AN INVESTIGATION OF AGGREGATE CHANNEL FEATURES OBJECT DETECTOR FOR UAS APPLICATION
Nataliia Kuzmenko, Volodymyr P. Kharchenko, Ivan Ostroumov · Proceedings of National Aviation University · 2019
Purpose: The paper is aimed to point out an artificial intelligence as a key priority for research and development issues. Consideration of existing methods of object detection is one of the important tasks of the paper. The represented research results are aimed to investigate a problem of moving object detection by visual sensor data for Unmanned Aerial System application. Methods: Represented approach is grounded on probabilistic and statistical methods of data processing, in particular on Aggregate Channel Features approach usage for object detection. Results: An Aggregate Channel Features approach for object detection by video-stream at different scenarios has been practically investigated. Results of experimental investigation of ACF usage for moving vehicles detection, such as cars and trams, indicate good performance characteristics. Also, a dependence between training time of detector and amount of object positive instances has been investigated for particular case. Discussion: Multiple advantages of Aggregate Channel Features object detector such as universality, simplicity of realization and good compromise between computation time and detection accuracy allow to use it in the tasks of people, vehicles, artificial and natural objects detection in UAS application. Represented results can be implemented in Unmanned Aerial Systems for searching and tracking of movable objects.