基于视频码流的交通标志检测

Traffic Sign Detection Based on Video Encoding Stream

为提高自然场景下交通标志检测实时性差的问题,提出了一种基于视频码流的交通标志检测方法。此方法充分利用编码块像素之间的相关性和码流中能体现的不同交通标志颜色特征和边缘特征,通过分析视频码流帧内预测模式、像素残差分布情况值和像素残差信息,对交通标志进行目标粗定位;通过分析视频码流帧间预测码流预测模式、像素残差分布情况值和运动矢量信息,对交通标志检测进行修正。此方法避免了视频解码过程中整数IDCT变换、反量化、重构和环路滤波等耗时的操作。基于中国交通标志数据库(CCTSDB)的实验获得了96.9%检测率和4.3%误检率,该方法检测时间较短,可满足实时检测要求。

In order to improve the poor real-time performance of traffic sign detection in natural scenes,a traffic sign detection method based on video stream was proposed.This method makes full use of the correlation between the pixels of the coding block,and the different color and edge characteristics of traffic signs that can be reflected in the code stream.By analyzing the intra-frame prediction mode,the value of coded_block_pattern,and the pixel residual information,the traffic signs are initially located.By analyzing the prediction mode of the inter-frame,the value of coded_block_pattern,and the motion vector information,the traffic sign detection is corrected.This method avoids time-consuming operations such as integer discrete cosine transform(IDCT),inverse quantization,reconstruction,and loop filtering in video decoding process.The experiment based on the China Traffic Sign Database(CCTSDB)obtained 96.9%detection rate and 4.3%false detection rate.The detection time of this method is greatly shortened and can meet the real-time detection requirements.