机器视觉技术是目前最常见的无标记人体步态分析手段之一，对于帕金森、脑卒中等患者的康复评估与治疗具有重要意义。现有的基于机器视觉技术的无标记步态分析方法对人体关节定位的准确性较差，导致步态参数精度较低，且需要手动对相机进行标定，操作繁杂。因此，本文研究了基于深度相机的无标记人体步态分析方法，以进一步提高人体步态参数的计算精度以及步态分析方法的自动化程度。

针对现有基于机器视觉技术的无标记步态分析方法标定过程复杂、关节定位不准确等问题，提出了基于人体骨架点云提取的步态分析方法。首先，在对点云进行背景去除后，利用点云分割算法与图像边缘检测算法将点云数据由相机坐标系变换至世界坐标系；其次，利用YOLO-Pose模型对人体点云进行分割，并引入点云骨架提取技术，通过点云中的单个点与其邻域点计算对应的骨架点的位置，从深度相机采集到的2.5D点云中提取人体下肢骨架点云，进而利用点云拟合算法拟合其位置信息，将拟合结果作为人体骨骼的位置信息，实现人体骨骼的准确定位；最后利用峰值检测算法对人体步态周期与步态时相进行划分，并根据划分结果计算相应的步态参数。

为了验证本文提出的步态分析方法的有效性，以基于标记的方法得到的步态参数为真值，对本文方法的检测结果进行评价分析，并和基于Kinect相机的传统步态分析方法进行对比。实验结果表明本文方法检测到步态参数与真值之间具有良好的相关性和一致性，且平均绝对误差较小，步长、抬脚高度等参数误差均在16~26mm之内，相较基于Kinect的方法提升了45.8%~66.7%，本文方法检测到的膝、髋关节角度误差分别为1.08°和2.62°，相较基于Kinect方法分别提升了64.5%与29.6%，可以对人体步态参数进行有效测量。

Machine vision technology is currently one of the most common unlabeled human gait analysis methods, which is of great significance for the rehabilitation evaluation and treatment of patients with Parkinson's disease, stroke, and other conditions. The existing machine vision technology-based unlabeled gait analysis methods have poor accuracy in human joint localization, resulting in low accuracy of gait parameters and the need for manual camera calibration, making the operation cumbersome. Therefore, this article studies an unmarked human gait analysis method based on depth cameras to further improve the calculation accuracy of human gait parameters and the automation level of gait analysis methods.

A gait analysis method based on human skeleton point cloud extraction is proposed to address the problems of complex calibration process and inaccurate joint positioning in existing machine vision technology-based unlabeled gait analysis methods. Firstly, after removing the background of the point cloud, the point cloud data is transformed from the camera coordinate system to the world coordinate system using point cloud segmentation algorithms and image edge detection algorithms; Secondly, the YOLO-Pose model is used to segment the human point cloud, and point cloud skeleton extraction technology is introduced. The position of the corresponding skeleton point is calculated based on a single point in the point cloud and its neighboring points. The human lower limb skeleton point cloud is extracted from the 2.5D point cloud collected by the depth camera, and then the position information is fitted using the point cloud fitting algorithm. The fitting result is used as the position information of the human skeleton, achieving accurate positioning of the human skeleton; Finally, the peak detection algorithm is used to partition the human gait cycle and gait phase, and the corresponding gait parameters are calculated based on the partition results.

In order to verify the effectiveness of the gait analysis method proposed in this article, the detection results of our method were evaluated and analyzed using the gait parameters obtained from the label based method as true values, and compared with traditional gait analysis methods based on Kinect cameras. The experimental results show that the proposed method has a good correlation and consistency between gait parameters and true values, and the average absolute error is small. The errors of parameters such as step length and foot lift height are all within 16~26mm, which is 45.8%~66.7% higher than the Kinect based method. The angle errors of the knee and hip joints detected by the proposed method are 1.08° and 2.62°, respectively, which are 64.5% and 29.6% higher than the Kinect based method. This method can effectively measure human gait parameters.

[1]王忠彦, 霍家康, 郭欣等. 康复学科发展的中国实践[J]. 中国科学:生命科学, 2023, 53(01): 131-134.

[2]国务院关于印发“十四五”残疾人保障和发展规划的通知[J]. 中华人民共和国国务院公报, 2021, 22: 25-38.

[3]Fu Cheng Cao, Xiao Xue Xing. Gait Analysis and Trajectory Planning for Lower Limb Rehabilitation Training[J]. Applied Mechanics and Materials, 2014, 3546(672): 1940-1943.

[4]王人成, 张美芹. 人体步态时相对称性评价指标的对比研究[J]. 中国康复医学杂志, 2011, 26(10): 957-959.

[5]T. Suzuki, Y. Ito, T. Ito, et al. Pathological gait in Rett syndrome: Quantitative evaluation using three-dimensional gait analysis[J]. European Journal of Paediatric Neurology, 2023, 42: 15-21.

[6]陈忠强, 康赵颖, 占道伟等. 基于反复促通疗法的步行训练对脑卒中患者下肢功能重建的影响[J]. 中国康复, 2024, 39(02): 81-84.

[7]孙中川, 王秋根, 张栋等. 步态分析应用于脊髓型颈椎病的研究进展[J]. 国际骨科学杂志, 2024, 45(01): 43-45.

[8]M. Mcgrath, J. Wood, J. Walsh, et al. The impact of Three-Dimensional Gait Analysis in adults with pathological gait on management recommendations[J]. Gait & Posture, 2023, 105: 75-80.

[9]励建安, 孟殿怀. 步态分析的临床应用[J]. 中华物理医学与康复杂志, 2006, 28(07): 500-503.

[10]M. V. Munteanu, M. D. Stanciu, S. Vlase. Image Correlation to Predict the Gait Recovery of Patients in the Post-Surgery Period[C]//Moldovan L. 13th International Conference Interdisciplinarily in Engineering (INTER-ENG 2019). Amsterdam: Elsevier Science Bvsara Burgerhartstraat, 2020: 499-508.

[11]Nathaniel Goldfarb, Alek Lewis, Alex Tacescu, et al. Open source Vicon Toolkit for motion capture and Gait Analysis[J]. Computer Methods and Programs in Biomedicine, 2021, 212: 106414.

[12]罗豪龙, 李建胜, 邹丹平等. 大场景光学运动捕捉系统标定方法研究[J]. 光子学报, 2023, 52(11): 222-236.

[13]Dimple Sethi, Sourabh Bharti, Chandra Prakash. A comprehensive survey on gait analysis: History, parameters, approaches, pose estimation, and future work[J]. Artificial Intelligence in Medicine, 2022, 129: 102314.

[14]Y. He, Y. X. Chen, L. Tang, et al. Accuracy validation of a wearable IMU-based gait analysis in healthy female[J]. Bmc Sports Science Medicine and Rehabilitation, 2024, 16(1): 2.

[15]徐婷, 孙哲, 方剑. 用于步态分析的足底压力监测系统研究进展[J]. 科学通报, 2024, 69(Z1): 565-577.

[16]余炳威, 赖舒婷, 詹润源等. 基于混合注意力的轻量级偏瘫步态评估系统[J]. 计算机系统应用, 2023, 32(10): 96-105.

[17]Takemura N, Makihara Y, Muramatsu D, et al. On input/output architectures for convolutional neural network-based cross-view gait recognition[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2017, 29(9): 2708-2719.

[18]Man Ju, Bhanu Bir. Individual recognition using gait energy image[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2006, 28(2): 16-22.

[19]Connor P, Ross A. Biometric recognition by gait: A survey of modalities and features[J]. Computer vision and image understanding, 2018, 167: 1-27.

[20]Megumi Ota, Hiroshige Tateuchi, Takaya Hashiguchi, et al. Verification of validity of gait analysis systems during treadmill walking and running using human pose tracking algorithm[J]. Gait & Posture, 2021, 85: 290-297.

[21]Jaime Diaz-Arancibia, Matias Cordova, Jeferson Arango-Lopez, et al. Automatic gait analysis through computer vision: a pilot study[J/OL]. Neural Computing & Applications, 2023-04-08.

[22]Saman Vafadar, Wafa Skalli, Aurore Bonnet-Lebrun, et al. Assessment of a novel deep learning-based marker-less motion capture system for gait study[J]. Gait & Posture, 2022, 94: 138-143.

[23]Jan Stenum, Cristina Rossi, Ryan T. Roemmich. Two-dimensional video-based analysis of human gait using pose estimation[J]. Plos Computational Biology, 2021, 17(4): e1008935.

[24]Wu J, Huang J, Wu X, et al. A novel graph-based hybrid deep learning of cumulative GRU and deeper GCN for recognition of abnormal gait patterns using wearable sensors[J]. Expert Systems with Applications, 2023, 233: 120968.

[25]Yoshitomo Saiki, Tamon Kabata, Tomohiro Ojima, et al. Reliability and validity of OpenPose for measuring hip-knee-ankle angle in patients with knee osteoarthritis[J]. Scientific Reports, 2023, 13(1): 3297.

[26]田皓宇, 马昕, 李贻斌. 基于骨架信息的异常步态识别方法[J]. 吉林大学学报(工学版), 2022, 52(04): 725-737.

[27]T. Usami, K. Nishida, H. Iguchi, et al. Evaluation of lower extremity gait analysis using Kinect V2 tracking system[J]. Sicot-J, 2022, 8:1-7.

[28]唐心宇, 宋爱国. 人体姿态估计及在康复训练情景交互中的应用[J]. 仪器仪表学报, 2018, 39(11): 195-203.

[29]黄秋晗. 基于彩色信息与深度信息的步态分析系统设计与实现[D]. 湖北: 华中科技大学, 2019.

[30]柯锦文. 基于Kinect的步态分析系统设计与实现[D]. 湖北: 华中科技大学, 2020.

[31]荀小飞. 基于Kinect的人体步态分析和重心估算方法研究[D]. 四川: 电子科技大学, 2021.

[32]K. Otte, T. Ellermeyer, T. S. Vater, et al. Instrumental Assessment of Stepping in Place Captures Clinically Relevant Motor Symptoms of Parkinson's Disease[J]. Sensors, 2020, 20(19): 5465.

[33]滕君敏, 马晔, 刘东威等. 基于深度学习融合算法的无标记点步态分析系统[J]. 医用生物力学, 2022, 37(04): 733-740.

[34]滕君敏. 基于Kinect深度传感器的无标记点步态分析系统研究[D]. 浙江: 宁波大学, 2021.

[35]Enrico Martini, Michele Boldo, Stefano Aldegheri, et al. Enabling Gait Analysis in the Telemedicine Practice through Portable and Accurate 3D Human Pose Estimation[J]. Computer Methods and Programs in Biomedicine, 2022, 225: 107016.

[36]刘朝云. 基于深度相机的无标记人体步态捕捉方法及系统研究[D]. 浙江: 浙江工业大学, 2020.

[37]Sumit Hazra, Acharya Aditya Pratap, Dattatreya Tripathy, et al. Novel data fusion strategy for human gait analysis using multiple kinect sensors[J]. Biomedical Signal Processing and Control, 2021, 67: 102512.

[38]刘驰,周晋,陈雨. Openpose算法在步态分析中的可靠性研究[J]. 皮革科学与工程, 2021, 31(01): 65-68.

[39]Wade Logan, Needham Laurie, Mcguigan Polly, et al. Applications and limitations of current markerless motion capture methods for clinical gait biomechanics[J]. PeerJ, 2022, 10: e12995.

[40]林绿开, 钮倩倩李毅. 基于棋盘标定板的优化相机参数标定方法[J]. 计算机技术与发展, 2023, 33(12): 101-105.

[41]秦仕昊. 下肢矢状面运动的穿戴式软传感测量系统研究[D]. 黑龙江: 哈尔滨工业大学, 2019.

[42]B. X. Wang, J. H. Lan, J. J. Gao. LiDAR Filtering in 3D Object Detection Based on Improved RANSAC[J]. Remote Sensing, 2022, 14(9): 2110-2128.

[43]Cao Junjie, Tagliasacchi A, Olson M, et al. Point cloud skeletons via laplacian based contraction[J]. 2010 Shape Modeling International Conference, 2010, 187-197.

[44]A. Tagliasacchi, H. Zhang, D. Cohen-Or. Curve Skeleton Extraction from Incomplete Point Cloud[J]. Acm Transactions on Graphics, 2009, 28(3): 71.

[45]H. Huang, S. H. Wu, D. Cohen-Or, et al. L1-Medial Skeleton of Point Cloud[J]. Acm Transactions on Graphics, 2013, 32(4): 65.

[46]H. X. Qin, J. Han, N. Li, et al. Mass-Driven Topology-Aware Curve Skeleton Extraction from Incomplete Point Clouds[J]. Ieee Transactions on Visualization and Computer Graphics, 2020, 26(9): 2805-2817.

[47]Jun K, Lee D W, Lee K, et al. Feature extraction using an RNN autoencoder for skeleton-based abnormal gait recognition[J]. IEEE Access, 2020, 8: 19196-19207.

[48]Zeng Y, Zhang X, Kawasumi Y, et al. A 2.5 D deep learning-based method for drowning diagnosis using post-mortem computed tomography[J]. IEEE journal of biomedical and health informatics, 2022, 27(2): 1026-1035.

[49]韩松魁, 于正林. 基于车载激光雷达点云数据的预处理研究[J]. 长春理工大学学报(自然科学版), 2024, 47(01): 85-91.

[50]舒军, 邓明舟, 雷建军等. 基于统计滤波和稀疏度优化的印花配准算法[J]. 华中师范大学学报(自然科学版), 2021, 55(04): 559-566.

[51]李睿, 林海荣, 吴小燕. 基于稳健移动最小二乘法的点云数据拟合[J]. 测绘与空间地理信息, 2017, 40(05): 122-124.

[52]Maji D, Nagori S, Mathew M, et al. YOLO-Pose: Enhancing YOLO for Multi Person Pose Estimation Using Object Keypoint Similarity Loss[C]//2022 IEEE/CFV Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2022. New York: IEEE, 2022: 2636-2645.

[53]J. J. Cao, H. Chen, J. Zhang, et al. Normal estimation via shifted neighborhood for point cloud[J]. Journal of Computational and Applied Mathematics, 2018, 329: 57-67.

[54]M. Yamamoto, K. Shimatani, M. Hasegawa, et al. Accuracy of Temporo-Spatial and Lower Limb Joint Kinematics Parameters Using OpenPose for Various Gait Patterns With Orthosis[J]. Ieee Transactions on Neural Systems and Rehabilitation Engineering, 2021, 29: 2666-2675.

[55]白桂峰. 基于MEMS惯性传感器的人体步态分析关键技术研究[D]. 山西: 中北大学, 2020.

[56]查祖福水,白梅娟,魏永勇等.基于行人检测与峰值密度聚类的行人多次徘徊检测算法[J].电脑与信息技术,2023,31(04):24-27.

[57]李硕,韩迎东,王双,等.基于Pearson相关系数的图像误匹配点剔除算法[J].激光与光电子学进展,2021,58(08):263-273.

[58]B. Langley, M. Greig. The gait abnormality index: A summary metric for three-dimensional gait analysis[J]. Gait & Posture, 2023, 105: 87-91.

[59]郭楚珊. 基于Pearson相关性的VAR模型煤电水煤比寻优应用[J]. 科学技术创新, 2024, 03: 223-228.

[60]朱红梅, 高蕾, 谢冰等.两种测量方法在成年女性冠状面形体测量中的Bland-Altman一致性评价[J]. 中国妇产科临床杂志, 2023, 24(03): 285-288.