Research on Marathon Directory Recognition Method Based on Deep Learning
In the identification task of athletes with abnormal behaviors,the identification method based on number plate is adopted to complete the identification task of number plate through two parts:regional location of number plate and number text recognition.Aiming at problems such as tilt distortion and small changes in length-to-width ratio,a clustering method is combined with rRetinaNet based on rotating target detection model to match multiple forms of number tags,so as to improve the detection accuracy of targets.In the stage of number text recognition,the text image is obtained by clipping the corner coordinates,and the text recognition algorithm based on CRNN combined with attention mechanism is used to recognize the number text,so as to improve the recognition performance of irregular number text.Through text synthesis algorithm,the image of approximate number text is generated for network pre-training,and the real number plate image is used for fine tuning of the pre-training model.The experimental results show that,on the self-created real marathon data set,the improved compound network has improved the detection and recognition accuracy of the number book,and the combination of these two algorithms can effectively solve the problem of the identification of athletes with abnormal behavior.
万方数据
维普
