Development of Robust Multiple Face Tracking Algorithm and Novel Performance Evaluation Metrics for Different Background Video Sequences

Abstract

In computer vision, face tracking is having wider opportunities for research activities using different background video sequences because of various factors and constraints. Due to the challenges that are increasing day by day, old/existing algorithms are becoming obsolete. There are many powerful algorithms that are limited to certain set of video sequences. In this paper, we are proposing an algorithm that detect and track multiple faces in different background video sequences. Viola-Jones face detection algorithm is used in such a way that, new face/first face need not to be in the starting frame of the selected video sequence. The proposed algorithm successfully detect new face(s) along with existing face(s) by keeping track of the facial data using BRISK feature points. The mean of the old points and new points are calculated based on the area of the facial data. The detected face(s) in further frames undergoes similarity check with existing facial data. If detected facial data and existing facial data mismatches, then the detected facial data is entered into face tracks structure. By using point tracker method, the proposed algorithm track those points that has been set for each of the facial data.

Color Based New Algorithm for Detection and Single/Multiple Person Face Tracking in Different Background Video Sequence

“…Segmented sections are detached out of the frame by applying a threshold; and facial features are used to know whether these regions comprise person face or not. Knowledge based approaches [8,9] adopt geometrical attributes of human face for detection and tracking. Using RGB color range, it is tough to decide whether the color is human skin color or not.…”

“…Ranganatha et al [9], [24,25] have proposed three algorithms for tracking multiple face in different background videos. The first algorithm was based on BRISK [26] features, the second algorithm was based on image training & LBPH [27], and the third algorithm used image training, corner [18] and FAST [28] features.…”

A Comprehensive Survey of Algorithms for Face Tracking in different Background Video Sequence

“…KLT algorithm is a core for the feature tracking algorithms as many of the feature trackers are based on the point values. Ranganatha et al [23][24][25][26][27] have proposed several algorithms that are based on KLT. These algorithms are robust enough to track both single and multiple face in different background video sequence.…”

“…These algorithms are robust enough to track both single and multiple face in different background video sequence. The algorithms addressed many questions which were left unanswered previously such as, 1) Tracking multiple face [24,25], 2) Selective face tracking [26] i.e. user specified number of face(s) tracking, 3) Selected single face tracking [27], 4) Occlusion (e.g.…”

“…spectacle, beard, headscarf etc.) [25,26], 5) Low resolution [23][24][25][26][27], 6) Face skin colored regions [26], 7) Abrupt motions [23][24][25][26][27], 8) Variations in pose, expression and illumination [23][24][25][26][27] etc. to name a few.…”