A Feasibility Study of Real-Time Image Processing Techniques for Small Flying Object Detection in Drones

Drone usage is increasing significantly in our daily life, from military to delivery purposes. Although drones are also used to detect objects by using different techniques, they are limited to detecting flying small objects such as birds and responding quickly not to cause unintended collisions while flying at high speed. In this paper, we investigate the feasibility of using machine learning and image processing methods in drones while detecting birds mid-flight and responding to ensure their safety. This Real Time Bird Detection system (RTBD) is designed to detect birds so that proper response or evasive action can be taken by the drone. To avoid erroneous responses and observe the auto-behavior of drones while acting not to collide, we have developed an application with a graphical interface to easily control the drone’s video feed and process that information using a machine-learning model. The application also has the capability to detect if a bird is close enough to interfere with the drone’s flight path. Our test results show that the drone identified bird images within a 50-millisecond window of time, with Precision exceeding 96%, when Confidence exceeded 80%.