular camera ( 102 ); 
 b. input of camera tilt angle from a tilt calibration module ( 103 ); 
 c. signage parameters ( 104 ); and 
 d. learning input for ML based models ( 105 ) which processes the inputs using several ML models to estimate the following:
 i. person gender; 
 ii. person head pose; and 
 iii. person face key points; 
 
 
 wherein, the person head pose is used to provide feedback for the gaze estimation module ( 106 ) to thereby gauge audience interest towards a displayed signage board ( 101 ). 
 
     
     
         2 . The system for audience head gaze estimation, as claimed in  claim 1 , wherein, the input from the monocular camera can be images or video. 
     
     
         3 . The system for audience head gaze estimation, as claimed in  claim 1 , wherein, the camera tilt calibration model provides tilt parameters with known intrinsic and extrinsic parameters. 
     
     
         4 . The system for audience head gaze estimation, as claimed in  claim 1 , wherein, the signage parameters include location, position, dimensions of the display signage board, and camera positioning height. 
     
     
         5 . A method for audience head gaze estimation comprising the steps of:
 Inputting images from a monocular camera;   Utilizing learning systems that process and analyse the images and produce output ( 107 ) to obtain head pose and gender;   Using pre-calibrated camera parameters, signage parameter and the gender to compute person height and person depth; and   Computing person gaze at target from the head pose, the person height, and the person depth.   
     
     
         6 . The method for audience head gaze estimation, as claimed in  claim 5 , wherein, the audience head gaze is estimated to provide information regarding user interest and engagement levels with a target signage board.