Research On Algorithm Of Monocular Vision-based Collision Avoidance System For UAV

With the continuous development of aviation science and technology, in order to perform the daunting tasks that are too difficult or dangerous to human pilots, the development of Unmanned Aerial Vehicle(UAV) has been widely concerned. For many civilian applications and tasks, UAVs are required to navigate in unknown low-altitude terrains where obstacles of various types and sizes exist. According to the statistics of aviation accidents, usually collisions occur because UAVs flying at low altitude fail to identify thin objects such as power lines. Therefore, in order to ensure the implementation of the task, it is critical that UAV system can recognize and avoid the power lines in the flight path.Monocular vision system has the characteristics of small volume, low power consumption,good concealment, cheap et al. This paper mainly research on preprocessing method and power line identification, meanwhile, the UAV's autonomous avoidance algorithm for distance estimation based on the researching on image matching from monocular vision is studied.According to the actual situation of aerial image obtained by the UAV image collection system, the preprocessing method is studied. By using optical correction, it can be improved the brightness unbanlanced and contrast descending of the images. Due to the motion blurring and the noise jamming in image, Weiner filtering algorithm and an improved wavelet de-noising which based on edge detection are used to enhance the quality of image restoration. According to the characteristics of power lines, we put forward the method which based on Harris feature clusters, Hough transform and power line classifier for power line extraction and identification. Find-Union based clustering process is implemented to obtain Harris feature clusters, which conform to the characters of lines. Hough transform and Segment-Tree based line growing process is adopted to acquire line ‘candidates', which can be power lines or noise lines. Then identify the power lines and give up the noise lines by the power line classifier. As the preliminary steps completed, we can extract the Harris feature points on the power lines of the image,implement Harris feature matching based on the IMED, estimate the distance between the UAV and the power lines through the relationship of the movement of the feature points on the image and the speed of the UAV. The experimental results show that the algorithm not only can effectively identify power lines, also can estimate the distance of the UAV and the power lines. In this paper, the UAV motion speed and distance model is proposed so that it can determine whether the UAV is in dangerous conditions through the distance parameter obtained by the result of distance estimation, according to the different status of the risks, the UAV can take the corresponding control strategies, such as security, vigilance and deceleration, to ensure the UAV avoid the power lines in time.