车载视频交通场景定位与文字识别方法的改进

Traffic Scene Position and Character Recognition Method Improvement Based on Vehicle Video

以行车记录仪作为车载视频传感器,提出一种基于车载视频的交通场景文字识别方法。采用Retinex算法、笔画宽度特征、级联分类器等技术方法选定文字候选区,使用投影法和连通域法将整个文字区域分割成单个文字,最后将单个文字送入训练好的CNN文字分类器进行文字识别。通过搭建测试系统、处理行车记录仪录制视频的实验可以得出,该识别方法对于交通场景文字的定位和识别精度比传统方法提高20%,在不同光照环境下有较好的鲁棒性。

In the paper,a traffic scene character recognition method based on vehicle video sensor using vehicle data recorder is proposed.Using the Retinex algorithm,the width features of the brushstrokes,and the cascading classifiers,the character candidate area is selected,using the projection method and the access method,the entire text area is separated into a single character.Finally,a single character is sent into the trained CNN classifier for the character recognition.Through constructing the test systemand the experiment of processing video recorded by vehicle traveling data recorder,it can be concluded that the identification method for traffic scene recognition and positioning precision is 20% higher than the traditional method,it has good robustness under different lighting conditions.