Recognition Of Athlete Number Plate Based On Deep Learning

With the widespread use of smartphones and digital cameras with high-definition cameras,the acquisition of images becomes more and more convenient.In sport events,a large number of images are usually taken.Images containing the particular athlete are mainly obtained by identifying the athlete number plate.Recognition of athlete number plate includes three parts: number plate region location,number plate character segmentation and number plate character recognition.The location of number plate area is the primary work of athlete number plate recognition.Its main function is to segment the number plate area quickly and accurately from the complicated background.However,the location of athlete number plate is sensitive to the light environment.Under poor lighting environment,the quality of number plate images will be seriously affected,resulting in adhesion of the number plate area to the background,making it impossible to accurately locate the number plate area.After location,it is necessary to preprocess the number plate area image so as to obtain the binary image that only contains the number plate characters and does not tilt.The top and bottom of the number plate images usually have lots of advertisement information.After preprocessing,they appear as white area as the number plate characters.Their existence will seriously affect the accuracy of character segmentation and the effect of character recognition.Therefore,it is of great significance to design a number plate recognition method with good robustness.The existing methods of number plate area location are easily affected by light and the robustness is poor.To accurately locate the number plate area under a complicated photographing environment,a method based on deep learning to locate the region of number plate is proposed.The accuracy rate of athlete number plate location is up to 96.83% by testing 400 pairs of samples with 2305 number plates.Compared with morphological method,the accuracy rate of location improves 3.86%.Then the number plate is subjected to preprocessing such as graying,binarization,and morphological operations.To remove the advertisement information from the number plate,this paper improves the original morphological operation.Different from the original morphological processing method,the improved method connects the structural elements of different dimensions of the same shape and connects the structural elements of different shapes in parallel.The weight-adaptive calculation method is used to filter the number plate.Experiment shows that the improved method is efficient.Further,the number plate characters are segmented.Finally,BP neural network is used to recognize the segmented characters.Experimental results show that the method proposed in this paper can quikly and accurately recognize the number plate.