Object Tracking Algorithm Based On Multi-Feature Fusion

Object tracking is an important research topic in the field of computer vision,pattern recognition,artificial intelligence,etc.It finds numerous promising applications,including video surveillance,human computer interaction,intelligent transportation,etc.In complex scenes,designing an effective object tracking algorithm,however,still has many key issues unresolved due to illumination changes,object occlusion,object appearance variations,etc.Therefore,designing an accurate,robust and real-time object tracking algorithm is an urgent need.Because of the timeliness of features,a single feature cannot adapt to dynamic changes of the scene,the tracking algorithm based on a single feature is difficult to achieve robust results.If multi-feature fusion is used in the tracking algorithm,it can make use of the complementarity among different features to better adapt to changes of the scene and achieve robust tracking results.Therefore,in this paper,we researched the key technology of classical MS algorithm and particle filter algorithm based on multi-feature fusion framework.The proposed algorithm based on multi-feature fusion improved the robustness and precision of tracking in complex scenes.The main research contents of this thesis are as follows:Firstly,the method of establishing multi-feature model based on the function of measuring the feature discriminability and stability is proposed,a generic model of multi-feature fusion object tracking has been designed,which is the basis of achieving multi-feature fusion object tracking under different algorithm framework.Based on the above generic model,a framework is established for MS algorithm and particle filter algorithm.Secondly,a novel MS object tracking algorithm based on multi-feature fusion is proposed.The classical MS algorithm has many defects,such as the single color feature cannot contain all the information of targets,the tracking window size cannot be adapted to changes of the scene,lacking an updating mechanism of the target model,blocking migration,etc.In order to solve the problems of the classical MS algorithm,the main improvements are as follows: Establishing the target appearance model based on multi-feature fusion strategy to solve the problem of classical MS algorithm only uses the color feature to describe the target model.According to the discriminability of each feature between target and background dynamic set the feature weights.According to the weights of different features,an adaptive model updated method and occlusion handling method are proposed.By adaptive adjust the bandwidth of kernel function to achieve scale adaptive.The forward-backward tracking is used as the basis to improve the tracking precision.Experimental results show the effectiveness of the proposed algorithm under the scenes of target occlusion and illumination changes.Thirdly,a novel particle filter object tracking algorithm based on multi-feature fusion is proposed.The method mainly solves the poor robustness problem of the classical particle filter algorithm based on single feature in the object tracking process.First,in order to avoid using rectangle description target includes some background pixels,this paper establishes the target appearance model based on local patch and multi-feature fusion strategy.Next,a novel feature fusion strategy is proposed based on the discriminability of each feature.Then,according to the stability function of each feature update the target model to achieve the proposed algorithm long-time and effective tracking.Finally,the simulation results show that the proposed algorithm can track target under complex scenes in robust performance.