面向智能交通引导的车辆检测算法改进

An Improved Vehicle Detection Algorithm for Intelligent Traffic Guidance

设置可变车道可以提高现有道路资源的利用率,在一定程度上缓解交通拥堵问题。然而目前的可变车道多是定时或人工切换,对交通流量的引导不够及时准确。利用车辆检测自动识别各个方向车道的车流密度,可以为可变车道的智能引导提供决策依据。通用的目标检测算法对特定场景缺乏针对性,存在优化空间。对智能交通引导应用场景中的车辆检测任务进行了分析,通过充分利用结构化场景的先验信息,对两阶段目标检测框架中的候选区域生成算法进行重新设计,提出了基于车道线检测直接生成候选区域的算法,提高了车辆检测的效率和准确率。针对该场景下常出现的车辆遮挡问题,采用一种高斯加权的非极大值抑制算法有效降低车辆的漏检率。在实际的交通引导场景数据集上验证了算法的有效性。

Traffic congestion can be alleviated to some extent by setting reversible lanes,which can improve the utilization rate of existing road resources.However,the current reversible lanes are mostly scheduled or manual switching,which is not timely and appropriate enough to guide the traffic flow.Using vehicle detection to automatically identify the density of traffic flow in each direction can provide decision basis for intelligent guidance of variable lanes.The general target detection algorithms are not specific to some scenarios,which can be further optimized.After analyzing the vehicle detection task in the application scenario of intelligent traffic guidance and making full use of the priori information of structured scene,the candidate region generation algorithm in two-stage target detection framework is redesigned,and an algorithm based on lane line detection is proposed to directly generate candidate region,which is more efficient and accurate in vehicle detection.Aiming at the problem of vehicle occlusion in this scenario,a Gaussian weighted non-maximum suppression algorithm is used to reduce the vehicle missed detection rate effectively.The effectiveness of the algorithm is verified on a real traffic guidance scenario data set.