MobileNet-SSD MicroScope using adaptive error correction algorithm: Real-time detection of license plates on mobile devices

Abstract

At present, many deep learning methods have been applied widely in the field of vehicle and license plate detection. These methods are quite effective in detecting large objects (like vehicles, pedestrians etc.). However, detection for small objects, such as license plates, is not ideal, especially in complex scenes such as license plate occlusion, so that it is difficult to meet the real-time requirements of the industry. To realise efficient real-time detection of small license plates on mobile devices, this study proposes a lightweight model for small object detection, named MobileNet-SSD MicroScope (MSSD MS). This model improves the accuracy of license plate detection, enhances the anti-interference capability and can be implemented in real time on the mobile device RK3399. Besides, for false detection objects, an adaptive error correction algorithm is proposed to reduce the false detection rate, which improves the precision rate of license plate detection and is adaptive to various scenes. The experimental results show that compared with MobileNet-SSD, MSSD MS using the adaptive error correction algorithm has stronger robustness, which can significantly improve the effect of small license plate detection and meet the requirements of real-time detection on mobile devices.