A method for adaptive multi-feature fusion for robust tracking

Abstract

The invention relates to a self-adaptive multi-feature fusion method for robust tracking. The self-adaptive multi-feature fusion method includes performing feature structure initialization, respectively calculating features of histograms of oriented gradients (HOG), local binary patterns (LBP) and color histograms, calculating principle components of the features by an incremental principle component analysis (IPCA) algorithm, acquiring stability characterizations of the features, updating sub-feature weights, and calculating updating rate of the sub-features according to tracking results of particle filtering. Target features of the self-adaptive multi-feature fusion method are expressed by gradient information, texture information and color information, so that descriptive power of the features is improved in complex conditions. According to different stability of different sub-features in tracking, the feature weights are updated in time to improve stability of fusion-improved features. The self-adaptive multi-feature fusion method has small calculated amount, and can meet the requirements on real-time performance.

Description

technical field [0001] The invention relates to the fields of video monitoring and pattern recognition, in particular to an adaptive multi-feature fusion method for robust tracking. Background technique [0002] Intelligent Video Surveillance (IVS: Intelligent Video Surveillance) is an emerging application direction in the field of computer vision in recent years. It uses computer vision technology to process, analyze and understand video signals, and controls the video surveillance system, thereby improving the intelligence level of the video surveillance system. Intelligent video surveillance system has great application prospects in both civil and military fields. Although monitoring cameras are generally installed in some important public places such as banks, shops, stations, and ports at present, the actual monitoring tasks still require more manual work to complete. In many cases, the information provided by the current video surveillance system is raw video data wi...

AI Technical Summary

Problems solved by technology

Such as adaptive model approach (Z.Kalal, K.Mikolajczyk, and J.Matas, "Tracking-learning-detection," PAMI, vol.34(7), July2012; H.Grabner, C.Leistner, and H.Bischof, "Semi-supervised on-line boosting for robust tracking," ECCV, 2008), but these methods of updating the model assume that the detection and tracking of the previous frame image are correct. Obviously, such an update method is difficult to guarantee the accuracy of tracking; in addition, there are multi-feature Combined methods (J.Kwon and K.M.Lee, "Tracking by sampling track-ers," ICCV, 2011; G. Shu, A. Dehghan, O. Oreifej, E. Hand, and M. Shah, "Part-based multiple-person tracking with partial occlusion- sion handling , "CVPR, 2012)

Images