Research On Real-time Object Tracking Algorithm In Complex Scenes

Object tracking is a fundamental problem in computer vision due to its wide range of applications,such as motion recognition,video surveillance and human computer interaction.Object tracking is usually formulated as an estimation of the target state,such as location,scale and so on.Specifically,the purpose of an object tracking algorithm is to estimate the state of a target in a continuous image sequence given the initial state of a target.Although many achievements have been made these years,to establish a real-time tracking system when the object suffer from large appearance changes caused by unstable illumination,occlusion and deformations of itself is still a challenging task.The appearance model is a basic element in object tracking,so this paper mainly focuses on algorithms which are based on the classification of the target and background and improves the appearance model in complex scenes.The main works are as follows:(1)We make two improvements under the framework of the CT(Compressive Tracking):firstly we extract the rectangle features based on the compressed sensing theory.The reliability of each feature is checked according to its classification performance for object tracking in the current frame.Then unreliable features are updated in time.Secondly,the values of reliable feature's weight are increased in real-time such that their importance can be emphasized.Finally,these new weighted candidate features are inputted into the beyesian classifier to distinguish the object from background in the next frame.Compared with traditional compressive tracking algorithm,experimental results show that our algorithm achieves better tracking results in complex scenes;(2)We analyze the defect of a tracking algorithm which considers only the appearance of the target,and then we deduct the auxiliary effect of background information on tracking an object in a complex scene.As a representative algorithm which takes background information into consideration,an algorithm called STC(Spatial-Temporal Context)tracking is introduced in detail;(3)We make two improvements under the framework of the STC:firstly four possible positions of the target in the current frame are predicted based on the displacement of the target between the previous two frames,then we calculate four confidence maps at these four positions;the target position is located at the position that corresponds to the maximum value.Secondly,we propose a target reliability criterion and design an adaptive threshold to regulate the updating speed of the model.Specifically,we stop updating the model when the reliability is lower than the threshold.Experimental results show that the proposed algorithm achieves better tracking results than traditional STC and other algorithms when there exists several challenge in a complex scene.