UC San Diego

Object searching is one of the most popular applications of unmanned aerial vehicles. Low cost small drones are particularly suited for surveying tasks in difficult conditions. With their limited on-board processing power and battery life, there is a need for more efficient search algorithm. The proposed path planning algorithm utilizes AZ-net, a deep learning network to process images captured on drones for adaptive flight path planning. Search simulation based on videos and actual experiments show significant reduction in search time under certain circumstances, compared to traditional linear search method. The thesis will discuss important design tradeoff between performance and battery life.