摘要In extreme bad weather conditions, vehicles can timely adjust the speed and distance according to the road information, which will effectively reduce the occurrence of traffic accidents. Road type and road adhesion coefficient are the main factors affecting road condition. The traditional identification method of road adhesion coefficient has the characteristics of high cost and low reliability. At the same time, the road surface information recognition technology based on machine vision has become a hot spot in the current research, but the low accuracy and poor robustness are always the difficulties in the research. Based on the semantic segmentation model which is popular in machine vision, a new pavement type recognition technology is proposed by improving the model output network. According to the existing literature, vehicle road surface can be divided into the following 9 categories, including wet asphalt road, dry asphalt road, wet concrete road, dry concrete road, wet soil road, dry soil road, gravel road, compacting snow road and icy road. The standard data set is made by collecting road pictures through taking photos, downloading and other ways, and the pre-processed data set is used to train the improved semantic segmentation network model. Through many tests, training parameters that achieve the desired effect are selected for model parameter solidification, and the semantic segmentation model after solidification is used to predict the road image acquired by the camera. According to the prediction results of the model, the categories of the current driving road surface are obtained. The test results of a large number of road images show that the average classification accuracy of the nine road types is about 94%, which effectively improves the identification accuracy and robustness of the current road types. At the same time, the predicted time of a single frame image on a specific test platform is about 0.028 6 s, which meets the real-time requirements.
Abstract：In extreme bad weather conditions, vehicles can timely adjust the speed and distance according to the road information, which will effectively reduce the occurrence of traffic accidents. Road type and road adhesion coefficient are the main factors affecting road condition. The traditional identification method of road adhesion coefficient has the characteristics of high cost and low reliability. At the same time, the road surface information recognition technology based on machine vision has become a hot spot in the current research, but the low accuracy and poor robustness are always the difficulties in the research. Based on the semantic segmentation model which is popular in machine vision, a new pavement type recognition technology is proposed by improving the model output network. According to the existing literature, vehicle road surface can be divided into the following 9 categories, including wet asphalt road, dry asphalt road, wet concrete road, dry concrete road, wet soil road, dry soil road, gravel road, compacting snow road and icy road. The standard data set is made by collecting road pictures through taking photos, downloading and other ways, and the pre-processed data set is used to train the improved semantic segmentation network model. Through many tests, training parameters that achieve the desired effect are selected for model parameter solidification, and the semantic segmentation model after solidification is used to predict the road image acquired by the camera. According to the prediction results of the model, the categories of the current driving road surface are obtained. The test results of a large number of road images show that the average classification accuracy of the nine road types is about 94%, which effectively improves the identification accuracy and robustness of the current road types. At the same time, the predicted time of a single frame image on a specific test platform is about 0.028 6 s, which meets the real-time requirements.
通讯作者:
WANG Zhi-hong
E-mail: wangzhihong@whut.edu.cn
引用本文:
王志红, 王少博, 颜利蓉, 袁雨. Road Surface State Recognition Based on Semantic Segmentation[J]. Journal of Highway and Transportation Research and Development, 2021, 15(2): 88-94.
WANG Zhi-hong, WANG Shao-bo, YAN Li-rong, YUAN Yu. Road Surface State Recognition Based on Semantic Segmentation. Journal of Highway and Transportation Research and Development, 2021, 15(2): 88-94.
[1] WANG B, GUAN H, LU P, et al. Road Surface Condition Identification Approach Based on Road Characteristic Value[J]. Journal of Terramechanics, 2014, 56:103-117.
[2] QI Yan. Study on Road Condition Recognition Based on Vehicle Anti-collision Warning System[D].Wuhan:Huazhong University of Technology,2013. (in Chinese)
[3] YU Zhi-sheng, XIA Qun-sheng. Automobile Theory[M]. Beijing:China Machinery Press, 2009:93-94. (in Chinese)
[4] EICHHORN U, ROTH J. Prediction and Monitoring of Tyre/Road Friction[C]//XXIV FISITA Congress. London:FISITA, 1992.
[5] ERDOGAN G, ALEXANDER L, RAJAMANI R. Estimation of Tire-road Friction Coefficient Using a Novel Wireless Piezoelectric TIRE Sensor[J]. IEEE Sensors Journal, 2011, 11(2):267-279.
[6] YUAN Chao-chun, ZHANG Long-fei, CHEN Long, et al. A Research on the Algorithm for Identifying the Peak Adhesion Coefficient of Road Surface[J]. Automotive Engineering, 2017,39(11):1268-1273.(in Chinese)
[7] ALONSO J, LOPEZ J M, PAVON I, et al. On-board Wet Road Surface Identification Using Tyre/Road Noise and Support Vector Machines[J]. Applied Acoustics, 2014, 76:407-415.
[8] WANG Bo, LU Ping-ping, GUAN Xin, et al. A Review on the Development Status of Road Adhesion Coefficient Identification Approach[J]. Automotive Technology, 2014(8):1-7, 37.(in Chinese)
[9] SONG T, ZHOU H L, LIU H F. Road Friction Coefficient Estimation Based on BP Neural Network[C]//Proceedings of the 36th Chinese Control Conference. Dalian:IEEE, 2017:9491-9496.
[10] ZHAO J D, WU H Q, CHEN L L. Road Surface State Recognition Based on SVM Optimization and Image Segmentation Processing[J]. Journal of Advanced Transportation, 2017, 2017:1-21.
[11] WANG Shi-feng, DU Kai-yue, MENG Ying, et al. Machine Learning-based Road Terrain Recognition for Land Vehicles[J]. Acta Armamentarii, 2017,38(8):1642-1648.(in Chinese)
[12] ROYCHOWDHURY S, ZHAO M, WALLIN A, et al. Machine Learning Models for Road Surface and Friction Estimation Using Front-camera Images[C]//2018 International Joint Conference on Neural Networks (IJCNN). Rio de Janeiro:IEEE, 2018:1-8.
[13] WULFF F, SCHAUFELE B, SAWADE O, et al. Early Fusion of Camera and Lidar for Robust Road Detection Based on U-net FCN[C]//2018 IEEE Intelligent Vehicles Symposium. Changshu:IEEE, 2018:1426-1431.
[14] CALTAGIRONE L, BELLONE M, SVENSSON L, et al. LIDAR-camera Fusion for Road Detection Using Fully Convolutional Neural Networks[J]. Robotics and Autonomous Systems, 2019, 111:125-131.
[15] RUSSELL B C, TORRALBA A, MURPHY K P, et al. LabelMe:A Database and Web-based Tool for Image Annotation[J]. International Journal of Computer Vision, 2008, 77(1/2/3):157-173.
[16] XU X W, LIN J C, WANG X D. An Improved Densenet Method Based on Transfer Learning for Fundus Medical Images[C]//20187th International Conference on Digital Home(ICDH), Guilin:IEEE,2018:137-140.
[17] HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely Connected Convolutional Networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu:IEEE, 2017:4700-4708.
[18] JEGOU S, DROZDZAL M, VAZQUEL D, et al. The One Hundred Layers Tiramisu:Fully Convolutional Densenets for Semantic Segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. Honolulu:IEEE, 2017:11-19.
2021, Vol. 15 
