Object Tracking Platform for Color Object Detection using Genetic Algorithm Optimization

摘要

The aim of this work is to resolve typical tracking's challenge, which is object detection in the scene. In this context, a new robotic system to detect objects in unknown indoor and outdoor environment is realized. The developed robotic system is equipped with ultrasonic sensor and camera, gives visually information about the environment around detected object. Therefore, the realized system works in real time by analyzing camera data making it possible to detect color, size and position of the object. The aim of the proposed strategy is to find a compromise between robustness and processing speed of color detection based on rapid threshold, which produces a much smaller number of edge pixels compared to standard approaches based on a simple threshold. This reduction significantly reduces the number of votes required for robust real-time detection of object parameters. The approach consists of two main phases. In the first calibration phase, which takes place offline, a heuristic method of color classification using genetic algorithms is learned from a copy of an image of our colored environment. Then, in the real-time monitoring phase, the color classification is applied to the input images, the object is detected and its position is returned. To boost the detection accuracy the fixed and adaptive threshold methods are tested and compared with the genetic algorithm method, where the last method presents a good results. Accordingly, an exact determination of position, orientation of mobile platform, and accurate determination of color object in the environment is succeed.