Vision-Based Obstacle Avoidance For Power Lines Inspection With UAVs

Recently, the commercialization of small multi-rotor UAVs(unmanned aerial vehicles) has been taking off. In the field of high-voltage power lines inspection, the operating of UAVs can achieve higher efficiency and minimize the need of labor. To prevent the accidents of UAVs hitting the power equipment in unexpected situations, this paper has presented an obstacle avoidance algorithm and video remote monitoring system for UAVs based on the binocular stereo vision. Obstacle avoidance algorithm can achieve distance information of obstacle accurately in long-baseline binocular vison system, which can provide operation basis for flight control system. Taking the VR(Virtual reality) devices as terminal, video remote monitoring system can inspect power equipment scene obtained by short-baseline binocular vision airborne system in real-time.In this paper, the main contents are as follows:(1) Binocular vision system designing and stereo calibration. To complete the work of obstacle avoidance, this paper has designed a large-scale calibration board and long-baseline binocular vision system so as to perceive distance information accurately. The long-baseline system was used for photogrammetry in a long-distance range, which verified the validity of the calibration results and reliability of vision measurement method. In the video remote monitoring system, a short-baseline binocular vision framework, which is regarded as front-end video sampling device built up on the airborne platform, has been designed.(2) Binocular measurement algorithm designing and simulation for practical inspection scenario. Long-baseline binocular vision system which is set up vertically was used for image collection in inspection scene. This paper has analyzed how to extract counterpart points in the following two inspection scenario. a) Power lines inspection. Normally, there are no obvious feature points on the power lines. So, the power lines were detected in the first place. Secondly, the epipolar line is plumbed with forward direction of power lines in the vision system, which is a crucial property can be used to extract counterpart points. b) Pylons inspection. Different from some active depth perception methods such as laser ranging, ultrasonic and structured light, this paper has proposed a passive depth perception model, which is based on the optical flow theory, to perceive pylons in the front of UAVs roughly. Then counterpart points lied on the pylons were extracted by Scale Invariant Feature Transform(SIFT). The coordinates of counterpart points combined with calibration parameters can be used to calculate the distance information of pylons and power lines.(3) The designing of video remote monitoring system based on binocular vision system and hardware implementation. This paper has explained the key ideas and main functional modules(short-baseline binocular vision module, FPGA and image processing module with FPGA and video compression chip at the core, wireless transceiver module, virtual reality devices) of video remote monitoring system. Power workers can feel more vivid and inspect the sites in real time with this system.The research work and main results of this paper can be viewed as a positive attempt for robot vision methods applied to the obstacle avoidance with UAVs for power lines inspection.