Research On Key Technology Of Fast Image Fusion,Object Detection And Tracking Based On UAV

In order to realize the goal that UAV can reconnoitre interesting objects in all-weather and at day and night,we need to fuse the infrared and visible images captured from the same scene,so as to get the fused image with richer information;then,we need to detect and recognize objects in region of interest,and finally track target.Because of short control period,the visual algorithms should be real-time in order to keep pace with the control system of UAV.In this paper,image fusion,objects detection and objects tracking of UAV will be studied on the base of computer vision and flight control theory.The main content and innovation is as follows.1.Based on IFBSC algorithm,CU-IFBSC algorithm has been proposed and realized in this paper.(1)IFBSC can only fuse images with single channel,while CU-IFBSC can fuse images with 3 channels such as RGB images;(2)CU-IFBSC algorithm has accelerated by around 20 times with respect to IFBSC.(3)IFBSC can only process images with the same resolution while CU-IFBSC based on affine transformation can fuse the original images captured from visible and infrared cameras whose resolution are different.(4)CU-IFBSC algorithm has been realized on UAV embedded platform.2.SSD algorithm,which can detect objects(such as people and cars)has been realized on UAV embedded platform.By contrast,the valid detection ratio of fused video is more than that of visible video by 35.6%,and is more than that of infrared video by 88.4%;while of fused video is greater than that of visible video by 23%,and is greater than that of infrared video by 24%.3.A new object tracking algorithm which can choose the tracked object autonomously is proposed and realized based on TLD and SSD algorithm and we named it SSD_TLD.We use SSD_TLD to process visible video,infrared video and fused video respectively,and contrast the result.It is found that the successful tracking rate of fused video is greater than that of visible video by 13.3%,and greater than that of infrared video by 31.9%.