3D OBJECT RECOGNITION SYSTEM FOR UAVS BASED ON KINECT AND ML

Abstract

Unmanned vehicles (smart cars, drones, robots) must understand and respond to their environment in order to perform their tasks. Therefore, they must have vision to determine objects and their coordinates. Kinect, an alternative to computer vision, was used to detect objects and measure their distance. The results showed that the Kinect sensor can, together with machine learning algorithms, determine the presence of objects in its field of view and measure the distance to them. Thus, it is justified that Kinect can be installed on unmanned vehicles as a vision sensor instead of a conventional video camera, as well as on manned vehicles to notify drivers, but mainly computer vision plays a crucial role in vehicles without human intervention. The article presents a technical vision system for a drone using Kinect, instead of a conventional camera to process streaming video. The system allows the drone to perform various tasks, such as detecting and tracking objects, building a map of the environment, planning a trajectory and avoiding obstacles. The system consists of three main modules: an image processing module, a localization module and a control module. The image processing module is responsible for acquiring and analyzing kinetic data, the localization module is responsible for determining the position and orientation of the drone in space, and the control module is responsible for generating commands for the drone's motors. The system was tested in real conditions and showed good results in performing the assigned tasks.