OBJECT DETECTION AND VISUAL TRACKING SYSTEM

The aim of the project is to design visual object detection and target tracking system. Such a system has been extensively used in military applications, but these days they find use in the civilian domain too, such as fault detection, geological surveys, remote sensing and domestic policing. Object detection and target tracking systems find use in robots and unmanned vehicles. In order to navigate and perform various operations on its own, these vehicles need to have intelligence. Therefore, developing an algorithm that would enable the robotic device to act on its own and make necessary changes to its trajectory would be a major breakthrough. Many algorithms have been developed to achieve this objective but they usually run in a pre-defined environment. In this project it is assumed that the environment is unknown and the drone has only visual data obtained by its frontend camera. The algorithm for object detection is based on the concept of determining image contours and utilizing the principle of image moments. Contours help detect the object and moment of the object is used to determine its size and therefore, its distance from camera. The algorithm for tracking has been developed using the Shi Tomasi algorithm for corner point detection and the pyramidal Lucas- Kanade optical flow algorithm to determine the motion vectors of the corner points. This is followed by the calculation of the local scale variance of these points and then estimating the time to collision. With these results the robotic device would be in a position to change its direction or trajectory. Therefore this algorithm performs real-time image processing of visual data provided by the drone and gives some kind of visual intelligence to the drone.M.S. in Electrical Engineering, December 201