Research On UAV System Based On Visual Detection And Tracking

With the development of the automatic control, Unmanned Aerial Vehicle(UAV) systems have received widespread attention with its extensive application prospect. Especially in recent years, the domestic civil UAV market develops rapidly. The design of ground missions UAVs is one of the open areas need to be further researched. This paper focuses on a four-rotor unmanned aircraft aiming to search and track the intelligent ground vehicles by utilizing visual processing algorithms, embedded software, aircraft control, etc.Firstly, to complete the detecting and tracking missions, a platform including UAV and ground station was established. The UAV is with an ODROID-U3 ARM development board, which built Linux OS, ROS extension framework, and Open CV visual environment. The programmings of image acquisition, image processing, UAV control, the communication module are accomplished in embedded environment. The remote control system is programmed on the ground station. As the target of searching and tracking, the ground vehicle moves through a path which is not foreknown by the UAV. The tracking mission is totally achieved by visual recognition.Secondly, the visual recognitions of artificial marks and natural target are invested respectively. For artificial marks, an adaptive color segmentation algorithm is proposed. Then, the target will be located according to the extracted and matched outline. This algorithm is capable of working in different lamination conditions. For natural targets such as vehicles, this paper compares the three feature detection algorithms: SIFT, SURF, and ORB and then choose ORB method for its real-time performance and wild detectable features. The ORB points in real-time images are compared with templates and the noise points are eliminated according to the geometric distribution of macheted area. Finally, the detection of targets is completed. In addition, some preliminary research on target detection ba sed on AdaBoost Cascade classifier was carried out, the results show that in some cases, this method has a good performance.Then, target tracking technology is studied, which includes the multiple feature fused Camshift algorithm, and the research and improvement of LK optical flow method. Multiple feature fusion algorithm makes the Camshift algorithm has better anti-jamming properties for color close background; the improvement of LK optical flow method makes it successful to track feature points wtih big movements. Kalman filter is applied to avoid tracking failure problem for the causes such as short-term shelter.Finally, this paper analyzes the simplified control model of UAV, coordinate transformations of positioning data combined with the attitude da ta. Moreover, we design PID controllers for the UAV system, and successfully complete the tracking car experiments.