基于匹配分布和混合高斯模型的车辆检测算法

为解决高斯混合模型GMM（Gaussian Mixture Model）在车辆检测中存在的车辆断裂等问题, 提出了一种基于匹配度分布的混合高斯车辆检测算法。该算法采用c均值聚类法计算混合高斯模型初始值, 得到初步的背景模型； 匹配度分布的提出充分考虑了背景变化的时间性和空间性的特性； 根据前几帧检测结果得到每个点的匹配度分布, 对当前图片改变背景学习的规则, 去除了干扰, 适应了背景的变化。实验结果表明, 该算法较传统的混合高斯检测方法检测率平均提高16%以上, 使背景也更稳定和准确, 克服了车辆检测的断裂以及光照突变等问题, 提高了车辆区域检测的准确性。

Vehicle Detection Algorithm Based on Distribution of Matching and Gaussian Mixture Model

Aiming at solving several problems in vehicle detection,a GMM (Gaussian Mixture Model) detection algorithm based on distribution of matching rate was proposed. The algorithm got the initial value of GMM with c meanclustering method, which formed the original background model. It fully took the timeliness and spatiality into consideration by proposing the concept of distribution of matching. And it changed the learning rule of background of current frame to eliminate interference and adapt to the change of background according to the distribution of matching rate corresponding to the former several frames. The experimental results indicate that the algorithm improves the detection rateby at least over 16%. The updated background is also more stable and accurate. It overcomes the problems of fracture of vehicle detection and saltation of illumination and improves the accuracy of detection of vehicle region.