Visual Object Tracking Using Color And Texture Features Based On Mean-shift And Particle-kalman Filter Algorithm

Due to increasing demand of video surveillance system,intelligent video surveillance system has become challenging subject in the field of computer vision research.In intelligent video surveillance system there are four key steps,i.e.object detection,object classification,object tracking,and object analysis.Among these steps,object tracking is considered to be a key and important task in intelligent video surveillance system.Object tracking is considered as difficult task because of several problems such as illumination variation,tracking non-rigid object,non-linear motion,occlusion,and requirement of real-time implementation.Every single algorithm in visual object tracking always has advantages and disadvantages.Tracking with only a single algorithm is therefore generally considered to be inadequate and inefficient because each individual algorithm has its limitations.Classic Mean Shift tracking algorithm always suffers from large position errors,which may lead to the failure on tracking target in complex environment.To handle this problem,an improved Mean-shift Particle-Kalman Filter tracking algorithm based on texture and color features is proposed.In the proposed method,a color feature based Mean-shift is used as the main tracking algorithm when the target-object is moving in linear motion or when there is no occlusion occurred.Conversely,when occlusion occurs or when the target-object is moving in non-linear motion,color and texture features fusion based Particle-Kalman Filter tracking algorithm is used as main tracking algorithm.The experiment results show that the proposed method can be implemented in single object tracking and able to cope with several tracking problems such as illumination variation,non-rigid deformation,non-linear movement,similar color interference,and occlusion.Furthermore,the experiment results show that the proposed tracking algorithm which utilizes multi features(color and texture features)to represent the target model has better performance evaluation results compare to other comparator algorithms including proposed method which only utilizes color feature to represent the target model.