Akademik Veri Yönetim Sistemi
7th International Conference on Smart City Applications, SCA 2022, Castelo Branco, Portekiz, 19 - 21 Ekim 2022, cilt.629 LNNS, ss.463-472
To cope with modern requirements, recognition of combined multi-digit numbers are necessary. State-of-the-art methods have used object detection models to recognize the individual digits and this way makes the automatic multi-digit detection/recognition techniques not effective, not certain and not stable. For this reason, it is necessary to apply a learning-based method/system to detect the multi- digits. In this study, YOLOv3-tiny and YOLOv5 based real-time multi-digit number detection/classification system has been designed. Images have been taken from the camera, the multi-digit number's region has been localized and finally, the multi-digit recognition process has been performed. In cases where more than one digit number is sent to the system, it is possible to find out how many digits this multi-digit number has by processing the probabilities of what each written number may be, separated by deep learning. MNIST and Street View House Number (SVHN) datasets have been used in the experimental study. The results have been analyzed comparatively and the training and test results have been given.