Research On Image Tracking Method Of Incoordinate Maneuvering Target

Target tracking is a hot research direction,and it is also an emerging application direction and a topic of high concern in the field of computer vision.Because the speed and direction of the non-cooperating maneuver target are random,the target's visual tracking method is more complicated due to the uncertainty of the target's own state and the external environment.The tracking process is prone to target errors or loss.Psychological research shows that the human visual system has the ability to pay attention to the selection of visual stimuli.Based on behaviors and tasks,irrelevant information is filtered out and only regions or objects of interest are selected and processed,so the human eye can easily track down the target.Making use of the reference frame which provided by nature----visual attention mechanism,this paper analyzes the human vision system by bionics method and establishes the computational model of the process that is changed by visual focus.Then,by simulating human visual characteristics,a tracking method for non-cooperating maneuvering targets is proposed,which is a nuclear correlation tracking algorithm based on visual saliency.The main work is as follows:1.Since the motion of incoordinate maneuvering target does not prior knowledge,it may lead to motion blur and defocus blur in acquiring image.Blurry image will have a great impact on target identification,therefore this paper does the research about the method of image restoration with mixed-blur image.The motion blur and defocus blur images are analyzed respectively,and then different restoration methods are carried out according to the causes of the mixed-blur image.On the basis of experimental comparative analysis,a method for restoration image is presented that restores motion blur first,and then deals with defocus blue.This method can restore the target image clearly without obvious ringing effect.2.In the tracking process of incoordinate maneuvering target,it is necessary to accurately identify the tracking target in the complex background,that is,to detect the saliency of targets.Under the visual attention mechanism,the image saliency is shown as the degree to which the human eye attaches importance to some regions of the image.Methods commonly used target detection are mainly applied to areas with a single background and a small interference environment,and the tracking effect on non-combined maneuvering targets is not ideal.Hence,a method for image saliency detection based on contrast enhancement is introduced.In this method,the contrast of input image is full-resolution enhanced,and salient regions are extracted via Gaussian bandpass filter,then three characteristic channels of Lab color space are normalized respectively.According to the power of each channel,dynamic weight is distributed to combine into a final saliency map.The results show that the clarity of the target profile and internal details are improved,and the salient region is highlighted clearly,much easier for target detection and recognition.3.Based on the principle of kernelized correlation tracking,an adaptive kernelized correlation tracking method on the basis of visual saliency is proposed to achieve automatic tracking.Using the KCF tracking algorithm as the basic framework,basic sample for the tracking is training,and the classifier is initialized.FHOG feature of the target is used for key point matching,which can well solve the problem of uncertain speed,direction and scale change of maneuvering target.4.Firstly,the video sequence of cars,trucks and motorcycles is verified by the tracking method.Then,the non-cooperative maneuvering target tracking experiment is carried out on the actual model car.The tracking image shows that the tracking algorithm is designed to match the non-coordination.The maneuvering target has a good tracking effect,and can also effectively deal with some incomplete tracking problems of the target images.