高德报告数据显示，全国有1/3的城市高峰通勤受到拥堵的威胁，通过对城市交叉口交通拥堵问题分析，发现造成现有交叉口拥堵的原因有：车道功能不合理、绿信比分配不合理、相位相序设置不合理、路口内部渠化设计缺失、交通供需分布不均、进出口车道数量不匹配等。随着人脸识别、指纹识别技术的迅速发展，我们思考，类比人脸识别将交叉口交通数据来进行特征识别，通过识别进行匹配，发现已知交叉口管理或规划上的缺陷。从而优化或解决现有交叉口拥堵问题。本设计分为知识图谱模块，数据采集模块，控制模块三大模块。通过知识抽取和知识融合，建立案例数据库，经过深度学习和大量训练，建立“AI+深度学习”式的交叉口优化知识图谱。通过数据采集模块中的多目标雷达监测车流量等信息，智能交通摄像机、照相机等前端设备对交叉口道路特征元素的捕捉，经图像信息预处理后，运用图像识别技术对拥堵交叉口进行识别，对已知交叉口和标准案例库中交叉口标准化模型进行匹配度判断并根据匹配度不同匹配不同的优化策略。
According to the data of Gaud report, 1/3 of urban peak commuting in China is threatened by con-gestion. Based on the analysis of traffic congestion at urban intersections, it is found that the causes of existing intersection congestion are: unreasonable lane function, unreasonable distribution of green signal ratio, unreasonable phase sequence setting, lack of internal channelization design, uneven distribution of traffic supply and demand, and mismatched number of import and export lanes Match etc. With the rapid development of face recognition and fingerprint recognition technology, we think that analogical face recognition will recognize the intersection face for feature recognition, match through recognition, and find the known defects in intersection management or planning and optimize or solve the existing intersection congestion problems. The design is divided into three modules: knowledge map module, data acquisition module and control module. Through knowledge extraction and knowledge fusion, the case database is established. After deep learning and a lot of training, the knowledge map of intersection optimization of “AI + deep learning” is established. Through the multi-target radar in the data acquisition module to monitor the traffic flow and other information, the intelligent traffic camera, camera and other front-end equipment to capture the road characteristic elements of the intersection, after image information preprocessing, use image recognition technology to identify the congested intersection, judge the matching degree of the known intersection and the standardized model of the intersection in the standard case base, and then match them according to the matching degree Different degrees match different optimization strategies.
