Research And Design Of Global Vision In Small-Sized Intelligent Soccer Robot System

The year of 1997 is remembered as a turning point in the history of artificial intelligence and robotics.The fifth month of this year, IBM's supercomputer called "Deep Blue" defeated a man who was a chess champion in the world, which means the challenge in artificial intelligence lasted for 40 years has finally been successful. After that, Soccer Robot was established to promote researches and education on artificial intelligence and intelligent robot technology as a standard issue. Its purpose is to promote the development of related fields by technology competitions in the same stage. Soccer Robot is a frontier research and technology integration involving artificial intelligence, robotics, communication, sensor, image treatment and many other areas, which is typical in researches.Vision subsystem is one of four subsystems of soccer robot.All information about environment are obtained by the vision subsystem during the competation, therefore it is so important for the whole system.Vision subsystem mainly deals with instability in the effect of image recognition caused by variations in color informations and lighting conditions; barrel distortion produced by cameras requires vision subsystem robust ; small size soccer robots who moved in high-speed need to process images in real-time; meanwhile, informations provided by vision subsystem must be accurate for the precise position and orientation of soccer robots.Based on small size RoboCup, a whole design scheme of the system was put forward in this pape after the analysis of functions and performance requirements on vision subsystem. In order to meet these requirements, vision subsystem would be divided into three functional modules to study respectively including distortion correction, image segmentation, and object recognition. The image segmentation technology based on BP neural network has improved the recognition rate of vision subsystem in unstable light conditions. The processing speed of target identification has been increased by areas segmentation recognition algorithm of global scan based on subtracted background. Main contents of this paper are as follows: (1) In the module of image segmentation, color image segmentation technology based on BP neural network of HSI color space has been proposed, and good segmentation results has been achieved in unstable lighting conditions.(2) A better correction effect has been prepared for the precise location concerning to goal after using two-step calibration method of Tsai in the module of camera calibration and distortion correction.(3) The processing speed of target identification has been greatly improved by targets recognition algorithm of connected domain based on modeled background.The experimental results of overall operation show that the scheme and algorithm design have been in line with basic requirements of robot soccer, meanwhile its vision subsystem has been greatly improved in light adaptability, accuracy and real time.