停车管理中车牌识别技术的研究

Research on key methods of roadside parking based on deep learning

本文研究了基于视频的路边停车管理中车牌识别问题,提出了动态车牌识别算法。在视频场景下基于打分机制选取最佳结果,充分利用视频的特性提高车牌识别率,有效地解决了车牌遮挡导致无法识别的问题。对于单帧车牌的识别,使用场景文本识别算法EAST定位车牌,再通过投影变换进行矫正,以提高在倾斜状态下的识别率。实验证明该方法对倾斜扭曲车牌的识别率有一定提升。

This paper studies the problem of license plate recognition based on video in roadside parking management and proposes a dynamic license plate recognition algorithm.In the video scene,the best result is selected based on the scoring mechanism,and the license plate recognition rate is improved by making full use of the characteristics of video,which effectively solves the problem of license plate occlusion leading to unrecognition.For the recognition of single frame license plate,the scene text recognition algorithm EAST is used to locate the license plate,and then the projection transformation is used to correct the license plate,so as to improve the recognition rate in the tilt state.Experimental results show that this method can improve the recognition rate of skewed license plate to a certain extent.