基于视频图像的多特征车位检测算法  

Parking Cell Detection Algorithms of Multiple Characteristics Based on Video Image

作　　者：蒋大林[1] 邓红丽[1] 平彧[1] 韦燕凤[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100022

出　　处：《北京工业大学学报》2008年第2期137-140,共4页Journal of Beijing University of Technology

基　　金：国家'八六三'计划资助项目(2003AA103960)

摘　　要：车位检测是停车场监控和管理系统的重要内容,针对停车场的复杂背景和不同环境光照条件,提出了多特征的车位检测方法.通过充分利用车位信息的几何特点和纹理特征来提取车位的特征参数,从数学建模的角度设计了3种不同判决函数的方案并采用实际现场的视频监测图像数据对设计方案进行了测试和比较.实验结果表明,3种方案在运算的实时性、识别的准确性及鲁棒性方面均能满足实际停车场车位的监控和管理系统的需求,其中以主成分分析降维的贝叶斯判别方案效果最佳,识别率高达99.17%.Parking cell detection is one of the key technologies in parking lot monitoring and management system, this paper proposes a parking cell detection algorithm of multiple video features, in view of complex background and different illumination conditions in the parking lot. Firstly, this paper has fully used the geometrical and statistical features of the parking cell information, and extracted the features parameter of parking cell, and then designed three kinds of different solution from mathematics modeling angle. Finally, many images, which are regarded as the parking cell detection images in different weather and conditions are have been tested by those proposed algorithms continuously for 14 days, and comparisons have been made with the results in the proposed algorithms. The experimental result indicated that the algorithm of Bayes methods with PCA is superior to other algorithms in aspects of operation time, recognition accuracy, and the robustness. Its detection rate reaches as high as 98.99 %.

关 键 词：分类器 特征测度 主成分分析 贝叶斯方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]