Research On Target Detection And Tracking Algorithms Based On Sparse Representation

As a hot issue in computer vision areas,the research of tracking and detecting moving targets has been applied widely in various areas,such as Intelligent transportation,military,medical and video monitoring etc.In the process of tracking and detecting moving targets,impact of illumination,deformation,occlusion and background changes may affect the accuracy of experimental result.In order to improve the effectiveness and robustness of tracking and detecting,people can analyze the characteristics of foreground target,the external structure and the background of space.Sparse representation theory is about the sparse-representation of signal,which has become a hot topic in recent years and changed the status of sampling theories.This theory appears continuously in the field of target detecting and tracking.Although many algorithms have a certain increase,the technology is still immature.In accordance with the achievements summarized by predecessors,this paper mainly studies the target detection and makes some improvements.The main work is as follows:1.On the basis of sparse signal representation,this paper proposes an algorithm of moving target detection,which improves the method of three-frame differencing with Prewitt operator.First use the sparse-representation of compressive sensing to carry out the image de-noising processing.Because of the adaptivity of general principal component algorithm,the edges and local details can be preserved well in images.Secondly make sequential extraction of three frames from processed image,and do three-frame differencing operation on edges of moving target.Then do OR operation between result and the middle-frame which after background subtraction.Finally,execute binaryzation and morphological processing on the result.The experimental result shows that the algorithm is efficient,which can detect the target completely.This algorithm not only decreases the power dissipation,transmission quantity of image data and cost of video transmission,but also promotes the veracity of moving-target detection,even eliminates problems about double image and cavity.2.In order to solve the problem of target tracking drift under illumination change and fastmoving scenarios,this paper proposes a compressive tracking algorithm based on SURF.Firstly,choose the tracking target and extract SURF features.Then,do the sparse process on SURF features with the compressive sensing theory and reduce its dimensions.Finally,screening the features while use object samples to distinguish background from target,and update the classifier to determine its location.By the dimension method,the scale and time-cost of calculation in the tracking process can be greatly reduced.Furthermore,this algorithm considerably promote the accuracy and timeliness of tracking.