动捕设备判断不良姿势训练行为数据

瑜伽或者健美操对身体形态的改善有很好的效果，但是不规范的动作可能会产生众多负面作用。通过惯性动作捕捉设备捕捉瑜伽或健美操练习者的姿势，帮助练习者更准确地执行相关动作，结合智能优化算法，实现对操作者的实时反馈、姿势分析、进展跟踪和准确检测。本项目通过结合惯性动作捕捉设备（3DSuit Motion Capyure System），实现对作业过程中肢体三维动作的行为数据的收集与分析。我们使用鲸鱼优化算法优化支持向量机分类算法（WOA-SVM）中的超参数（惩罚参数C和核函数系数gamma）来实现训练方案的生成。在这个过程中，首先对收集到的数据进行逆运动学处理得到三维坐标，通过REBA标准对操作姿势进行量化分析，最终，得到用于WOA-RF训练的数据集，利用标注数据对随机森林模型进行训练，并通过交叉验证等方法评估模型的准确性和泛化能力。根据模型训练和测试结果，对模型参数进行优化调整，以提高模型的预测能力，得出多个指标（如预测分数、正确率、召回率等）。在此过程中，我们建立了一个动捕数据库，招募了不同性别、年龄的受试者作为数据样本，采集了他们的操作行为学数据。数据包括：1.生成的bvh文件workpose.bvh；2.借助bvh-converter库得到三维坐标点workpose_worldpos.csv；3.参照REBA标准得到的训练数据workpose.csv。Yoga or aerobics can effectively improve physical shape, but non-standard movements may lead to various adverse effects. By capturing the postures of yoga or aerobics practitioners via inertial motion capture devices, we help practitioners perform relevant movements more accurately, and combine intelligent optimization algorithms to provide real-time feedback, posture analysis, progress tracking and accurate detection for the practitioners. This project collects and analyzes the behavioral data of three-dimensional limb movements during the exercise process by integrating the 3DSuit Motion Capture System, an inertial motion capture device. We use the Whale Optimization Algorithm (WOA) to optimize the hyperparameters (penalty parameter C and kernel function coefficient gamma) in the Support Vector Machine (SVM) classification algorithm (abbreviated as WOA-SVM) to generate the training scheme. During this process, the collected data is first processed via inverse kinematics to obtain three-dimensional coordinates, and the exercise postures are quantitatively analyzed based on the REBA standard. Finally, a dataset for WOA-RF training is obtained. We train the Random Forest (RF) model using labeled data, and evaluate the model’s accuracy and generalization ability via methods such as cross-validation. According to the model training and test results, we optimize and adjust the model parameters to improve the model’s predictive performance, and derive multiple indicators including prediction score, accuracy rate, recall rate, etc. During this process, we established a motion capture database, recruited subjects of different genders and ages as data samples, and collected their operational behavioral data. The collected data includes: 1. The generated bvh file workpose.bvh; 2. The three-dimensional coordinate point file workpose_worldpos.csv obtained via the bvh-converter library; 3. The training data workpose.csv generated in accordance with the REBA standard.