空地协同场景下特殊障碍物数据集与检测算法评估系统构建

doi:10.12382/bgxb.2024.0583

摘要/Abstract

摘要：

在空地协同场景下,特殊障碍物的识别与处理对地面装备安全运行至关重要。针对非结构化环境中样本稀缺的问题,构建了包含33 124张图像的检测数据集,覆盖多类典型特殊障碍物,支持复杂场景下的识别任务。为准确评估检测算法性能,设计融合类别信息与定位精度的综合评价指标,增强模型对比的科学性。提出结合物理属性与环境语义的可通行性分析方法,为地面无人系统路径规划提供依据。实验结果表明,该数据集与评估体系显著提升检测精度,所提方法能有效识别坑洞、水面等典型特殊障碍物。

关键词: 特殊障碍物检测, 空地协同, 数据集构建, 通行性策略

Abstract:

The recognition and handling of special obstacles are critical to ensuring the safe operation of ground equipment in air-ground collaborative scenarios.To address the lack of data samples in unstructured environments,a detection dataset comprising 33124 images is developed,featuring a wide range of typical special obstacles to support the recognition tasks in complex scenes.A comprehensive evaluation index integrating the category information and localization accuracy is designed to enhance the objectivity and reliability of algorithm performance comparisons.Additionally,a passability analysis method is proposed,which combines the physical properties with the environmental semantics to guide path planning for unmanned ground systems.Experimental results demonstrate that the proposed dataset and evaluation framework significantly improve detection accuracy,and the method effectively identifies typical obstacles such as potholes and water surfaces in unstructured environments.

Key words: special obstacle detection, ground-air collaboration, dataset construction, passability strategy

中图分类号:

TP242

冷呈宇, 赵津, 刘畅, 杨世凤. 空地协同场景下特殊障碍物数据集与检测算法评估系统构建[J]. 兵工学报, 2025, 46(6): 240583-.

LENG Chengyu, ZHAO Jin, LIU Chang, YANG Shifeng. Construction of Special Obstacle Dataset and Detection Algorithm Evaluation System in Air-ground Collaborative Scenarios[J]. Acta Armamentarii, 2025, 46(6): 240583-.

图/表 18

图1 基于特殊障碍物检测的实现流程

Fig.1 Implementation process based on special obstacle detection

表1 障碍物数据集视角、类别与场景覆盖对比分析

Table 1 Comparative analysis of viewpoints,categories,and scene coverage in obstacle detection datasets

数据集	视角	障碍物类别	场景覆盖
GRACO数据集^[6]	地面, 空中	建筑相关障碍物	静态城市建筑
UDTIRI数据集^[7]	地面	道路坑洞	城市道路
GRDDC数据集^[8]	地面	道路裂缝、坑洞	道路环境
NPDOID数据集^[21]	地面	坑洼障碍物	道路环境
本文数据集	地面, 空中	常规障碍物(行人、车辆等) 特殊障碍物(石块、水坑等)	空地协同动态场景

表2 潜在特殊障碍物、特殊障碍物类别名称及标注数目

Table 2 Category names and annotation counts of potential special obstacles and special obstacles

特殊障碍物对象	标注数目	潜在特殊障碍物对象	标注数目
坑洞	48509	物料堆	1684
水面	22384	管道	2383
沟槽	1315	石块	21948

表3 常规障碍物类别名称及标注数目

Table 3 Category names and annotation counts of regular obstacles

常规障碍物对象	标注数目	常规障碍物对象	标注数目
三轮车	3844	摩托车	36107
自行车	11678	行人	95716
公交车	8241	卡车	14626
汽车	168283	面包车	29948

图2 目标检测结果

Fig.2 The results of object detection

图3 采用RGB-D相机识别坑洞到UAV的距离

Fig.3 The distance from pothole to UAV identified by RGB-D camera

图4 边界扩展策略计算坑洞深度

Fig.4 Boundary extension strategy to calculate the depth of potholes

图5 使用RGB-D相机识别水面、水坑到UAV的距离

Fig.5 Use RGB-D camera to identify the distance between the water surface, puddle and UAV

图6 水面、水坑等障碍物与视觉地图的交互

Fig.6 Interaction between obstacles,such as water surfaces and puddles,and visual maps

图7 算法1:坑洞、沟槽检测与深度计算

Fig.7 Algorithm 1:pothole and trench detection and depth calculation

图8 算法2:水面、水坑检测与三维建模

Fig.8 Algorithm 2:water surface and pothole detection and 3D modeling

图9 UGV可通行性的工况分析

Fig.9 Condition analysis of UGV passability

表4 硬件配置

Table 4 Hardware configuration

硬件	类型	设备规格
智能装备1	UAV	JETSON Xavier开发板;8核;32GB内存;RGB-D深度相机,IMU和GPS传感器
智能装备2	线控底盘车	Nuvo-8108GC;Intel Core i9-9900K CPU;32GB

表5 YOLO系列模型训练过程可视化

Table 5 Visualization of YOLO series model training process

模型	训练损失	CS-mAP	模型	训练损失	CS-mAP
YOLOv2			YOLOv5
YOLOv3			YOLOv6
YOLOv4			YOLOv7
YOLOv8			YOLOv10
YOLOv9			YOLOX

表6 特殊障碍物数据集的目标检测精度对比

Table 6 Comparison of object detection accuracies for special obstacle datasets

网络模型	参数量/ 10⁶	坑洞 CSP/%	水面 CSP/%	沟槽 CSP/%	CS-mAP/%	CEI
YOLOv2	23	81.02	83.24	77.45	80.57	24.25
YOLOv3	240	85.95	89.13	84.26	86.44	28.49
YOLOv4	195	84.33	89.39	81.62	85.11	26.81
YOLOv5	27	88.29	87.82	83.35	86.48	27.31
YOLOv6	80	85.66	92.46	86.15	87.09	27.33
YOLOv7	145	97.61	93.05	90.92	93.86	30.26
YOLOv8	173	97.64	93.15	90.38	93.16	30.26
YOLOv9	122	98.27	94.23	91.12	95.11	31.08
YOLOv10	211	99.11	96.16	93.27	96.27	31.97
YOLOX	67	89.93	90.19	89.86	89.99	27.22

表7 特殊障碍物检测可视化

Table 7 Visualization of special obstacle detection

模型	序号
模型	1	2	3	4	5	6
YOLOv2
YOLOv3
YOLOv4
YOLOv5
YOLOv6
YOLOv7
YOLOv8
YOLOv9
YOLOv10
YOLOX

图10 UAV与坑洞、沟槽的交互过程

Fig.10 The interaction process among UAV and potholes and trenches

图11 UAV与水面、水坑的交互过程

Fig.11 Interaction process between UAV and water surfaces/puddles

参考文献 22

[1]	MASIHULLAH S, GARG R, MUKHERJEE P, et al. Attention based coupled framework for road and pothole segmentation[C]// Proceedings of the 2020 25th International Conference on Pattern Recognition. Milan,Italy: IEEE, 2021:5812-5819.
[2]	SUN X Z, XIE Y C, JIANG L M, et al. DMA-Net:deeplab with multi-scale attention for pavement crack segmentation[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(10):18392-18403.
[3]	LIU Z J, FAN G Y, RAO L, et al. Positive and negative obstacles detection based on dual-LiDAR in field environments[J]. IEEE Robotics and Automation Letters, 2024, 9(8):6768-6775.
[4]	RAN H H, SUN L, CHENG S, et al. A novel cooperative searching architecture for multi-unmanned aerial vehicles under restricted communication[J]. Asian Journal of Control. 2022, 24(2):510-516.
[5]	LIU C, ZHAO J, SUN N Y. A review of collaborative air-ground robots research[J]. Journal of Intelligent & Robotic Systems, 2022, 106(3):1-23.
[6]	ZHU Y L, KONG Y, JIE Y G, et al. GRACO:a multimodal dataset for ground and aerial cooperative localization and mapping[J]. IEEE Robotics and Automation Letters, 2023, 8(2):966-973.
[7]	GUO S C, LI J H, FENG Y, et al. UDTIRI:an online open-source intelligent road inspection benchmark suite:arXiv:2304.08842[R/OL]. Ithaca,NY,US:Cornell Universit, 2023:2304.08842. https://arxiv.org/abs/2304.08842.
[8]	ARYA D, MAEDA H, GHOSH S K, et al. Global road damage detection:state-of-the-art solutions[C]// Proceedings of the 2020 IEEE International Conference on Big Data.Atlanta,GA,US:IEEE, 2020:5533-5539.
[9]	KULAMBAYEV B, NURLYBEK M, ASTAUBAYEVA G, et al. Real-time road surface damage detection framework based on mask R-CNN model[J]. International Journal of Advanced Computer Science and Applications, 2023, 14(9):757-765.
[10]	HERNANDA Z S, MAHMUDAH H, SUDIBYO R W. CNN-based hyperparameter optimization approach for road pothole and crack detection systems[C]// Proceedings of the 2022 IEEE World AI IoT Congress.Seattle,WA,US:IEEE, 2022:538-543.
[11]	FENG Z, GUO Y N, SUN Y X. Segmentation of road negative obstacles based on dual semantic-feature complementary fusion for autonomous driving[J]. IEEE Transactions on Intelligent Vehicles, 2024, 9(4):4687-4697.
[12]	FENG Z, GUO Y N, NAVARRO-ALARCON D, et al. InconSeg:residual-guided fusion with inconsistent multi-modal data for negative and positive road obstacles segmentation[J]. IEEE Robotics and Automation Letters, 2023, 8(8):4871-4878.
[13]	赵一兵, 郭烈, 李明丽, 等. 无人地面车辆野外水体障碍物识别[J]. 兵工学报, 2014, 35 (增刊1):62-67.
	ZHAO Y B, GUO L, LI M L, et al. Recognition of water obstacles for unmanned ground vehicles in the field[J]. Acta Armamentarii, 2014, 35(S1):62-67. (in Chinese)
[14]	刘家银, 唐振民, 王安东, 等. 基于多激光雷达与组合特征的非结构化环境负障碍物检测[J]. 机器人, 2017, 39(05):638-651.
	LIU J Y, TANG Z M, WANG A D, et al. Detection of negative obstacles in unstructured environments based on multi-lidar and combined features[J]. Robot, 2017, 39(5):638-651. (in Chinese)
[15]	WANG H, ZHANG Q, FAN Z Q, et al. Towards an obstacle detection system for robot obstacle negotiation[J]. Industrial Robot-the International Journal of Robotics Research and Application, 2024, 51(2):236-245.
[16]	RUAN S L, LI S B, LU C W, et al. A real-time negative obstacle detection method for autonomous trucks in open-pit mines[J]. Sustainability, 2022, 15(1):120.
[17]	阮顺领, 鄢盛钰, 顾清华, 等. 基于多特征融合的露天矿区道路负障碍检测[J]. 煤炭学报, 2024, 49(5):2561-2572.
	RUAN S L, YAN S Y, GU Q H, et al. Obstacle detection on open-pit mining roads based on multi-feature fusion[J]. Journal of China Coal Society, 2024, 49(5):2561-2572. (in Chinese)
[18]	DODGE D, YILMAZ M. Convex vision-based negative obstacle detection framework for autonomous vehicles[J]. IEEE Transactions on Intelligent Vehicles, 2022, 8(1):778-789.
[19]	付浩龙, 赵津, 席阿行, 等. 无人机视觉下基于改进OTSU的负障碍物检测[J]. 计算机应用与软件, 2020, 37(9):105-110.
	FU H L, ZHAO J, XI A X, et al. Obstacle detection based on improved OTSU in drone vision[J]. Journal of Computer Applications and Software, 2020, 37(9):105-110. (in Chinese)
[20]	GOODIN C, CARRILLO J, MONROE J G, et al. An analytic model for negative obstacle detection with lidar and numerical validation using physics-based simulation[J]. Sensors, 2021, 21(9):3211.
[21]	金政北, 金贝贝, 宋晓辉, 等. 改进YOLOv10算法及其在路面坑洼检测中的应用[J/OL]. 计算机应用与软件, 2025(2025-03-11). http://kns.cnki.net/kcms/detail/31.1260.tp.20250307.0824.002.htm.
	JIN Z B, JIN B B, SONG X H, et al. Improvement of YOLOv10 algorithm and its application in pavement pothole detection[J/OL]. Computer Applications and Software, 2025(2025-03-11). http://kns.cnki.net/kcms/detail/31.1260.tp.20250307.0824.002.htm. (in Chinese)
[22]	LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft coco:common objects in context[C]// Proceedings of the 13th European Conference on Computer Vision.Zurich,Switzerland:Springer, 2014:740-755.