Research On Target Tracking Technology Based On Machine Vision

In recent years,with the continuous development of computer and image processing technology,object tracking has become a kind of intersecting frontier technology.It has a very profound application prospect and practical value in the security monitoring,robot,remote sensing technology,military and other fields.Machine vision is the key technology for achieving object tracking,which uses the machine to simulate the human eye system to capture the visual information in reality and locate the captured information.This paper mainly studies three aspects of the content.Firstly,the requirements of object tracking are analyzed.Then the machine vision system based on MF-DSC03 serial camera is established as the hardware platform to capture the object frame image.Secondly,aiming at the problem of tracking error or failure caused by using a single feature to describe the object in the current object tracking technology,the fusion of the color and contour features was used to describe the object.The fusion feature can solve the problem of incomplete description of the object with a single feature.On the basis of previous research,a fusion feature algorithm based on particle filter and Mean Shift is proposed.According to the distance between the state of each particle and the current state of the object,the Mean Shift algorithm is used to iteratively optimize the fusion to suppress noise before particle resampling.Then the particle converges on the local maximum value of the probability density function.In the stage of the feature fusion,the color and contour features are fused by adaptive fusion strategy.Different weights are given to each feature by measuring the similarity of each feature.In order to prevent the divergence of the object particles caused by the filtering process,the kernel function is adopted to adjust the weight of the system before resampling.In order to ensure the number of large weight particles,the filter divergence phenomenon occurred in the resampling process is effectively improved.And the accuracy and robustness of object tracking are greatly improved.Thirdly,the tracking drift phenomenon caused by the invariance of Mean Shift algorithm window is improved.In this paper,a scale-varying tracking algorithm is adopted,which adaptively adjusts according to its own scale in the process of object tracking.It effectively solves the above problems.Finally,the two methods proposed in this paper are verified.For the first algorithm,three groups of contrast tests are set up,including partially occluded pedestrian tracking test,illumination tracking test and tracking experiment in complex scene.Taking partial occlusion experiment is an example.The time consuming and error performance of the algorithm are analyzed.And the results show that the algorithm is effective and accurate.The results show that the algorithm can effectively improve the error caused by Mean Shift algorithm.There are 50 figures,8 tables and 59 references in this paper.