Research On Camshift Tracking Algorithm Based On Multi-feature Fusion

Due to the wide range of applications in the field of intelligent human-computerinteraction, medical diagnosis, video surveillance and military, target trackingcontinues to highlight the value of its application. It has also become a hot researchfield of computer vision, image processing and pattern recognition. This research hasgreat value both in the theoretical study and practical application.At the present stage, target tracking technology can not be perfect applied topractical uses. How to achieve accurate identification and tracking of a target is arecognized problem, one of the key issues is how to adaptively choose the appropriatefeatures for the accurately target modeling. This thesis has done some exploration andresearch to solve this problem, extracting multi-feature information from the imageand fusing them to express the target model, and has improved the trackingperformance of the CamShift algorithm. The main work results can be summarized asthe following aspects.First in the terms of multi-feature fusion, mainly study the typical local invariantfeatures, and edge texture features. With the characteristics、the pros and cons of eachfeature, weightedly fuse color features and edge texture features to form the targetmodel, to make early preparations for the follow-up tracking algorithm.Secondly in the terms of target tracking algorithm, study the CamShift trackingalgorithm based on SURF feature match. In order to solve the problems of poor orinvalid tracking performance which is caused by the color sensitivity of Continuouslyadaptive Mean Shift (CamShift) algorithm, a new CamShift tracking algorithm basedon local feature matching is proposed. The new algorithm used the method ofSpeeded Up Robust Features (SURF) to extract the local feature points containing theimage information from the target and search areas of multi-channel images, and thenmatched the feature points by the method of approximate nearest neighbor searching;the location, scale and orientation information of the feature points are obtainedutilizing the purified matching results, therefore the CamShift method is constrainedand updated.Further,in the terms of target tracking algorithm,study the CamShift tracking algorithm which is based on feature fusion and motion estimation. CamShiftalgorithm only uses the color feature for the target modeling, does not use the otherfeatures containing the target information. In view of this, the paper proposed aCamShift tracking algorithm based on feature fusion and motion estimation.Weightedly fuse the color information and the edge information to form theprobability distribution of the target model, using second-order autoregressive modelto correct target position and velocity. Then in the LSV combination color space, takethe SURF method to correct the region of interest to improve the accuracy andstability of tracking.Finally, using VC++and OpenCV complete realization of the tracking algorithm,and design experiments and analysis for the algorithm. The experimental results showthat, compared with the classic CamShift algorithm and similar improved algorithms,the proposed algorithm can achieve real-time tracking of the rotation and zoommoving target against complex backgrounds, and has excellent anti-interferenceability.