Target Tracking Technique Based On Visual Saliency In Complex Environment

Accurate feature extraction of targets,occlusion of target and complex tracking scenes are common problems in object tracking,these problems can be effectively solved is helpful to improve the stability and robustness of target tracking algorithm.Based on the framework of saliency tracking,this thesis studies three problems in target tracking: effective description of target appearance,multi-feature fusion and multi-feature adaptive combination,this thesis proposes some new tracking methods.The main research work of this thesis are as follows:(1)Tracking algorithm base on super-pixel and fragments is researched.Under the saliency target tracking framework,using the adjacent pixels with similar texture,color and brightness characteristics of super-pixels,the pixels are grouped into visual blocks,and the condition number of Heisen matrix is used to realize the target segmentation.At the same time,the blurring of each block is judged,and the corresponding fragments are modified.According to the distance between the fragments and the center of the target,the correlation of each fragments is established to achieve the target sample.Finally,the final target location is determined according to Naive Bayesian classifier.(2)Tracking algorithm base on multi-feature fusion and fuzzy processing is researched.Under the saliency target tracking framework,initialization of target fragments is realized by combining super-pixel and condition number,and the adjacent fragments are merged to form a pixel block to adapt the trackers.The occlusion problem is solved by local tracking.Two independent and complementary features are used to describe the target pixel block,and multi-feature fusion is realized to improve the robustness of target tracking in complex environment.Finally,the feature similarity between the pixel block and the observation model is used as the input of the fuzzy logic system to realize the adaptive weighting of multiple trackers and determine the final target location.(3)The target tracking technique combining feature adaptive combination and fuzzy processing is researched.Under the saliency target tracking framework,the adaptive update strategy of the target feature template is introduced.The similarity of feature template in observation model and the current frame's feature template is used to update the feature template.The feature template is updated using the maximum likelihood ratio,and a feedback-based parameter update strategy is added to the classifier.The thesis selects the open test sequence OTB to analyze the research algorithm.The experimental results show that the saliency tracking method based on super-pixel and conditional number fragments can effectively solve the ambiguity problem in target tracking and improve the real-time performance.The saliency tracking method based on multi-feature fusion and fuzzy processing can deal with the local occlusion problem in target tracking.Based on the saliency tracking method of multi-feature selection and fuzzy processing,long-term target tracking in various tracking scenarios is realized.