作者：高继东1, 焦鑫2, 刘全周1, 贾鹏飞1, 李占旗1, 杨伟东2
作者单位：1. 中汽研（天津）汽车工程研究院有限公司，天津 300300;
2. 河北工业大学机械学院，天津 300130
关键词：机器视觉;毫米波雷达;深度学习;信息融合;卡尔曼滤波
摘要：
为提升辅助驾驶系统（ADAS）对于行驶环境中车辆的感知能力，运用机器视觉与毫米波雷达信息融合技术对车辆进行检测。融合系统中将摄像头和雷达进行联合标定，确定两者的转化关系，对深度学习算法SSD进行改进，提升对于小目标车辆的检测精度，同时对雷达数据进行处理，借助雷达模拟器确定合适阈值参数实现对车辆目标的有效提取，并采用卡尔曼滤波算法对目标数据进行处理，根据雷达有效目标数据对摄像头采集的图像进行选择与建立感兴趣区域，通过改进的SSD车辆识别算法对区域中的车辆进行检测，经测试，车辆的检测准确率最高达到95.3%，且具备较高的实时性和环境适应性。
Research on vehicle detection based on data fusion of machine vision and millimeter wave radar
GAO Jidong1, JIAO Xin2, LIU Quanzhou1, JIA Pengfei1, LI Zhanqi1, YANG Weidong2
1. CATARC (Tianjin) Automotive Engineering Research Institute Co., Ltd., Tianjin 300300, China;
2. School of Mechanical Engineering, Hebei University of Technology, Tianjin 300130, China
Abstract: In order to improve the vehicle detection capabilities of Advanced Assisted Driving System (ADAS), the data fusion of machine vision and millimeter wave radar is used to detect vehicles. In the fusion system, the camera and radar were jointly calibrated to determine their conversion formula. The deep learning algorithm SSD was improved to improve the detection accuracy of small target vehicles. The appropriate threshold parameters of radar data were determined by radar simulator and the effective vehicle target was extracted. Then Kalman filter algorithm was used to process the target data. According to these effective target data, the image collected by the camera was selected. Vehicles in the selection area were detected with the improved SSD algorithm. In the test, the vehicle detection rate is 95.3%, and it has high real-time performance and environmental adaptability.
Keywords: machine vision;millimeter wave radar;deep learning;data fusion;Kalman filter
2021, 47(10):33-40 收稿日期: 2020-12-16;收到修改稿日期: 2021-01-15
基金项目: 国家自然科学基金面上项目(51975428)；国家自然科学基金项目(51975426)；天津市科技计划项目（17YDLJGX00020，18YFCZZC00150）
作者简介:
参考文献
[1] 罗栩豪, 王培, 李绍华, 等. 汽车辅助驾驶系统动态目标检测方法[J]. 计算机工程, 2018, 44(1): 311-316
[2] WANG T, XIN J, ZHENG N. A method integrating human visual attention and consciousness of radar and vision for autonomous vehicle navigation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015, 39(6): 1137-1149
[3] FLORIAN F, HERMANN R. An automotive radar net based on 77 GHz FMCW sensors[J]. International Radar Conference, 2005, 18(6): 871-876
[4] NATNAEL S, YOUGMIN W, JINTAE K. Millimeter-wave radar and RGB-D camera sensor fusion for real-time people dection and tracking[J]. 2019 7th International Conference on Robot Intelligence Technology and Applications(RiTA), Daejeon, Korea(South), 2019, 20(12): 93-98
[5] 李鹏, 彭嘉潮, 黄培炜, 等. 基于双目标传感器分布优化的转向架构状态监测[J]. 中国测试, 2020, 46(9): 131-147
[6] 陈晓伟. 汽车前方车辆识别的雷达和视觉信息融合算法开发[D]. 长春: 吉林大学, 2016.
[7] EVERINGHAM M, ESLAMI S, VAN G, et al. The pascal visual object classes challenge: A retrospective[J]. International Journal of Computer Vision, 2015, 111(1): 98-136
[8] 王昕, 赵飞, 蒋佐富, 等. 迁移学习和卷积神经网络电力设备图像识别方法[J]. 中国测试, 2020, 46(5): 108-113
[9] GIRSHICK R. Fast r-cnn[C]// 2015 IEEE International Conference on Computer Vision(ICCV), Santiago, 2015: 1440-1448.
[10] 王宝峰, 齐志权, 马国成, 等. 一种基于雷达和机器视觉信息融合的车辆识别方法[J]. 汽车工程, 2015, 37(6): 674-678
[11] 张家旭, 李静. 基于交互式多模型和容积卡尔曼滤波的汽车状态估计[J]. 汽车工程, 2017, 9(39): 977-983
[12] 李辉, 宋耀良, 杨余旺. 宽带混沌信号在汽车防撞雷达中的应用[J]. 现代雷达, 2006(11): 54-57
[13] 公路工程技术标准：JTG B01—2003[S]. 北京: 中华人民共和国交通部, 2003:12-13.
[14] 罗逍, 姚远, 张金换. 一种毫米波雷达和摄像头联合标定方法[J]. 清华大学学报, 2014, 3(54): 289-293
[15] 乐亚南. 基于曲线拟合理论的点云数据处理分析[D]. 西安: 西安交通大学, 2015.
[16] WU X, REN J, WU Y, et al. Study on target based on vision and radar sensor fusion[J]. Wcx World Congress Experience, 2018, 11(5): 1-8
