低等级路面病害的轻量化边缘实时检测方法

Lightweight Edge Real-Time Detection Method for Low-Grade Pavement Diseases

摘要

针对传统路面检测车巡检效率低、设备成本高、实时性差等问题，提出一种基于边缘智能的轻量化路面病害实时检测方法。该方法采用YOLO-Trip模型，高效提取颜色和空间位置特征，结合TensorRT技术实现边缘端实时检测；针对现有里程测量存在的精度不足与部署困难等问题，设计了IMU和GNSS自校准高频里程计，结合卡尔曼滤波与线性插值算法，实现高频里程测量；构建了低功耗车载边缘计算平台，无需额外供电即可实时采集和检测路面图像。里程计对比实验表明，在0~40 km/h速度范围内，系统采样最大误差与车轮编码器相比仅为0.9%，显著优于单GNSS方案；模型对比实验表明，YOLO-Trip模型在召回率和平均精度上较基准模型均有所提升，同时参数量更少，计算量更小，减轻了边缘计算压力。系统能实时检测横向裂缝、纵向裂缝、网状裂缝和凹坑等病害，准确记录位置信息，适用于农村水泥路和山区柏油路，为路面养护提供数据支持。

Abstract

Aiming at the problems of low efficiency, high cost and poor real-time performance of traditional pavement detection vehicles, a lightweight edge real-time detection method for low-grade pavement distresses is proposed. This method adopts a novel YOLO-Trip model to efficiently extract color and spatial features, and integrates TensorRT technology to achieve real-time detection on the edge. For the existing mileage measurement challenge, an IMU and GNSS self-calibrated high-frequency odometer is designed, combined with Kalman filter and linear interpolation algorithm to realize ultra-high-frequency mileage measurement. A low-power onboard edge computing platform is built to collect and detect road surface images in real time without additional power supply. In the mileage measurement comparison experiment, the maximum sampling error of the system is only 0.9% different from that of the wheel encoder in the speed range of 0~40 km/h, which is significantly better than the single GNSS scheme. The model comparison experiment shows that the YOLO-Trip model leads the benchmark model by in recall rate and average precision, while the parameter quantity and the computational load are reduced, which alleviates the edge computing pressure. The system can detect transverse cracks, longitudinal cracks, alligator cracks and potholes and other diseases in real time, and accurately record the location information, which is suitable for rural concrete roads and mountainous asphalt roads, providing data support for road maintenance.

关键词

边缘智能 / 目标检测 / 卡尔曼滤波 / 路面病害 / 轻量化

Key words

edge intelligence / target detection / Kalman filter / pavement distress / lightweight

参考文献

[1] 易钰程, 王靖智, 朱路, 等 . 路面缺陷智能检测系统与方法综述[J]. 华东交通大学学报, 2023, 40(5): 19-31.

[2] YI Y C, WANG J Z, ZHU L, et al. Review of the intelligent pavement defect detection system and methods[J]. Journal of East China Jiaotong University, 2023, 40(5): 19-31.

[3] LIU Y Y, YE S T, WANG J Z, et al. Deep—supervised pavement crack segmentation algorithm for multiscale attention and feature fusion[J]. Photoelectron Laser, 2025, 36(11): 1140-1150.

[4] YAO H, LIU Y H, LI X, et al. A detection method for pavement cracks combining object detection and attention mechanism[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 22179-22189.

[5] LIU F Y, LIU J, WANG L B . Asphalt pavement crack detection based on convolutional neural network and infrared thermography[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 22145-22155.

[6] 胡鹏, 夏晓华, 钟预全, 等 . 采用多尺度特征增强的路面病害检测模型[J]. 西安交通大学学报, 2025, 59(2): 156-169.

[7] HU P, XIA X H, ZHONG Y Q, et al. Pavement disease detection model based on multi—scale feature reinforcement[J]. Journal of Xi’an Jiaotong University, 2025, 59(2): 156-169.

[8] 邓天民, 陈月田, 余洋, 等 . 聚焦形状特征的路面病害检测算法[J]. 计算机工程与应用, 2024, 60(24): 291-305.

[9] DENG T M, CHEN Y T, YU Y, et al. Pavement disease detection algorithm focusing on shape features[J]. Computer Engineering and Applications, 2024, 60(24): 291-305.

[10] 高晶, 姚金杰, 刘鹏杰, 等 . 基于改进YOLOv5s的路面病害检测[J]. 计算机系统应用, 2024, 33(9): 253-260.

[11] GAO J, YAO J J, LIU P J, et al. Pavement disease detection based on improved YOLOv5s[J]. Computer Systems and Applications, 2024, 33(9): 253-260.

[12] 白锋, 马庆禄, 赵敏 . 面向航拍路面裂缝检测的AC—YOLO[J]. 计算机工程与应用, 2025, 61(1): 153-164.

[13] BAI F, MA Q L, ZHAO M . AC—YOLO for aerial pavement crack detection[J]. Computer Engineering and Applications, 2025, 61(1): 153-164.

[14] KAISER S, TOSUN A Ş, KORKMAZ T . Benchmarking container technologies on ARM—based edge devices[J]. IEEE Access, 2023, 11: 107331-107347.

[15] WU Y R, GUO H F, CHAKRABORTY C, et al. Edge computing driven low—light image dynamic enhancement for object detection[J]. IEEE Transactions on Network Science and Engineering, 2022, 10(5): 3086-3098.

[16] 王葱, 宋伟东, 孙尚宇, 等 . 基于编码器和PLC的公路智能巡检测距模块设计[J]. 自动化仪表, 2023, 44(4): 12-17.

[17] WANG C, SONG W D, SUN S Y, et al. Design of highway intelligent inspection distance measurement module based on encoder and PLC[J]. Process Automation Instrumentation, 2023, 44(4): 12-17.

[18] LIU L, LI Z, LU R, et al. An enhanced smartphone GNSS/MEMS—IMU integration seamless positioning method in urban environments[J]. IEEE Sensors Journal, 2024, 24(24): 41251-41263.

[19] WANG J Q, XIONG Z, ZHAO Y, et al. Covariance matrix transformation method for absolute/relative measurements fusion of vision/IMU/GNSS integration in parafoil landing[J]. IEEE Sensors Journal, 2024, 24(10): 16673-16687.

[20] 陈博怀 . 基于GNSS/IMU/UWB融合的扑翼机器人定位技术研究[D]. 南京: 东南大学, 2023.

[21] CHEN B H . Research on positioning technology of flapping wing robots based on GNSS/IMU/UWB fusion[D]. Nanjing: Southeast University, 2023.

[22] XIAO J P, CHEE J H, FEROSKHAN M . Real—time multi—drone detection and tracking for pursuit—evasion with parameter search[J]. IEEE Transactions on Intelligent Vehicles, 2024, 9(10): 6027-6037.

基金资助

江西省研究生创新专项资金项目(2023-S478)

江西省交通运输厅科技项目(2024QN008)