Sitting posture recognition is essential in preventing work-related musculoskeletal disorders (WMSDs). WMSDs are of huge concern for office workers whose working process is averagely 81.8% sedentary. Prevailing studie...
详细信息
Sitting posture recognition is essential in preventing work-related musculoskeletal disorders (WMSDs). WMSDs are of huge concern for office workers whose working process is averagely 81.8% sedentary. Prevailing studies have utilized cameras, wearables, and pressuresensors to recognize sitting postures. The cameras and wearables can achieve accurate recognition results, while personal privacy concerns and inconvenience for long-term use impede their adoption. Meanwhile, the pressuresensors are privacy-preserving and convenient. However, they cannot accurately recognize the sitting posture with different states of the trunk, head, upper extremity, and lower extremity. Considering the pros and cons of those approaches, this study proposes a novel privacy -preserving and unobtrusive sitting posture recognition system, which combines a pressure array sensor with another privacy-preserving sensing technology, i.e., an infrared array (IRA) sensor. Moreover, a deep learning -based sitting posture recognition algorithm is developed, which adopts a feature-level fusion strategy and does not require a complex handcrafted feature extraction process. Based on the ergonomics studies, ten daily sitting postures with the states of different body parts are selected. This system achieved an overall 90.6% ac-curacy using the leave-subject-out validation approach based on the self-collected dataset from 21 subjects. It has a great potential for privacy-preserving and unobtrusive related applications for sitting posture management.
In this paper, we focus on diagnosing Parkinson's patients using dynamic plantar pressure data collected via sensor devices. We employ data preprocessing methods, including clustering, dimensionality reduction, an...
详细信息
In this paper, we focus on diagnosing Parkinson's patients using dynamic plantar pressure data collected via sensor devices. We employ data preprocessing methods, including clustering, dimensionality reduction, and multichannel feature screening. Our approach proposes a comprehensive set of data processing techniques, including data cleaning, constrained clustering, and dimensionality reduction, to convert sensor data into a multichannel multivariate time series suitable for neural network input. Unlike current methods that use all features for automatic filtering by the network - adding complexity and resource burden - we introduce a data analysis method combining statistical features and Recursive Feature Elimination. This reduces the number of channels and simplifies the model. We used a simplified 1D-convnet model, achieving a 10-fold accuracy of 91.09%, segmentation accuracy of 95.54%, individual accuracy of 97.33%, weighted precision of 95.71%, weighted recall of 95.56%, and a weighted F1-score of 95.61%. Our results validate the effectiveness of our data acquisition and feature screening methods, and notably, our processing speed is nearly three times faster.
暂无评论