Research on An Autonomous Tunnel Inspection UAV based on Visual Feature Extraction and Multi-sensor Fusion Indoor Navigation System

UAVs have been widely used in line erection traction and inspection of outdoor high-voltage power grids due to its flexibility and low cost. Most of the inspection systems inside underground cable tunnels now adopt track-type and trolley-type robots, which are stable and safe, but also have disadvantages of high laying cost, low utilization efficiency, sensitive to terrain and so on. A UAV system for autonomous tunnel inspection based on visual feature extraction and multi-sensor fusion in GPS-denied environments is proposed in this paper. Kalman filter and GCC1 fusion method is the algorithm utilized for UAV indoor positioning using the optical flow sensor and the UAV inertial sensor. And then the flight stability was further enhanced by attitude decoupling. For visual feature extraction, this paper used the LAB color gamut threshold segmentation method to extract the foreground, the Hough transform matching straight line to extract the features of the horizontal deviation and forward direction, and designed a method to separate straight lines, curves and corners, thereby improved the UAV's perception of the tunnel environment. A series of inspection processes such as steering, height change, hovering and return voyage mission with vision feedforward PID controller were designed as tunnel inspection tasks. The on-site flight verification was conducted in the power transmission underground tunnel. Experimental results show that this system has the advantages of light weight, low cost, reliability, highly efficiency, low operation and maintenance costs. It can be deployed quickly and efficiently, and has certain engineering application prospect in the autonomous tunnel inspection.