Humanoid Robot Vision System Implementation

Since the first birth of a robot, robot technology and intelligent levels have made great progress, due to its combination of mechanism, automatic control technology, artificial intelligence, microelectronics, communication technology, bionics, sensor technology, etc. multi-discipline, so it caused around the world attach great importance to research institutions and universities. In particular, the United States, Germany, Japan and other industrial powers have made remarkable achievements in robotics research areas. RoboCup, as the world's largest exchange platform of robot technology, has played a huge role in promoting the development of robot technology. Humanoid Robot, with its humanoid modality, will come into home and take the place of human being to do dangerous work in future. Therefore, research on humanoid robots is not only academic significance, as well as a high application value. This paper studied the robot's vision model under the humanoid robot platform.In this study, we described the design and implementation of PC-side visual debugging system through analysis of the humanoid visual model. The debugging system consists of camera acquisition module, the network transmission module, manual threshold setting module, camera parameters setting module, and feature color table saving module. Camera acquisition module is mainly responsible for reading image files from the camera; network module transmits the images in the camera from the robot's host computer to a PC debug terminal; manual threshold setting module is responsible for manually extracting features color from obtained image; camera parameters setting module is responsible for adjusting the camera parameters in order to more easily identify the markers. Feature color table saving module is responsible for extracting the feature color table to compress and store in accordance with the Run-Length EncodingThis article described the design and implementation of the robot host computer vision subsystem, and how to identify the ball, the goal and landmark poles from the background in detail. We deduced the formulae of image coordinate system with the robot's own coordinate system, and the formulae can be used to accurately position the ball marker in the carpet. Finally according to the robot camera model and site characteristics we proposed a fast extraction of linear and self-positioning algorithm, which can complete their positioning through one point observation. It expanded vision in another way, at the same time, it saved self-positioning time.