A Comprehensive Study of MeanShift and CamShift Algorithms for Real-Time Face Tracking

Abstract

Human face detection is an essential area of computer vision that harness the digital vision for detecting, recognizing, and tracking human faces. Face detection and tracking is widely being used in different applications like security monitoring, home video surveillance, etc. This paper demonstrates an effective human face tracking system for a real-time video stream. In particular, we investigate the two widely used algorithms, MeanShift and CamShift algorithms, for face tracking. However, face detection is performed only in the first video frame using basic Haar cascade features and AdaBoost classifiers. The advantages and drawbacks of MeanShift and CamShift algorithms are discussed in detail with the help of real-time video frames. Results show that the CamShift algorithm is more relevant in head and face tracking, which makes the algorithm highly robust and practically adaptive.