Navigation And Obstacle Avoidance For Miniature UAV Based On Binocular Stereo Vision

Navigation system plays an important role in autonomous flight of UAVs(unmanned aerial vehicles). The flight control system, sensor system and other electronic equipments are critical parts of the navigation system. The navigation system is designed to help UAVs detect and estimate their current flight status and position, so that the UVAs can develop the most optimal path planning and the most appropriate control strategy to ensure the UAVs to arrive at destination. Last century, research on the navigation of UAVs focused on the high-altitude areas, but with the expansion of the UAV’s application in low-altitude areas, such as search and rescue operations, recycling and hazardous substances, fire monitoring in forest, low-altitude UAV autonomous navigation gradually becomes a hot spot research.Binocular stereo vision simulates human’s eyes with which people perceive 3D space. Binocular stereo vision uses two cameras to capture the same scene at the same time. After image acquisition, stereo rectifying and stereo matching, disparity map is generated. Depth information of the scene can be calculated by the disparity map and then the 3D information will be obtained. Binocular stereo vision has hidden features and the ability to obtain comprehensive information, including 3D depth information. Due to the advantages above, binocular stereo vision will be widely used in the navigation system of UAVs.In this paper, we explore and study on the key technologies of UAV’s navigation based on binocular stereo vision. The main contents are as follows:(1) A stable binocular stereo vision system on the PC and embedded platform has been set up. The calibration of cameras is completed. This paper introduced an effective binocular vision system in which the baseline and angle can be adjusted independently.(2) In this paper, we made a comparison and analysis of a variety of existing stereo matching algorithms and also presented an improved matching algorithm and optimization strategies based on the semi-global matching algorithm which taking into account the matching accuracy and real-time computing.(3) For obstacle detection based on vision seems difficult in the fog, this paper did some meaningful work on image dehazing and a new improved single image dehazing approach which based on dark channel prior was introduced. This work will do a positive effect on UAV’s aerial and visual navigation.(4) This paper also proposed an obstacle detection method which combined growth characteristics based on the disparity of regional and another image segmentation method called Normalized Cut. It works a good result of obstacle detection. And the distance measurement of the obstacle was also finished.(5) Finally, we designed some experimentals on obstacle avoidance and also completed some work about 3D reconstruction, image stitching, route planning.