Research On Visual Tracking Algorithm Based On Target Feature Matching

Computer vision is to quickly and accurately obtain the target motion state parameters such as the position,velocity and acceleration of the target through the computer.Based on this information,the target can be further analyzed.the deterministic visual target tracking algorithm based on feature matching,because of its high real-time and high accuracy,has achieved good tracking effect in some tracking scenarios.However,in some complex tracking scenarios,such methods are prone to losing targets under various interference factors.Therefore,there are still many defects in this type of visual target tracking algorithm that need to be improved.Aiming at the problem that is easy to fail in tracking micro-sized targets.Taking the neurofilament protein image from the fluorescence microscope as the experimental object.Using the color histogram in HSV color space to establish a target model and weighting target feature points according to prediction point.introducing a kernel function into the target color probability model so that uses the kernel density gradient to Search object,finally obtains the specific location of the object in each frame of image.thanks to the particularity of the neurofilament protein,this paper also compares and analyzes the tracking effects of two other probabilistic prediction algorithms.The experimental results show that this method can quickly and stably track neurofilament proteins and provide a new approach for neurofilament protein research.Aiming at the target modeling of traditional camshift algorithm and matching method that is susceptible to interference pixels,the target modeling and matching method are improved accordingly.The improved target modeling method selects the most obvious color features in the target by block,making the distinction between the target and the background more obvious.In addition,during the target matching process in the tracking process,the tracking method based on the traditional model needs to refer to the pixels of the candidate target region one by one,and the calculation amount is large.The improved method divides the candidate target regions into blocks,and then extracts the features of each block to compare with the targetmodel.This method not only reduces the amount of calculation,but also spreads the influence of the interfering pixels to the surrounding area,so that the interfering pixels can be eliminated through The means of binarization.The experimental results show that the tracking accuracy and speed of the target tracking method based on the improved model are greatly improved.Aiming at the problem that the KCF algorithm is difficult to deal with complex tracking scenarios,An evaluation method for tracking effect of each frame is proposed.First,describe the search area in the HSV color space and build a discriminant model.Generating a color probability distribution map of the search area during the tracking process and evaluating the initiative result of KCF algorithm in it.Then based on the evaluation,the correction algorithm is enabled to correct the tracking result.In the aspect of target model updating,through the evaluation of the initial tracking results to identify complex scenes,a model updating strategy of adaptive learning factors is proposed to avoid introducing too much background or obscuration information.The comparison of experimental results shows that the improved algorithm can effectively deal with complex tracking scenarios.Finally,the paper summarizes the research work of the full text,including the problems that have been solved and the remaining imperfections,and points out the key points that need further research in this research direction in the future.