Visual-based Localization Algorithm Design on Humanoid Robot Soccer
|School||Zhejiang University of Technology|
|Course||Control Theory and Control Engineering|
|Keywords||HSI color model image segmentation image recognition vision-based localization|
In the last decade the fast development of robot technology along with its wide application in many fields, has further promoted the development of social productive forces and provided great convenience for the people.Vision system is the most important sense organ of the robot. The task of vision system is to collect surroundings information by their own (carry-on) vision sensors, to process the collected information in real time, and ultimately to achieve the objects localization and the robot self-localization. Due to its powerful function of perceiving surroundings, navigation, positioning and autonomous moving, vision based mobile robot has a broad application future and becomes the focus in the robotics research. Based on RoboCup humanoid league, the vision system is researched in this paper. The research includes image capture, image preprocessing, image segmentation, features extraction, image recognition, objects localization. The results have practical value for the robot application of industrial detection, biomedicine and intelligent navigation, and so on.On the base of reviewing current situation home and abroad, and analysis of the characteristic of the mobile robot, the image segmentation is emphatically studied in this paper and results are as follows:(1) Image Capture and Preprocessing. A CMOS color camera and DirectShow software is used to obtain original images which are in RGB color space. According to the characteristic of the RoboCup humanoid robot, comparing the different color models on theory and experiment, finally the HSI color model is chosen to be the base of image processing and converse the RGB images to HSI space as well.(2) Image Segmentation. Image segmentation is the focus of this research. Combining with the histogram thresholding algorithm and the K means clustering algorithm, a fast and adaptive clustering segmentation method is proposed. In addition, the scan line seed filling algorithm is improved to accomplish the color clustering process and meanwhile to obtain the description of interested color blobs. The simulation results show that the fast and adaptive clustering segmentation method has stronger advantages on real time ability and stability than the conventional fixed thresholding or K-means clustering algorithm.(3) Image Recognition. This process contains two steps: building the classification model; filtering and defining the descriptions of color blobs obtained by the image segmentation. Finally important objects such as the ball, goals and color columns are mapped to color blobs. Experimental results show that the algorithm has high accuracy and good anti-jamming capability.(4) Object Localization. Camera’s internal parameters are drawn by camera calibration algorithms based on linear model. Combining the pinhole imaging geometric model and robot’s physical model, a geometric localization method based on monocular vision is proposed. And object tracking is achieved by integrating the data of motion sensors. Software model is compiled and applied to the robot’s hardware platform and satisfactory effects were obtained, thus the theoretical part of the research is verified.