A Soccer Robot Target Recognition Algorithm Based On Autonomous Vision

Using visual sensors to achieve target recognition by embedded wheeled robots is currently one of the research hotspots.The key problem is how to design a target recognition method to help robots to complete the task of target recognition in complex and specific application scenarios quickly and effectively.At the same time,it can provide guarantee for wheeled robots to execute instructions more timely and move more sensitively in motion.In this paper,with a background of wheeled soccer robots in soccer competition,aiming at the problem of identifying elements such as football and soccer robot in specific application scenarios,the main work content of the following four aspects is completed in this paper:(1)In this paper,a target recognition method for soccer robot based on autonomous vision is designed,of which the algorithm combines Haar-like features,SVM and AdaBoost learning classification algorithms to complete the task of acquiring training sample sets and constructing strong classifiers on the server side.Through the strong classifier model,the target is traversed in the robot field of view.It reduces the dependence of robots on color coding and other aspects in target recognition.(2)A target recognition algorithm based on dynamic target region estimation is proposed.The three-frame difference method is used to mark the predicted region of the dynamic target.When the visual field image is traversed,the predicted region is detected first.This method further improves the real-time performance of the target recognition system.(3)A multi-sensor information fusion algorithm for wheeled soccer robot is applied in this paper.Through this algorithm,the differential and integral terms in the formula of PID algorithm are improved,the efficiency of fusion calculation is improved,and the output data is more accurate.Thus,it provides control-level support for accurate and fast robot target recognition.(4)The content of this research is to complete the target recognition of low-cost wheeled soccer robots with limited resources.The training of small sample sets can be done by ordinary computers and not necessarily with high performance computers.It has good recognition effect to meet the needs of football robot competition.The validation experiments of multiple groups of recognition targets are carried out for this paper.The results show that the recognition method designed in this paper has a good effect.In real application,it also significantly improves the real-time performance of the recognition process,which achieves good recognition in the case of non-static targets.Moreover,this method also has a good recognition effect when the robot is separated from the pre-recognition target by the occluded object or when the illumination is complex and changeable in the application.Besides,after collecting and training samples of common objects,the target recognition algorithm based on autonomous vision can also be applied to target recognition in daily life.