基于图像纹理的车辆检测算法

A New Vehicle Detection Algorithm Based on Image Texture

传统的视频检测方法是基于面阵图像的,其背景图像复杂,不利于目标分割和特征提取,而线阵CCD成像中背景图像相对简单,并且其帧速率远远高于面阵CCD的帧速率,从而可以实现高精度的检测,特别是车辆存在和车辆速度的检测。本文提出了适用于线阵CCD图像的车辆检测算法。其基本思想是:利用小波变换提取路面的纹理特征,然后对采集的每线数据进行二值化,并在此基础上,逐线进行车辆分割。实验表明:该算法能有效抑制车灯和阴影对车辆分割的干扰,实现车辆的实时准确分割。

Traditional video detection algorithms are based on plane array CCD images that have complex background, so it＇s not beneficial for object segmentation and feature extraction. While the background of linear CCD images is relatively single, and the frame rate of linear CCD is far above the frame rate of plane array CCD. Linear CCD can complete detection with high precision, especially in vehicle existence and vehicle speed. A new vehicle detection algorithm based on linear CCD images is proposed. It includes three parts, Firstly, texture feature of freeway is extracted using wavelet transform. And then vehicle segmentation is performed on the result of binary conversion line by line. The results of experiments prove that the algorithm is effective to reduce the effect brought by car light and shadow on vehicle segmentation and can work in real-time.