Lane detection and tracking control for unmanned aerial vehicles using image segmentation

Abstract

Lane detection has been widely used in land-based vehicles to carry out autonomous driving. The same concept can be applied to aerial vehicles to enable autonomous maneuvering based on the detected runway. This opens the possibilities of automated capabilities such as path tracking, automatic landing, and automatic take-off.

However, the integration of real-time lane detection on an UAV is still an unconventional topic. Thus, this project aims to develop a robust runway detection system through the combination of image segmentation and computer vision techniques. The Convolutional Neural Network architectures used were E-Net and U-Net. Training data were self-generated and labelled using MATLAB image segmenter.

The optimized model weights were integrated with the DJI Tello drone using the DJITelloPy Application Programming Interface (API). Real-time detection was carried out to fulfill the autonomous actions such as path tracking, automatic landing, and automatic takeoff. This final year project serves as a study to future research on the integration of autonomous navigation and UAVs.