面向自动驾驶的驾驶环境视觉感知复杂度量化评估方法

doi:10.3976/j.issn.1002-4026.2022.03.008

摘要/Abstract

摘要：

为优选自动驾驶汽车开放测试、示范道路并支撑其驾驶环境的优化,提出面向自动驾驶的驾驶环境视觉感知复杂度量化评估方法。以百度街景地图作为驾驶环境数据源,运用脚本文件以及截图工具PicPick搭建自动化街景图像数据提取平台,并在不同区域、不同道路等级下采集上海市50条典型道路的驾驶环境数据;从行人、交通标志、交通标线、红绿灯、车辆5方面出发,构建驾驶环境要素感知平台,并开展感知精度的量化评估;在单要素感知准确率的基础上,采用熵权法确定多维感知要素权重,计算各道路综合感知准确率,并应用轮廓系数法与K-means++聚类算法进行视觉感知复杂度分级。结果表明,上海市典型道路的驾驶环境视觉感知复杂度分为三级,大部分道路的视觉感知复杂度属于2级;对比不同等级道路发现,支路的视觉感知复杂度总体上低于主干路。

关键词: 视觉感知复杂度, 复杂度分级, 感知关键要素, 驾驶环境, 自动驾驶汽车

Abstract:

To support the optimal selection of public testing road sites and provide instructions for improving driving environments for autonomous vehicles, a quantitative evaluation method for driving environment visual perception complexity was proposed. Based on the driving environment dataset from Baidu street view map, an automated extraction platform of street view image data was established using a script file and the screenshot tool PicPick. Driving environment data of 50 typical Shanghai roads of different areas and different road grades were collected. Then, an element perception platform of driving environment was established from five aspects—pedestrians, traffic signs, road markings, traffic lights, and vehicles—to quantitatively evaluate perception accuracy. Based on single-element perception accuracy, the weights for multidimensional perception elements were determined using the entropy method, and integrated perception accuracy at the road level was calculated. The silhouette coefficient method and K-means++ clustering algorithm were used to propose the classification of visual perception complexity. Results showed that the visual perception complexity of the road network in Shanghai can be divided into three levels. The majority of driving environment visual perception complexities of roads belong to level 2. In addition, comparisons among roads of different grades indicated that the visual perception complexity of local roads is generally lower than that of arterial roads.

Key words: :visual perception complexity, complexity classification, perception key elements, driving environment, autonomous vehicles

中图分类号:

U471

余荣杰,赵岁阳,董浩然. 面向自动驾驶的驾驶环境视觉感知复杂度量化评估方法[J]. 山东科学, 2022, 35(3): 62-71.

YU Rong-jie,ZHAO Sui-yang,DONG Hao-ran. Method for the quantitative evaluation of the visual perception complexity of driving environments for autonomous vehicles[J]. Shandong Science, 2022, 35(3): 62-71.

图/表 12

图1

表1

图2

图3

图4

表2

表3

图5

图6

表4

图7

表5

参考文献 27

[1]	California Department of Motor Vehicles. Autonomous vehicle disengagement reports 2018[R]. California: California Department of Motor Vehicles, 2018.
[2]	张迪. 基于单目视觉的自动驾驶视觉感知算法设计[D]. 武汉: 华中科技大学, 2019.
[3]	BRUMMELEN J V, O'BRIEN M, GRUYER D, et al. Autonomous vehicle perception: the technology of today and tomorrow[J]. Transportation research part C: emerging technologies, 2018, 89: 384-406. doi: 10.1016/j.trc.2018.02.012
[4]	NITSCHE P, MOCANU I, REINTHALER M. Requirements on tomorrow's road infrastructure for highly automated driving[C]// 2014 International Conference on Connected Vehicles and Expo (ICCVE). Vienna: IEEE, 2014: 939-940.
[5]	北京市交通委员会, 北京市公安局公安交通管理局, 北京市经济和信息化委员会. 北京市自动驾驶车辆测试路段道路要求(试行)[EB/OL]. [2021-07-03]. http://jtw.beijing.gov.cn/xxgk/tzgg/201803/P020191218698265121072.pdf.
[6]	深圳市交通运输委员会, 深圳市公安局交通警察局, 深圳市发展和改革委员会, 等. 深圳市智能网联汽车道路测试开放道路技术要求(试行)[EB/OL]. [2021-07-03]. http://jtys.sz.gov.cn/zwgk/zdxzjc/2018zdxzjc/content/post_4282901.html.
[7]	ROXAS E A, ACILO J N, VICERRA R R P, et al. Vision-based traffic sign compliance evaluation using convolutional neural network[C]// Proceedings of IEEE International Conference on Applied System Innovation 2018 (ICASI). Chiba: IEEE, 2018: 120-123.
[8]	LU X, MADADI B, FARAH H, et al. Scenario-based infrastructure requirements for automated driving[C]// ZHANG L, MA J, LIU P. 19th COTA International Conference of Transportation Professionals. Nanjing: ASCE, 2019: 5684-5695.
[9]	ZHU H, YUEN K V, MIHAYLOVA L, et al. Overview of environment perception for intelligent vehicles[J]. IEEE Transactions on Intelligent Transportation Systems, 2017, 18(10): 2584-2601. doi: 10.1109/TITS.2017.2658662
[10]	ZOU B, LI W, WANG D. Analysis on current situation of China's intelligent connected vehicle road test regulations[C]// MONTEMANNI R. MATEC Web of Conferences. Bangkok: EDP Sciences, 2019: 1-6.
[11]	BILAL M, HANIF M S. Benchmark revision for HOG-SVM pedestrian detector through reinvigorated training and evaluation methodologies[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 21(3): 1277-1287. doi: 10.1109/TITS.2019.2906132
[12]	FAN Q, BROWN L, SMITH J. A closer look at faster R-CNN for vehicle detection[C]// 2016 IEEE intelligent vehicles symposium (IV). Gothenburg: IEEE, 2016:124-129.
[13]	MOIZUMI H, SUGAYA Y, OMACHI M, et al. Traffic light detection considering color saturation using in-vehicle stereo camera[J]. Journal of Information Processing, 2016, 24(2): 349-357. doi: 10.2197/ipsjjip.24.349
[14]	郭笙听, 李子印, 赵储. 结合 Hough 变换与 Kalman 滤波的车道线检测方法[J]. 中国计量大学学报, 2017, 28(4): 460-466.
[15]	RAJENDRAN S P, SHINE L, PRADEEP R, et al. Real-Time traffic sign recognition using YOLO v3 based detector[C]// 2019 10th International Conference on Computing, Communication and Networking Technologies (ICCCNT). Kanpur: IEEE, 2017:1-7.
[16]	SUN D, WATADA J. Detecting pedestrians and vehicles in traffic scene based on boosted HOG features and SVM[C]// 2015 IEEE 9th International Symposium on Intelligent Signal Processing (WISP) Proceedings. Siena: IEEE, 2015: 1-4.
[17]	REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 39(6):1137-1149. doi: 10.1109/TPAMI.2016.2577031
[18]	GONG J, JIANG Y, XIONG G, et al. The recognition and tracking of traffic lights based on color segmentation and CAMSHIFT for intelligent vehicles[C]// IEEE. Intelligent Vehicles Symposium. San Diego: IEEE, 2010: 431-435.
[19]	VLADIMIR T, JEON D, KIM D H. Hough transform with kalman filter on GPU for real-time line tracking[C]// 2013 Seventh International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing. Taichung: IEEE, 2013: 212-216.
[20]	NGUYEN T T, SPEHR J, XIONG J, et al. A survey of performance measures to evaluate ego-lane estimation and a novel sensor-independent measure along with its applications[C]// 2017 IEEE International Conference on Multi-sensor Fusion and Integration for Intelligent Systems (MFI). Daegu: IEEE, 2017: 239-246.
[21]	中华人民共和国交通运输部. 公路项目安全性评价规范: JTG B05—2015[S]. 北京: 人民交通出版社, 2016.
[22]	李贤钰, 郭忠印, 苏东兰, 等. 基于操纵信号的小客车专用高速平曲线驾驶特性[J]. 同济大学学报(自然科学版), 2020, 48(3): 368-376.
[23]	CHEN F, SONG M, MA X, et al. Assess the impacts of different autonomous trucks' lateral control modes on asphalt pavement performance[J]. Transportation Research Part C: Emerging Technologies, 2019, 103: 17-29. doi: 10.1016/j.trc.2019.04.001
[24]	冀杰, 李以农, 郑玲. 车辆自动驾驶系统纵向和横向运动综合控制[J]. 中国公路学报, 2010, 23(5):119-126.
[25]	张文会, 李德才, 罗文文. 基于熵权的交通事故现场安全属性识别模型[J]. 交通运输系统工程与信息, 2013, 13(2): 136-140.
[26]	余荣杰, 龙晓捷, 涂颖菲, 等. 基于低频轨迹数据的分时租赁驾驶人驾驶风格分析[J]. 同济大学学报(自然科学版), 2019, 47(10): 1463-1469.
[27]	魏广奇, 苏跃江, 吴德馨, 等. 基于高速公路流水数据的通勤车辆特征研究[J]. 交通运输系统工程与信息, 2019, 19(3): 237-244.

要素分类	感知关键要素	关键算法及原理
动态驾驶环境	行人	HOG+SVM的行人检测算法^[16]
	车辆	快速卷积神经网络^[17]
	红绿灯	RGB色彩空间分割^[18]
静态驾驶环境	交通标线	Hough变换与卡尔曼滤波^[19]
静态驾驶环境	交通标志	YOLO v3目标检测算法^[15]

指标	交通标线	红绿灯	交通标志	行人	车辆
信息熵值	0.990 4	0.985 9	0.992 0	0.989 8	0.990 4
信息效用值	0.009 6	0.014 1	0.008 0	0.010 2	0.009 6
权重	0.18	0.27	0.16	0.20	0.19

聚类区间	视觉感知复杂度等级L	样本数
[0,0.47)	3	15
[0.47,0.55)	2	22
[0.55,1]	1	13

基准道路等级	综合感知准确率					视觉感知复杂度
基准道路等级	平均值	标准差	对比道路等级	F值	P值	平均值	标准差	对比道路等级	F值	P值
主干路	0.471 4	0.074 8	次干路	4.073	0.053	2.466 7	0.743 2	次干路	2.291	0.141
			支路	9.033	0.005			支路	10.435	0.003
次干路	0.513 8	0.047 8	主干路	4.073	0.053	2.066 7	0.703 7	主干路	2.291	0.141
			支路	1.513	0.227			支路	2.513	0.122
支路	0.534 0	0.048 6	主干路	9.033	0.005	1.700 0	0.657 0	主干路	10.435	0.003
			次干路	1.513	0.227			次干路	2.513	0.122