Optical Flow Obstacle Avoidance for UAV

To improve UAV safety and navigation, we're developing a sense-and-avoid system that detects any dynamic object without needing specific labels for each. This unsupervised approach focuses on the movement characteristics to identify potential obstacles. This solution adapts to a wide range of unforeseen objects, enhancing collision avoidance and flight path planning.

Our research delves into optical flow techniques to determine their effectiveness in obstacle detection for UAVs. We compared sparse and dense optical flow methods. We note sparse optical flow's speed and efficiency versus its lower detail accuracy, while dense optical flow provided higher accuracy but at a cost of increased computational demands.

Ultimately, we found that combining both sparse and dense optical flow, alongside cost-effective image analysis, presents a viable path forward. This hybrid approach balances accuracy and processing speed, making it suitable for real-time obstacle detection in diverse and unpredictable environments.