Research Of Object Tracking Algorithms Based On Compressive Sensing

Object tracking is a hot research direction in the field of computer vision.Its task is toestimate locations of an object in an image sequence.Object tracking technology is playing an increasingly important role in military tracking,intelligent transportation,robotics and other fields.However,the interference factors such as illumination change,fast motion,occlusion and deformation can affect the accuracy,robustness and real time of object tracking during the tracking process.Therefore,it is necessary to design a real-time object tracking algorithm with good robustness and high accuracy.This paper studies the tracking algorithms based on compressive sensing,and two tracking algorithms are proposed on the basis of the existing research.In order to solve the problem that the traditional compressive tracking algorithms are vulnerable to occlusion,illumination change and the change of object's scale,a compressive tracking algorithm based on color attributes and online boosting feature selection is proposed.Firstly,the particle filter framework is used to predict the possible positions of the object,and the six parameters of affine transformation are used to describe the state of the object.Secondly,color attributes are introduced into this algorithm,and are used to describe the image combined with the luminance information,which can balance the extracted features between discriminative power and photometric invariance.Thirdly,an online boosting feature selection method based on optimizing Fisher information matrix is proposed,which is used to select features and build a strong classifier.Finally,candidate samples are classified by the strong classifier,and the sample with the maximum return value is the tracking result.Traditional compressive tracking algorithms have poor stability because the collected features are single and the updating parameter of the classifier is fixed.In order to solve this problem,an adaptive compressive tracking algorithm based on dual features is proposed.Firstly,the algorithm improves the measurement matrix in traditional compressive tracking algorithms,and obtains gray features and texture features to describe the object.Then,candidate samples are classified by the classifier,and the sample with the maximum return value is the tracking result.Finally,the perceptual hash algorithm is used to obtain the similarity between the object and the tracking result.According to the similarity degree,the updating parameter is adjusted adaptively to avoid the situation that the classifier can not cope with the over update and the appearance change of the object.In this paper,the two tracking algorithms are tested qualitatively and quantitatively onseveral video sequences containing different interference factors,and compared with other tracking algorithms that have excellent performance.The experimental results show that the tracking algorithms proposed in this paper has good accuracy and robustness,basically meets the requirement of real time,and are better than other algorithms.