基于图像处理的交通事故检测及责任判定

为了在发生轻微交通事故时, 快速使事故车辆驶离现场, 保证道路畅通, 提出了一种车辆碰撞检测及责任判定模型. 首先结合SSD目标检测算法(single shot multibox detector)和MobileNet轻量级深度网络模型, 对其进行改进以获取每一帧视频图像中运动目标的位置和大小信息, 实现对车辆识别与检测. 其次, 利用卡尔曼滤波器对连续图像帧之间的运动目标建立对应匹配关系, 预测目标的运动状态, 对目标的位置及运动趋势做出判断, 实现车辆轨迹跟踪. 随后通过车辆目标检测框的交并比判断是否发生碰撞. 最后针对直行道路中车辆的速度、方向信息结合道路安全条例及机动车事故快速方法对事故车辆进行责任判定. 结果分析表明, 该研究可实现直行道路场景下的追尾及变道引发的车辆碰撞检测及责任判定.     

基于PSO-IGA算法的高性能十字阵列的设计与综合

前碰撞预警系统是安全辅助驾驶领域的一项重要部分,通过计算机处理交通环境信息,当检测到潜在危险时,及时提醒并辅助驾驶员。采用计算机视觉方法,通过目标检测和跟踪算法,获取图像中目标车辆的位置和轨迹信息,并利用相机标定,计算当前车辆和前方车辆在世界坐标系中的距离、速度及轨迹等信息,综合该信息,实现前车碰撞时间预警、前车并线预警以及非机动车预警算法。在前车并线过程中,利用轨迹信息实时检测前车并线意图,及时提示驾驶员注意避让前方车辆。实验表明,本文提出的预警算法具有较高的准确性和鲁棒性,特别在高架或高速道路场景下,并线预警算法能检测到前车的并线意图,及时预警。     

结合状态机和动态目标路径的无人驾驶决策仿真

目的 决策系统是无人驾驶技术的核心研究之一。已有决策系统存在逻辑不合理、计算效率低、应用场景局限等问题,因此提出一种动态环境下无人驾驶路径决策仿真。方法 首先,基于规则模型构建适于无人驾驶决策系统的交通有限状态机;其次,针对交通动态特征,提出基于统计模型的动态目标路径算法计算状态迁移风险;最后,将交通状态机和动态目标路径算法有机结合,设计出一种基于有限状态机的无人驾驶动态目标路径模型,适用于交叉口冲突避免和三车道换道行为。将全速度差连续跟驰模型运用到换道规则中,并基于冲突时间提出动态临界跟车距离。结果 为验证模型的有效性和高效性,对交通环境进行虚拟现实建模,模拟交叉口通行和三车道换道行为,分析文中模型对车流量和换道率的影响。实验结果显示,在交叉口通行时,自主车辆不仅可以检测冲突还可以根据风险评估结果执行安全合理的决策。三车道换道时,自主车辆既可以实现紧急让道,也可以通过执行换道达成自身驾驶期望。通过将实测数据和其他两种方法对比,当车流密度在0.20.5时,本文模型的平均速度最高分别提高32 km/h和22 km/h。当车流密度不超过0.65时,本文模型的换道成功率最高分别提升37%和25%。结论 实验结果说明本文方法不仅可以在动态城区环境下提高决策安全性和正确性,还可以提高车流量饱和度,缓解交通堵塞。     

基于物联网的城市道口机动车智能监控和行为检测系统

物联网架构的城市道口机动车监控和行为分析系统在传统的物联网定义基础上,提出了包含感知／发布层、智能中间层、网络层、服务层和应用层的智能监控架构、以高清网络监控相机为采集终端,使用自行开发的图像分析算法和软件实现了道路拥堵状态判定以及多目标车辆追踪,实时采集视频流中道口监控区域内车辆移动轨迹、速度、前方道路拥堵状态等基本参数;在通信层使用3G／Wi—Fi等高速无线通信技术与有线通信相结合实时传输关键参数和监控图像,应用层使用自行开发的中心监控软件平台实现了对关键数据的存储、统计、业务应用的服务支持和管理,应用层针对电子警察、智能交通监控和信息发布展开,结合交通控制信息和我国交通法规为违章实时判定、交通信息采集等提供了基础、     

RFID，GPS和GIS技术集成在交通智能监管系统中的应用研究*

为实现在城市复杂路网情况下对交通车辆的实时监控,并且能通过一定数量的车辆运行状态来判断道路交通的拥挤状况,采用射频识别技术(RFID)对道路上运行的车辆进行动态识别和数据信息交换;依靠全球定位系统(GPS)技术实时获得目标车辆的位置信息,并通过地理信息系统(GIS)将车辆的运行状况以及路网的交通状况以电子地图形式实时地展现给用户。将GPS、GIS与RFID技术综合应用于城市道路交通管理系统中,在此基础上设计出道路交通车辆的全程监控模型和系统框架。对交通监管的信息化建设具有一定的借鉴意义。     

车联网环境下高速公路合流区换道协调决策与控制系统设计

为了降低高速公路合流区车辆换道碰撞风险、提高道路通行质量,在车联网环境下,优化设计高速公路合流区换道协调决策与控制系统;通过安装车联网终端芯片、改装协调控制器,完成硬件系统的优化;根据高速公路合流区的物理结构,设置换道区域约束条件;在车联网环境下采集高速公路合流区车路信息,根据换道紧迫程度、跟随车辆换道行为、车间距离安全性的计算结果,作出换道协调决策;规划合流区最优协调换道轨迹作为系统控制目标,最终计算位置、速度控制量,在控制器支持下,实现高速公路合流区换道协调控制功能;通过系统测试得出结论：所设计系统的车辆换道碰撞概率和道路拥塞程度分别为0.15%和0.20%,道路车流量得到明显提升,车流量的平均值为356 pcu/d;车辆速度控制误差为0.05 km/h,实验结果表明所设计系统在决策与控制性能方面具有明显优势。     

基于道路风险评估的城市路网实时路径选择

为了缓解城市交通拥堵、避免交通事故的发生,城市路网的路径选择一直以来是一个热门的研究课题.随着边缘计算和车辆智能终端技术的发展,城市路网中的行驶车辆从自组织网络朝着车联网(Internet of vehicles,IoV)范式过渡,这使得车辆路径选择问题从基于静态历史交通数据的计算向实时交通信息计算转变.在城市路网路径选择问题上,众多学者的研究主要聚焦如何提高出行效率,减少出行时间等.然而这些研究并没有考虑所选路径是否存在风险等问题.基于以上问题,首次构造了一个基于边缘计算技术的道路风险实时评估模型(real-time road risk assessment model based on edge computing, R³A-EC),并提出基于该模型的城市路网实时路径选择方法(real-time route selection method based on risk assessment, R²S-RA). R3A-EC模型利用边缘计算技术的低延迟,高可靠性等特点对城市道路进行实时风险评估,并利用最小风险贝叶斯决策验证道路是否存在风险问...     

Joint ego-motion and road geometry estimation

We provide a sensor fusion framework for solving the problem of joint ego-motion and road geometry estimation. More specifically we employ a sensor fusion framework to make systematic use of the measurements from a forward looking radar and camera, steering wheel angle sensor, wheel speed sensors and inertial sensors to compute good estimates of the road geometry and the motion of the ego vehicle on this road. In order to solve this problem we derive dynamical models for the ego vehicle, the road and the leading vehicles. The main difference to existing approaches is that we make use of a new dynamic model for the road. An extended Kalman filter is used to fuse data and to filter measurements from the camera in order to improve the road geometry estimate. The proposed solution has been tested and compared to existing algorithms for this problem, using measurements from authentic traffic environments on public roads in Sweden. The results clearly indicate that the proposed method provides better estimates.     

Heuristic based evolution for the coordination of autonomous vehicles in the absence of speed lanes

The current state of the art in the planning and coordination of autonomous vehicles is based upon the presence of speed lanes. In a traffic scenario where there is a large diversity between vehicles the removal of speed lanes can generate a significantly higher traffic bandwidth. Vehicle navigation in such unorganized traffic is considered. An evolutionary based trajectory planning technique has the advantages of making driving efficient and safe, however it also has to surpass the hurdle of computational cost. In this paper, we propose a real time genetic algorithm with Bezier curves for trajectory planning. The main contribution is the integration of vehicle following and overtaking behaviour for general traffic as heuristics for the coordination between vehicles. The resultant coordination strategy is fast and near-optimal. As the vehicles move, uncertainties may arise which are constantly adapted to, and may even lead to either the cancellation of an overtaking procedure or the initiation of one. Higher level planning is performed by Dijkstra's algorithm which indicates the route to be followed by the vehicle in a road network. Re-planning is carried out when a road blockage or obstacle is detected. Experimental results confirm the success of the algorithm subject to optimal high and low-level planning, re-planning and overtaking.     

Multi-Level Planning for Semi-autonomous Vehicles in Traffic Scenarios Based on Separation Maximization

The planning of semi-autonomous vehicles in traffic scenarios is a relatively new problem that contributes towards the goal of making road travel by vehicles free of human drivers. An algorithm needs to ensure optimal real time planning of multiple vehicles (moving in either direction along a road), in the presence of a complex obstacle network. Unlike other approaches, here we assume that speed lanes are not present and that different lanes do not need to be maintained for inbound and outbound traffic. Our basic hypothesis is to carry forward the planning task to ensure that a sufficient distance is maintained by each vehicle from all other vehicles, obstacles and road boundaries. We present here a 4-layer planning algorithm that consists of road selection (for selecting the individual roads of traversal to reach the goal), pathway selection (a strategy to avoid and/or overtake obstacles, road diversions and other blockages), pathway distribution (to select the position of a vehicle at every instance of time in a pathway), and trajectory generation (for generating a curve, smooth enough, to allow for the maximum possible speed). Cooperation between vehicles is handled separately at the different levels, the aim being to maximize the separation between vehicles. Simulated results exhibit behaviours of smooth, efficient and safe driving of vehicles in multiple scenarios; along with typical vehicle behaviours including following and overtaking.     

Motion planning of autonomous vehicles in a non-autonomous vehicle environment without speed lanes

Planning is one of the key problems for autonomous vehicles operating in road scenarios. Present planning algorithms operate with the assumption that traffic is organised in predefined speed lanes, which makes it impossible to allow autonomous vehicles in countries with unorganised traffic. Unorganised traffic is though capable of higher traffic bandwidths when constituting vehicles vary in their speed capabilities and sizes. Diverse vehicles in an unorganised exhibit unique driving behaviours which are analysed in this paper by a simulation study. The aim of the work reported here is to create a planning algorithm for mixed traffic consisting of both autonomous and non-autonomous vehicles without any inter-vehicle communication. The awareness (e.g. vision) of every vehicle is restricted to nearby vehicles only and a straight infinite road is assumed for decision making regarding navigation in the presence of multiple vehicles. Exhibited behaviours include obstacle avoidance, overtaking, giving way for vehicles to overtake from behind, vehicle following, adjusting the lateral lane position and so on. A conflict of plans is a major issue which will almost certainly arise in the absence of inter-vehicle communication. Hence each vehicle needs to continuously track other vehicles and rectify plans whenever a collision seems likely. Further it is observed here that driver aggression plays a vital role in overall traffic dynamics, hence this has also been factored in accordingly. This work is hence a step forward towards achieving autonomous vehicles in unorganised traffic, while similar effort would be required for planning problems such as intersections, mergers, diversions and other modules like localisation.     

自动驾驶在拥堵路段的道路几何信息估计

道路几何信息是自动驾驶系统中重要的信息来源,也是后续路径规划的关键参考信息之一。 该研究针对城市内车道线遮挡及多路径效应导致的全球定位系统失效等问题,提出了一种基于前车信 息的道路几何估计方法。通过对当前车辆、前车以及道路之间关系的建模,获得了系统的运动模型和 观测模型。采用无损卡尔曼滤波框架对观测到的前车相对位置、相对速度、相对角度和本车角速度进 行滤波处理,估计出当前车道的曲率参数。在仿真软件 Car learning to Act(Carla)上的实验结果表明, 相比地图匹配方法,在无法获取车道线目标及精确定位信息的情况下,该方法道路几何精度得到了显 著提升。     

3D Pose Estimation by Directly Matching Polyhedral Models to Gray Value Gradients

This contribution addresses the problem of pose estimation and tracking of vehicles in image sequences from traffic scenes recorded by a stationary camera. In a new algorithm, the vehicle pose is estimated by directly matching polyhedral vehicle models to image gradients without an edge segment extraction process. The new approach is significantly more robust than approaches that rely on feature extraction since the new approach exploits more information from the image data. We successfully tracked vehicles that were partially occluded by textured objects, e.g., foliage, where a previous approach based on edge segment extraction failed. Moreover, the new pose estimation approach is also used to determine the orientation and position of the road relative to the camera by matching an intersection model directly to image gradients. Results from various experiments with real world traffic scenes are presented.     

基于稀疏帧检测的交通目标跟踪

为了获取高速公路交通视频中目标车辆的行驶轨迹,提出一种基于视频的多目标车辆跟踪及实时轨迹分布算法,为交通管理系统和交通决策提供目标车辆交通信息.首先,使用YOLOv4算法检测目标车辆位置及置信度.其次,在不同场景条件下,使用提出的基于稀疏帧检测的跟踪方法,结合KCF跟踪算法,将车辆数据进行关联获取完整轨迹.最后,用车辆分布图和交通场景俯视图显示轨迹,便于交通管理与分析.实验结果表明,提出的跟踪方法在车辆跟踪中有较高的跟踪正确率,同时基于稀疏帧检测的跟踪方法处理速度也较快,实时轨迹分布正确反映了真实场景的车道信息以及目标车辆运动信息.     

基于视频的多目标车辆跟踪及轨迹优化

为了获取交通视频中车辆的运动轨迹,提供道路动态交通信息,提出一种基于Yolo3目标检测和KCF目标预测相结合,关联历史轨迹预测结果和检测结果的长时间多目标车辆跟踪算法;对采用机器视觉获取的车辆轨迹非平滑现象,提出通过Savitzky-Golay滤波器对原始的车辆轨迹进行平滑优化。对比测试场景中车辆轨迹优化前后,优化后的轨迹在保留原有车辆运动特征的前提下,改善了轨迹平滑性,提供的动态交通信息更能反映车辆真实运动状况。     

一种基于OBB包围盒算法的交通事故检测方法

给出了一种用虚拟环境下的OBB碰撞检测算法对真实环境下城市道路交通事故进行检测的方法。该方法的主要思想是利用OBB算法计算车辆的包围盒,通过对包围盒的地面平面投影得到车辆矩形二维包围盒,然后检测这些矩形在同一平面上是否相交从而判断车辆是否发生碰撞,最后根据车辆的碰撞方式确认事故的类型,从而完成事故检测。为了提高算法效率,还给出了OBB包围盒数量的控制方法。实验结果表明,该方法能够快速地检测到交通事故的发生。     

基于交互车辆轨迹预测的自动驾驶车辆轨迹规划

针对城市道路等复杂行车场景,提出了一种基于交互车辆轨迹预测的自动驾驶车辆轨迹规划方法,将高维度的轨迹规划解耦为低维度的路径规划和速度规划;首先,采用五次多项式曲线和碰撞剩余时间规划车辆行驶路径;其次,在社会生成对抗网络Social-GAN的基础上结合车辆空间影响和注意力机制对交互车辆进行轨迹预测;然后,结合主车规划路径、交互车辆预测轨迹及碰撞判定模型得到主车S-T图,采用动态规划和数值优化方法求解S-T图,从而得到满足车辆动力学约束的安全、舒适最优速度曲线;最后,搭建PreScan-CarSim-Matlab&Simulink-Python联合仿真模型进行实验验证。仿真结果表明,提出的轨迹规划方法能够在对交互车辆有效避撞的前提下,保证车辆行驶的舒适性和高效性。     

基于云网格集成调度的防拥堵车辆路径规划算法

在道路交通路网中,车辆拥堵问题是流量与路网结构之间相互作用的一个复杂动态过程,通过车辆路径规划,实现对路网网格集成调度,从而提高路网通行吞吐量。传统方法采用并行微观交通动态负载平衡预测算法实现车辆拥堵调度和车辆路径规划,不能准确判断路面上的车辆密度,路径规划效益不好。提出一种基于云网格集成调度的防拥堵车辆路径规划算法,即构建基于Small-World模型的云网格路网模型,采用RFID标签信息进行路况信息采集,实现交通网络拥堵评估信息特征的提取,采用固有模态函数加权平均求得各车道的车辆拥塞状态函数,对所有车道内车辆密度取统计平均可获得簇内的车辆密度。设计交通路网拥堵检测算法来对当前个体道路信息进行一维邻域搜索,从而实现车辆路径规划控制目标函数最佳寻优。通过动态博弈的方式求得车辆防拥堵路径的近似最优轨迹,实现路径规划算法的改进。仿真结果表明,该算法能准确规划车辆路径,实现最优路径控制,从而提高严重拥堵路段的车流速度和路网吞吐性能,性能优越。     

短时记忆与CenterTrack的车辆多目标跟踪

目的 车辆多目标跟踪是智能交通领域关键技术,其性能对车辆轨迹分析和异常行为鉴别有显著影响。然而,车辆多目标跟踪常受外部光照、道路环境因素影响,车辆远近尺度变化以及相互遮挡等干扰,导致远处车辆漏检或车辆身份切换（ID switch,IDs）问题。本文提出短时记忆与CenterTrack的车辆多目标跟踪,提升车辆多目标跟踪准确度（multiple object tracking accuracy,MOTA）,改善算法的适应性。方法 利用小样本扩增增加远处小目标车辆训练样本数;通过增加的样本重新训练CenterTrack确定车辆位置及车辆在相邻帧之间的中心位移量;当待关联轨迹与检测目标匹配失败时通过轨迹运动信息预测将来的位置;利用短时记忆将待关联轨迹按丢失时间长短分级与待匹配检测关联以减少跟踪车辆IDs。结果 在交通监控车辆多目标跟踪数据集UA-DETRAC （University at Albany detection and tracking）构建的5个测试序列数据中,本文方法在维持CenterTrack优势的同时,对其表现不佳的场景获得近30%的提升,与YOLOv4-DeepSort（you only look once—simple online and realtime tracking with deep association metric）相比,4种场景均获得近10%的提升,效果显著。Sherbrooke数据集的测试结果,本文方法同样获得了性能提升。结论 本文扩增了远处小目标车辆训练样本,缓解了远处小目标与近处大目标存在的样本不均衡,提高了算法对远处小目标车辆的检测能力,同时短时记忆维持关联失败的轨迹运动信息并分级匹配检测目标,降低了算法对跟踪车辆的IDs,综合提高了MOTA。     

智能车逆向超车控制算法

针对双向车道因受限于道路条件及交通特性仅能借用对向车道完成超车（逆向超车）的问题,通过采用车联网以及车载传感器获取环境车辆的速度、加速度等全局信息,将多车场景中各个实体所造成的影响纳入超车决策中,从而提出一种基于图搜索和模型预测控制（Model Predictive Control, MPC）的逆向超车控制方法。首先,根据车车通信获取的全局信息,结合非合作博弈,对各车在整个时段内的行为进行预测,并根据预测情况对道路的各个区域进行安全评估,评估依据为该区域在下一时刻出现车辆的概率。对道路完成评估后,得到碰撞概率热区图,之后采用A*算法搜索安全路径,根据安全路径完成目标车辆的轨迹规划,并设计模型预测控制器来对主车进行实时控制,使车辆按照既定轨迹行驶。最后,借助Carsim与MATLAB/Simulink搭建联合仿真平台,对提出的算法进行验证。仿真实验结果表明,该模型的控制误差最大不超过0.15 m,平均误差率约为1.7%,能实现对车辆的精准控制,保证被控车辆安全完成逆向超车。