The Hardware Construction Of Soccer Robot And Research On Key Techniques Of Vision Under Complex Conditions

As yet, the research of soccer robot has achieved great accomplishments. In academe, RoboCup middle-size league exhibits its top level. The design of soccer robots needs many different courses: mechanics, kinematics, kinetics, control theory ,machine vision and artificial intelligence etc. The structure of soccer robot has three parts: basic motion system, motion control system and vision system. The vision provides a great deal of environmental information, it's the most promising sensor in the future. Autonomous machine vision is not only sensation, but also the explanation for sensed data. However, confined by complex conditions, many vision algorithms feasible in simulation experiments don't work in reality environment. The complex conditions faced by autonomous mobile machine vision include sensation and understanding complexity. The sensation complexity includs sensor errors, motion noises of mechanism, complex illumination noises, fancy shade of static landmarks, dynamic targets and background. The understanding complexity includes dynamic scene etc.The main work of the thesis is as follows:We have established soccer robot hardware platform, on which robotic programs can be performed and algorithms can be valued. The soccer robot hardware platform including two important structures which are omni-vision system and omni-directional motion structure. The omni-vision system is installed on the top of robot, and can obtain vision information of the whole field without turning about. The uni-vision system can see the front place of robot, and it provides useful information for controlling and shooting the ball.At the stage of shape recognition in image processing, strong noise condition is that shape contour is intended. While the least square method(LSM) can avoid the noise's error. Under strong noise, the average and uniformity of noise is unknown. In this paper, we propose a new adaptive method of strong-noise corner detection algorithm based on LSM. The method computes the residual sum of squares (RSS) of variable-length points set, detects corners at the minimum of the characteristic curve. This algorithm avoid the disadvantages of threshold methods. Meanwhile, it uses the half-division searching to reduce the computation complexity. Based on the one-corner detection method, we discuss how to reduce the time complexity on multi-corners recognition. Both the theoretical analysis and the experiment results indicate the validity of the method in increasing the veracity and reducing the time complexity of corner recognition of monotone edge on strong noise level.At the stage of object recognition in machine vision, the complexity is caused by complex illumination and covering, and by the complex relative position of dynamic objects. Complex conditions lead to incompleteness and uncertainty of features. For the anomalistic shape of gate caused by covering and shadowing, this paper present a omnidirectional radius scan method to detect gates and flagpoles based on the pantomorphism of omnidiectional reflector. According to the strong constraint that the object stands on the ground, this paper presents a unidirectional looking-down method to detect ball, and uses arc detection to eliminate the interference under complex conditions.This paper modularly designed RoboCup middle size vision system independently, totally about 10000 lines.