基于视频的交叉口排队过程感知及预测

在研究交叉口排队过程中，全面、精细地感知及预测方法一直是难点. 传统的排队过程感知方法缺乏不同排队阶段的精细感知和关联，无法得到全面、精细的交通参数. 对此，提出一种基于视频、分阶段感知再关联的排队过程感知及预测方法. 该方法包括3 部分：第一，通过目标检测和边缘检测提取道路和车辆信息；第二，对道路和车辆边缘信息进行卷积融合，检测排队区域和排队长度；第三，利用车辆位置重构和KCF跟踪等方法感知并关联不同排队阶段，重构并预测交叉口车辆运行轨迹，得到精准到车辆个体的排队车辆数和停车延误等参数. 实验结果表明，所提方法在检测正确的情况下，平均95.7%的排队车辆轨迹重构结果是准确的；并且在实际应用场景测试中，在不同车流量和光照条件下均可取得满意的效果.

Queueing Process Sensing and Prediction at Intersection Based on Video

Queueing process consisting of queueing, accumulation, and disappearing phase plays a very important role for the analysis and optimization of traffic management and signal controls. However, most of existing methods are lack of detailed sensing and correlation between different queue stages, which leads to uncomprehensive and rough detection in traffic parameters. In this paper, we present a video- based method, by sensing in stages first and correlating later to sense and predict queueing process. Firstly, the detection algorithm of objects and edges are adopted to obtain information of road and vehicle. Secondly, fusing road information and vehicle edges to detect queueing area and queueing length. Finally, reconstructing vehicle trajectory by sensing and correlating different queue stages, which are divided by queueing area. On this basis, the various parameters are calculated with vehicle trajectory, and the experimental results show that the reconstruction accuracy reaches 95.7% when there are no detection errors, in addition, proposed method is robust in detecting various traffic parameters in practical applications.