Fused faster RCNNs for efficient detection of the license plates

Abstract

Automatic License Plate Detection and Recognition (ALPD-R) is an important and challenging application for traffic surveillance, traffic safety, security, services purposes and parking management. Generally, traditional image processing routines have been used in ALPD-R. Although the general approaches perform well on ALPD-R, new and efficient approaches are needed to improve the detection accuracies. Thus, in this paper, a new approach, which is based on fusing of multiple Faster Regions with Convolutional Neutral Network (Faster- RCNN) architectures, is proposed. More specially, the Deep Learning (DL) is used to detect license plates in given images. The proposed license plate detection method uses three Faster- RCNN modules where each faster RCNN module uses a pre-trained CNN model namely AlexNet, VGG16 and VGG19. Each Faster-RCNN module is trained independently and their results are fused in fusing layer. Fusing layer use average operator on the X and Y coordinates of the outputs of the Faster-RCNN modules and maximum operator is employed on the width and height outputs of the Faster-RCNN modules. A publicly available dataset is used in experiments. The accuracy is used as a performance indicator of the proposed method. For 100 testing images, the proposed method detects the exact location of license plates for 97 images. The accuracy of the proposed method is 97%.