基于视频的行人车辆检测与分类

Pedestrian-vehicle Detection and Classification Based on Video

针对传统智能监控中行人车辆检测与分类算法存在目标分割不完整、分类准确率低等问题,提出一种基于视频的行人车辆检测与分类算法。利用领域信息动态调整置信区间构造混合高斯模型,采用卡尔曼滤波预测目标下一帧的位置。通过自适应EM聚类方法提取目标长宽比和面积作为特征,将目标分为行人和车辆。在模型估计过程中假设相邻帧目标做匀速直线运动,推导出目标面积变化满足线性关系,并对目标跟踪和分类进行修正,进一步提高检测准确性。实验结果表明,该算法的人车检测准确率达到90%以上,分类准确率达到80%以上。

Aiming at the problem of incomplete target segmentation and low classification accuracy of traditional pedestrian-vehicle detection and classification algorithm in intelligent monitoring,this paper presents a pedestrian-vehicle detection and classification algorithm based on video. The algorithm dynamically adjusts confidence intervals for constructing Gaussian mixture model using neighborhood information,and uses the Kalman filter to predict the position of the target in the next frame. It extracts the target aspect ratio and area through adaptive EM clustering as a feature,then divides target into pedestrians and vehicles. Assume that target makes the uniform linear motion in adjacent frame and derive the target area to meet the linear relationship change. Thus target tracking and classification can be modified to improve the detection accuracy in the end. Experimental result show that the algorithm detection rate is over 90% and classification rate is over 80% .