Research On Moving Object Detection And Tracking Based On Mobile Robot

In recent years,with the rapid development of robotics technology,mobile robots have played an extremely important role in daily life,especially in areas such as intelligent transportation,military security,and smart medical care.At present,moving object detection and tracking technology in static background have become more and more mature;however,the moving target detection and tracking algorithms in dynamic backgrounds have developed relatively slowly.Therefore,it is extremely important to study the moving object detection and tracking algorithms in mobile robot vision systems.The research status of moving object detection and tracking technology at home and board is studied in the thesis,and the moving target detection and tracking method in dynamic background is specifically analyzed.The moving target detection algorithm based on motion compensation and the tracking algorithm based on classifier are chosen in the thesis,and the following work is completed:Firstly,the concrete steps of the motion compensation-based detection algorithm are studied in the thesis.In order to improve the accuracy of motion parameter estimation,a motion compensation algorithm for improved BRISK feature matching is chosen.The Gaussian filter is used to process the image and remove the noise points in the image.The image is divided into blocks and the image block is screened by the image entropy.Then the sub-blocks with too many corner points in the image are removed,and the filtered image blocks are used to extract the BRISK features.For reducing the rate of the false matching in feature matching,the k-nearest neighbor and Euclidean distance are used for feature matching.In order to further improve the accuracy of parameter estimation,the improved PROSAC method is used for feature point purifying,and the six-parameter affine model is used to complete background motion compensation.Frame difference method and morphological processing are used to extract moving objects.Secondly,an improved KCF's classifier of tracking algorithm is used to track object.The HOG algorithm is used to extract the target features,and the circulated matrix is used to obtain a series of positive and negative samples.The ridge regression classifier is trained by the sampled positive and negative samples,and the tracked target is used to update the classifier.In order to cope with the problem of tracking failure of the KCF algorithm,the BRISK feature matching is used to detect the tracking failure,and the template matching is used to re-detect the target.In the end,the detected target is used to re-train the KCF classifier to continue the tracking process.Finally,the vs2015 platform and the opencv computer vision library are used to realize the proposed improved algorithm in the thesis,and multiple video sequences under dynamic background are used to test the proposed algorithm.Experimental results show that the improved BRISK feature motion compensation algorithm improves the accuracy of motion parameters,and it is suitable for video sequences in dynamic background.The improved KCF tracking algorithm can adapt the changing video background and has good real-time and robust performance.