城市道路排队车辆检测方法

针对城市道路环境下的排队车辆检测问题, 提出一种基于边缘信息和局部纹理特征的综合检测方法。根据交通环境的特点, 对比5种不同边缘检测方法的性能, 采用Canny算法提取边缘信息, 采用改进的LBP方法提取纹理特征, 得到车辆的综合检测结果, 提取车辆排队长度和车道占有率等交通参数。分别采用综合检测方法、高斯混合模型和帧差法处理快速路、交叉路口、阴雨天气、光线突变、大雪天气、浓雾天气等场景下的视频图像, 并采用ROC曲线对检测性能进行量化评价。分析结果表明: 在快速路和大雪天气场景中, 3种方法检测性能基本相似, 最佳检测率分别接近90.0%和60.0%, 虚警率分别不超过5.0%和10.0%;在交叉路口场景中, 3种方法的最佳检测率分别为77.1%、31.5%、13.6%, 虚警率分别为16.5%、3.2%、19.0%;在阴雨天气场景中, 3种方法的最佳检测率分别为65.2%、3.0%、62.4%, 虚警率分别为10.5%、5.0%、56.5%;在光线突变场景中, 3种方法的最佳检测率分别为62.0%、18.9%、39.7%, 虚警率分别为10.8%、55.1%、36.0%;在浓雾天气场景中, 当能见度较低时, 3种方法的检测率和虚警率均接近于0。

Detection method of queuing vehicles on urban road

Aiming at the detection problem of queuing vehicles under the condition of urban road, a synthesis method based on edge information and local texture feature was put out. According to the characteristics of traffic enviroment, the performances of five different edge detection methods were compared, Canny algorithm was used to extract the edge information, and the improved LBP method was used to extract texture feature. The comprehensive detection result of vehicle was obtained, and the traffic parameters such as vehicle queuing length and lane occupancy rate were extracted. The proposed method, Gaussian mixture model and frame difference method were used to treat the video images of different scenes such as expressway, intersection, rainy weather, illumination mutation, heavy snowy weather and dense fog weather, and the quantitative evaluation of detection performance was carried out through ROC curve. Analysis result shows that under the scenes of expressway and heavy snowy weather, the detection performances of three methods are almost same, the best detection rates are close to 90.0% and 60.0% respectively, and false alarm rates are no more than 5.0% and 10.0% respectively. Under the scene of intersection, the best detection rates of three methods are 77.1%, 31.5% and 13.6% respectively, false alarm rates are 16.5%, 3.2% and 19.0% respectively. Under the scene of rainy weather, the best detection rates of three methods are 65.2%, 3.0% and 62.4%, false alarm rates are 10.5%, 5.0% and 56.5% respectively. Under the scene of illumination mutation, the best detection rates of three methods are 62.0%, 18.9% and 39.7%, false alarm rates are 10.8%, 55.1% and 36.0% respectively. Under the scene of dense fog weather, when visibility is lower, the detection rates and false alarm rates of three methods are close to zero.