基于雷视一体机的交通流数据采集系统设计

Design of Data Acquisition and Fusion System for Highway Traffic Flow Based on Rayleigh Integrated Machine

摘要:

交通流数据是进行交通管理宏观决策的基础数据，交通流数据采集系统是交通管理信息化智能化的重要组成部分。随着我国交通领域的蓬勃发展，交通流量激增、高速交通拥堵、交通事故等突发事件频发，为此，设计一套实时性好、准确度高的交通流数据采集系统是十分必要的。论文基于雷视一体机开发了一套高速公路交通流数据采集系统，采用端-边-云分级传输的物联网架构，并结合了自主研发的雷视一体机，采用CNN神经网络技术提取图像信息后，在边缘计算机中通过一维数据最优估计、多传感器数据匹配、多传感器双向最优估计、多传感器目标特征融合的软件工作流程，将雷达与监控相机提取到的信息进行最优化估计，准确提取道路目标交通信息，仅将处理后的特征信息上传至云端服务器，实现交通流数据的精确、实时采集。系统试运行结果表明，该基于雷视一体机的交通流数据采集系统能够有效提高检测准确性，加强检测结果的实时性。

Abstract:

Traffic flow data is the basic data for the macro decision of traffic management, and the traffic flow data collection system is an important part of the traffic management information intelligence. With the vigorous development of China's traffic field, traffic flow surge, high-speed traffic congestion, traffic accidents and other emergencies occur frequently. Therefore, it is very necessary to design a traffic flow data collection system with good real-time and high accuracy. This paper develops a set of highway traffic flow data acquisition system based on the all-in-one machine, using the Internet of T architecture of terminal-edge-cloud hierarchical transmission. Moreover, the system combines the independently developed radar video all-one machine and uses CNN neural network technology to extract image information. Then, the edge computer optimizes the information extracted from the radar and the monitoring camera through the data workflow of one-dimensional data optimal estimation, multi-sensor data matching, multi-sensor bidirectional optimal estimation, and multi-sensor target feature fusion, to accurately extract the road target traffic information. Finally, the system uploads the processed feature information to the cloud server to realize the accurate and real-time collection of traffic flow data. The system test operation results show that the traffic flow data acquisition system based on radar video all-one machine can effectively improve the detection accuracy and strengthen the real-time performance of the detection results.