Research On Soccer Robot Whistle Recognition Method

As the robot becomes more and more intelligent,it is no longer a fantasy for the robot to complete a football game.In this background,it is important to add a whistle recognition module to the existing robot function module,so that the robot can distinguish the referee's instructions independently.In this thesis,we investigate whistle recognition technology,specifically,realize whistle signal collecting and preprocess,improve feature extraction and recognition models.The whistle recognition is realized on an ARM platform S32440.Firstly,basic theory of whistle recognition is studied,including front-end processing of whistle signal processing,endpoint detection,feature parameter extraction method and recognition algorithm.Then,some basic concepts of voice recognition technology are introduced in detail,including the preprocessing process of sound signal,the extraction process of feature parameters,and the recognition model commonly used in voice recognition.The principles of these recognition models are analyzed and compared.Furthermore,the indicators of the voice recognition system are listed.Then,the recognition process of the whistle is specifically realized,and the process of whistle recognition is improved on the basis of analyzing the uique characteristic of whistle.The template library is setup,providing the basis for the subsequent DTW algorithm and improvement.Based on the characteristics of the whistle,a preprocessing process more suitable for the whistle is selected.The feature parameters,namely the Mel frequency cepstral coefficient(MFCC),are extracted.Finally,according to the dynamic time rounding(DTW)principle,the recognition algorithm is designed and simulated,and the whistle recognition is realized initially.At the same time,anti-noise measures are taken to improve the recognition accuracy.Finally,above-mentioned procedure is experimentally verified.The whistle samples recorded respectively in a quiet and noisy background are extracted and recognized.And also,the correct ratio of recognition is counted.