METHOD FOR FACE RECOGNITION-BASED VIDEO HUMAN IMAGE TRACKING UNDER COMPLEX SCENES

摘要

The present invention relates to a method for face recognition-based video human image tracking under complex scenes, comprising the following steps: 1) decoding acquired video data to obtain a series of video frame images; 2) utilizing a face detection and object tracking method to process the obtained series of video frame images, so as to obtain all possible face trajectories; 3) judging similarities between the face trajectories obtained in the step 2) and subsequent adjacent face trajectories, and combining the face trajectories having higher similarities than a threshold value; and 4) outputting the combined face trajectories. As compared to the prior art, the present invention has the advantages of fast processing, high detection precision, low false positive rate and the like, and can address the problem of interruption and mixing-up of the face trajectories caused by shelter and head turning.