Vision and Control for UAVs: A Survey of General Methods and of Inexpensive Platforms for Infrastructure Inspection

Surveys vision and control methods usable on low-cost UAVs (feature tracking, optical flow, visual servoing, stabilization, planning) with a focus on infrastructure inspection. It sets up two railway inspection use cases and tests selected techniques on a small consumer drone: close-range target detection and positioning near structures, and long-range track following via the rails' vanishing point.

Key findings

• FAST keypoints with SIFT descriptors chosen for target detection; FAST/Shi-Tomasi ran under 10 ms per frame
• Rail detection ran at about 40-50 Hz including line extraction, with average vanishing-point error below 2.5% of image width
• Laplacian edge detection chosen over Canny for fewer false detections

Limitations

• Parameters tuned on the authors' own small image sets (50 and 165 images)
• Consumer GPS (~2 m) judged too coarse for safe flight close to structures
• No general obstacle-avoidance solution; battery and regulation limit operations