Research On The Integrated Method Of Automatic Detection And Tracking Of Image Targets

Object detection and tracking technology is one of the research hotpots in the field of computer vision,and it plays an important role in intelligent transportation,national security,public security and other fields.At present,domestic and foreign scholars have proposed many object detection and object tracking algorithms.However,in real scenes,subject to the scale change of the object,complex background,occlusion and many other factors,the object detection and tracking technology is still facing great challenges.This paper conducts research on detection technology based on object model and tracking algorithm based on correlation filtering,and designs the integrated approach.The main contents are as follows:(1)Aiming at the problem of poor real-time performance and high false detection rate of the HOG+SVM object detection algorithm,an improved Gaussian mixture model combined with the HOG+SVM detection algorithm is proposed.Firstly,use Gaussian mixture algorithm to establish the background model and extract the moving object area,then locate the moving object by the four-neighbor search method,finally,calculate the HOG feature and send to the SVM detector for specific object detection.The simulation results show that the improved algorithm can effectively improve the real-time performance and reduce the false detection rate.(2)An integrated detection and tracking method for specific types of objects is proposed,which combines the improved HOG+SVM detector with the KCF tracking algorithm for different scenes:In the single object scene,the detector captures the object and assigns the initial position of the tracker,then the maximum peak response value is used to judge the tracking reliability.When the tracking fails,the detector is called for object re-detection.In the multi-object scene,the detector and tracker adopt the correlation method based on IOU and Bhattacharyya distance.Experimental results show that this method can track specific objects with high precision and long time.(3)To solve the problem of poor detection effect of YOLOv3 algorithm on small and occluded objects,an improved YOLOv3 algorithm is proposed.First,use the L_GIOUloss function to replace the original loss function of YOLOv3,then optimize the NMS algorithm,and finally test on the COCO data set.The results show that the average accuracy of the improved YOLOv3 algorithm is 2.87 higher than the original algorithm,and it can accurately detect small objects and occluded objects.(4)An integrated method of object detection and tracking for complex scenes is proposed,which combines the improved YOLOv3 with the KCF tracking algorithm.Taking advantage of the high detection accuracy of the improved YOLOv3 algorithm,a collaborative work mechanism based on the IOU similarity measurement is designed to improve the detection accuracy and tracking stability of the algorithm.Experiments on different data sets show that the proposed method can track objects stably with high precision in complex scenes and has stronger robustness.