Control de un robot mediante técnicas de visión

Abstract

In this project, a robotic system provided with computer vision is designed and implemented. The robot is capable of detecting and tracking a specific target in a certain environment. The robot is designed according to the paradigm of behaviour-based robotics, and implemented under the Lego Mindstorms hardware. Due to the significant increase of mobile devices capabilities, the artificial vision system has been implemented through an Android device, using the OpenCV4Android library as a support for image processing. Regarding the integration of the two devices in one system, it has established a Bluetooth comunication as data exchange channel. Specifically, the mobile device is responsible for providing the data concerning the position of the object in the image to the robot. Thus, the capabilities of the robot have been extended by integrating the camera as another sensor of the NXT. A comparison was made between using a detection method in each frame to perform the tracking of the target, and the combination of the detector with a specialized tracking method. It has been observed that by using the Lucas-Kanade tracking algorithm once the target have been detected, a considerable performance increase is obtained.