Research On Moving Target Recognition And Tracking Technology Of Mobile Robot

With the on-going developments of mobile robots in various fields such as smart services and smart cities in recent years,their follow-up and motion response to moving targets have become more and more concerned.With the continuous advancement of machine vision related technology,the acquisition of target motion state based on sequential image analysis processing is widely used due to its low cost,wide adaptability and small deployment difficulty.However,the existing identification and tracking methods,especially the model-free tracking method with only the first frame initial information and no other prior knowledge,have the defects of insufficient environmental adaptability,large amount of calculation,and poor long-term robustness,making it difficult to deploy or well adapt to the working environment.This paper studies the model-free moving target recognition and tracking technology of mobile robot platform for giving a solution to above problems.Firstly,this paper makes a deep study on the reorganization of the interested foreground target from the sequence images based on the feature matching method.An adaptive Gamma illumination equalization method is designed for the unbalanced illumination of robotic scene images.ORB feature point extraction and description method is used to extract the feature points in the scene image,and then the target feature points in the current frame are obtained by k NN matching.Secondly,the target motion information hided in sequence images is captured by the pyramid LK optical flow method.Since the optical flow is highly susceptible to illumination interference in the environment,the forward-backward error control and the NCC similarity control methods are used for filtering the tracked points.Furthermore,in order to make the system adapt to the application environment,the feature point-based detection-tracking fusion framework is designed for keeping the system long-term robust and preventing from manual initialization again after the target lost.Because of the ambiguity of feature matching and motion expression,this paper designs the density reachability condition of spatial proximity and feature similarity,and excludes the outer pointby density clustering.At the same time,the method of selecting the robust key frame to continuously update the reference in the sequence image is designed for the problem that the initial information in the first frame continuous decreasing with the recognition and tracking process,making system constantly adapt to the various environment and deformed target.In addition,the target position and motion state obtained in the current frame are used to predict the target position in the subsequent frame because of the correlation between adjacent image frames,reducing the interference of redundant image information and improving the real-time performance of the system.Finally,this paper builds a mobile robot image processing platform,and test the performance in various by ample experiments in realistic scene.Besides,contrast experiments are completed on the standard data sets to display the excellent characters of this paper between our method and outstanding tracking method in recent years.