Drone Based Object Tracking with Camshift Algorithm and Neural Network

Integration of tracking system and the drone has been a novel research topic in recent years, especially when drones are required to implement complex tasks which cannot be done easily with human control. Usually, the drone uses camera to gather full information about environments. The main processor calculates all necessary trajectories for drones. However, such tracking methodology does not apply to real-world problems mostly due to the complexity which surrounded the environment. For example, when a large number of persons gathered closely together， the tracking system is difficult to find people that you want to track down. In addition, the similar background color is also one of the main reasons the system can't track people. This thesis proposes an approach that combines camshift algorithm and neural network to track the object. This approach is more cost-efficient and environment-adaptive compare to the previous approaches which use traditional tracking algorithms. Our model altered the previous Yolo neural network, combined camshift algorithm, and neural network. Results show the new approach is faster than Yolo neural network. On the other hand, it solves the problems of occlusion, illumination, scale, and noise in camshift algorithm.