The Application And Research Of Object Detection And Tracking Algorithm Based On Multi-sensor Fusion

With the development of society and the progress of science and technology,intelligent image algorithm has been widely used in military,medical,transportation and other fields,among which object detection and tracking algorithm is the most popular branch of intelligent image algorithm.A large number of false alarms in the engineering environment and the limitation of the performance of the computing platform bring greater challenges to the object detection and tracking.Perimeter monitoring project as the background,the real-time object detection and object tracking algorithm is studied,adopting multi-sensor fusion strategy with multiple IPC camera and one PTZ camera to automatically track and enlarging the object details while taking into account the panorama,in order to meet the needs of rapid warning,accurate positioning,panoramic detection of the actual project.First,for the traditional algorithm,based on the analysis and the study of common motion object detection algorithm,the Vi Be background model algorithm and frame difference method are combined to detect of the target area and solve the problem of ghost,the integrity of objects is improved by morphological operation,and the hue threshold method and hue-saturation two-dimensional histogram filter method are used to filter out false alarm interference,on the premise of guarantee the real time the detection error is reduced.For the deep learning algorithm,the object detection algorithm based on Mobile U-Net is proposed,studying the deep image segmentation network U-Net and combining the light-weight depth separable convolution,and applies the image segmentation algorithm to the target detection to realize the real-time and efficient end-to-end detection.Then,the online multi-object tracking algorithm framework is studied.The multi-object tracking problem is a task assignment problem.The coordinates,height and aspect ratio of the object bounding box obtained by Kalman filtering are correlated with the corresponding spatial information obtained by the object detection to get tracking object queue.The Hungarian algorithm based on GIo U as the cost function is used to improve the accuracy of the association,and use the object trajectory processing,so as to achieve the goal of starting,maintaining and eliminating the object.Finally,a set of multi-sensor linkage control system with two IPC camera,one fish-eye camera and one PTZ camera unit is designed.Firstly,fish-eye camera needs to be distorted because of its imaging characteristics.After the PTZ control thread obtains the coordinate information of the tracking object,according to the coordinate mapping relationship,it issues control instructions to the PTZ camera and drives the PTZ control of the PTZ camera,so as to observe the high definition details of the object.The PID algorithm is used for the pan-tilt control,and the table lookup method is used for the zoom control,in order to achieve accurate and efficient servo tracking.