Object Detection and Recognition Using Deep Learning

Many real-life machine learning applications are increasingly guiding into focus on object detection and recognition. The traditional computer vision approaches do not achieve the needed accuracies. Deep learning-based approaches have achieved high accuracy levels raising the interest in such approaches in recent years. License plate detection and recognition have been extensively studied over the decades. However, more accurate and national/language-independent approaches are still in the focus of today's demand. In this book, we discuss an approach to detect and recognize multinational and multilingual license plates. The approach has four modules and each module is implemented using convolutional neural network architecture. The YOLOv2 detector with ResNet core network is utilized for license plate detection module. Faster R-CNN detector with a custom core network architecture is used for character segmentation module. Low complexity convolutional neural network architectures for license plate classification and character recognition modules are analyzed and studied. Each module is trained and tested separately and used to build end-to-end license plate recognition system.

Autorentext

Mohammed Salemdeeb received B.Sc., 2004 and M.Sc., 2011 in Elect. Eng.\Comm. Syst. from IUG, Palestine, and PhD in Electr. & Comm. Eng. from Kocaeli University, Turkey, 2020. His research interest fields are Signal & Image Processing, Deep Learning, ITS and Wireless Comm. He is R&D group leader at Autonom-Turkey and IEEE student branch counselor.