This paper proposes a method for excavator pose recognition using synthetic image datasets for model training. Initially, virtual models and scenarios are established. Programming is utilized to randomize the excavator pose, virtual camera position, and scene parameters. Subsequently, keypoint coordinates and occlusion information are computed to construct synthetic image datasets. Finally, excavator key points estimation is performed using a monocular camera. Experimental results demonstrate that training with synthetic image datasets improves model recognition accuracy, with a normalized error of 0.005 6 and a percentage of correct keypoint of 97.64%. Therefore, this method can meet the practical application needs of monitoring excavator operation safety and work efficiency. It also avoids issues such as high safety risks, high time/economic costs, narrow working condition coverage, and low label accuracy associated with high-quality engineering dataset collection. This contributes to the application and deployment of deep learning and big data technologies in excavator work state recognition.
此外,有些研究识别了挖掘机关键点的3D坐标,并据此开发了3D位姿估计系统,分别为立体视觉法和平面视觉法[16]。立体视觉法一般利用具有特定位姿关系的多台相机,或者直接采用深度摄像机[18],但各有局限:结构光法深度相机由于使用红外线测量距离,不适合在室外使用,且其体积大、功耗高;飞行时间(Time of flight,TOF)深度相机需要多次采样积分,测量时间较长,在测量运动物体时可能产生运动模糊,不能满足挖掘机对实时性和精度的要求;双目相机需要精确标定两台相机的位姿关系,而挖掘机应用场景中的恶劣条件很容易破坏相机之间的相对位置,使标定数据失效,因此,该类方法在实际应用时流程烦琐、可控性差。而平面视觉法训练数据集的获取往往需要到挖掘机工作现场拍摄影像数据并依靠传感器及人工添加标签信息,不仅同样存在上述传感器方法的缺陷,还存在前述获取高质量工程数据集困难的问题[15]。此外,训练出的模型对与数据集中所包含的挖掘机有较大差别的目标识别效果较差,如需提高识别精度则需添加新的数据集。
YuXiang-jun, HuaiYuan-hui, YaoZong-wei, et al. Key technologies in autonomous vehicle for engineering[J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(4): 1153-1168.
FangCheng, YuSheng-xin, LiYong-gang, et al. Deep learning-based computer vision for health monitoring in civil engineering[J]. Journal of Tongji University(Natural Science), 2024, 52(2): 213-222.
[5]
YaoZ W, ZhaoS C, TanX D, et al. Real-time task-oriented continuous digging trajectory planning for excavator arms[J]. Automation in Construction, 2023, 152: No.104916.
[6]
ParkJ, ChenJ D, ChoY K. Self-corrective knowledge-based hybrid tracking system using bim and multimodal sensors[J]. Advanced Engineering Informatics, 2017, 32: 126-138.
[7]
VahdatikhakiF, HammadA, SiddiquiH. Optimization-based excavator pose estimation using real-time location systems[J]. Automation in Construction, 2015, 56: 76-92.
WangTai-hai, ChenJian-hong, JinJun. Mining attitude monitoring system based on GNSS precise positioning of excavator[J]. Gold Science and Technology, 2016, 24(4): 101-106.
[10]
YangW J, ZhangX H, MaH W, et al. Infrared leds-based pose estimation with underground camera model for boom-type roadheader in coal mining[J]. IEEE Access, 2019, 7: 33698-33712.
[11]
AssadzadehA, ArashpourM, BrilakisI, et al. Vision-based excavator pose estimation using synthetically generated datasets with domain randomization[J]. Automation in Construction, 2022, 134: No.104089.
WeiZhen-zhong, FengGuang-kun, ZhouDan-ya, et al. A review of position and orientation visual measurement methods and applications[J]. Laser & Optoelectronics Progress, 2023, 60(3): 135-167.
WangLian-ming, WuXin. Method for 3D motion parameter measurement based on pose estimation[J]. Journal of Jilin University(Engineering and Technology Edition), 2023, 53(7): 2099-2108.
WuHao. Research on deep learning methods for image processing based on synthetic datasets[D]. Lanzhou: School of Information Science and Engineering, Lanzhou University, 2021.
[20]
KimJ, KimD, LeeS, et al. Hybrid DNN training using both synthetic and real construction images to overcome training data shortage[J]. Automation in Construction, 2023, 149: No.104771.
[21]
AnX H, ZhouL, LiuZ G, et al. Dataset and benchmark for detecting moving objects in construction sites[J]. Automation in Construction, 2021, 122: No.103482.
[22]
LiangC J, KamatV R, MenassaC M. Real-time construction site layout and equipment monitoring[C]∥Construction Research Congress, New Orleans,USA,2018: 64-74.
[23]
ZhangS B, ZhangL J. Construction site safety monitoring and excavator activity analysis system[J]. Construction Robotics, 2022, 6: 151-161.
GuoXiao-xin, LiJia-hui, ZhangBao-liang. Joint segmentation of optic cup and disc based on high resolution network[J]. Journal of Jilin University(Engineering and Technology Edition), 2023, 53(8): 2350-2357.
[26]
TangJ Y, ZhangX, WongP K Y, et al. Method on pose estimation of excavators based on onboard depth camera[J]. IOP Conference Series: Earth and Environmental Science, 2022, 1101:No. 072005.
[27]
LinT Y, MaireM, BelongieS, et al. Microsoft coco: common objects in context[J]. The 13th European Conference on Computer Vision, 2014, 8693: 740-755.
[28]
IonescuC, PapavaD, OlaruV, et al. Human3.6M: large scale datasets and predictive methods for 3d human sensing in natural environments[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2014, 36(7): 1325-1339.
[29]
WenL Y, KimD, LiuM Y, et al. 3D excavator pose estimation using projection-based pose optimization for contact-driven hazard monitoring [J]. Journal of Computing in Civil Engineering, 2023, 37: 1-15.
ZhengYi-jie, LuoJian-xin, ChenWei-wei, et al. 3D multi-view virtual dataset construction based on Unity3D[J]. Computer Technology and Development, 2023, 33(5): 173-179.
[32]
LuoH, WangM Z, WongP K Y, et al. Full body pose estimation of construction equipment using computer vision and deep learning techniques[J]. Automation in Construction, 2020, 110: No.103016.
[33]
TorresC, RobertsD, GolparvarF M. Synthesizing pose sequences from 3D assets for vision-based activity analysis[J]. Journal of Computing in Civil Engineering, 2021, 35: No.04020052.
[34]
TianZ H, YuY, XuF, et al. Dynamic hazardous proximity zone design for excavator based on 3D mechanical arm pose estimation via computer vision[J]. Journal of Construction Engineering and Management, 2023, 149: 1-17.
[35]
LouH T, DuanX H, GuoJ M, et al. DC-YOLOv8: small-size object detection algorithm based on camera sensor[J]. Electronics, 2023, 12: No.2323.
[36]
MahmoodB, HanS, SeoJ. Implementation experiments on convolutional neural network training using synthetic images for 3D pose estimation of an excavator on real images[J]. Automation in Construction, 2022, 133: No.103996.