Depth cameras are developing widely. One of their main virtues is that, based on their data and by applying machine learning algorithms and techniques, it is possible to perform body tracking and make an accurate three-dimensional representation of body movement. Specifically, this paper will use the Kinect v2 device, which incorporates a random forest algorithm for 25 joints detection in the human body. However, although Kinect v2 is a powerful tool, there are circumstances in which the device’s design does not allow the extraction of such data or the accuracy of the data is low, as is usually the case with foot position. We propose a method of acquiring this data in circumstances where the Kinect v2 device does not recognize the body when only the lower limbs are visible, improving the ankle angle’s precision employing projection lines. Using a region-based convolutional neural network (Mask RCNN) for body recognition, raw data extraction for automatic ankle angle measurement has been achieved. All angles have been evaluated by inertial measurement units (IMUs) as gold standard. For the six tests carried out at different fixed distances between 0.5 and 4 m to the Kinect, we have obtained (mean ± SD) a Pearson’s coefficient, r = 0.89 ± 0.04, a Spearman’s coefficient, ρ = 0.83 ± 0.09, a root mean square error, RMSE = 10.7 ± 2.6 deg and a mean absolute error, MAE = 7.5 ± 1.8 deg. For the walking test, or variable distance test, we have obtained a Pearson’s coefficient, r = 0.74, a Spearman’s coefficient, ρ = 0.72, an RMSE = 6.4 deg and an MAE = 4.7 deg.
This paper investigates the feasibility of cross-domain recognition for human activities captured using low-resolution 8 × 8 infrared sensors in indoor environments. To achieve this, a novel prototype recurrent convolutional network (PRCN) was evaluated using a few-shot learning strategy, classifying up to eleven activity classes in scenarios where one or two individuals engaged in daily tasks. The model was tested on two independent datasets, with real-world measurements. Initially, three different networks were compared as feature extractors within the prototype network. Following this, a cross-domain evaluation was conducted between the real datasets. The results demonstrated the model’s effectiveness, showing that it performed well regardless of the diversity of samples in the training dataset.
The phase of the channel state information (CSI) is underutilized as a source of information in wireless sensing due to its sensitivity to synchronization errors of the signal reception. A linear transformation of the phase is commonly applied to correct linear offsets and, in a few cases, some filtering in time or frequency is carried out to smooth the data. This paper presents a novel processing method of the CSI phase to improve the accuracy of human activity recognition (HAR) in indoor environments. This new method, coined Time Smoothing and Frequency Rebuild (TSFR), consists of performing a CSI phase sanitization method to remove phase impairments based on a linear regression transformation method, then a time domain filtering stage with a Savitzy-Golay (SG) filter for denoising purposes and, finally, the phase is rebuilt, eliminating distortions in frequency caused by SG filtering. The TSFR method has been tested on five datasets obtained from experimental measurements, using three different deep learning algorithms, and compared against five other types of CSI phase processing. The results show an accuracy improvement using TSFR in all the cases. Concretely, accuracy performance higher than 90\% in most of the studied scenarios has been achieved with the proposed solution. In few-shot learning strategies, TSFR outperforms the state-of-the-art performance from 35% to 85%.
In this study, a new wireless electronic circuitry to analyze weight distribution was designed and incorporated into a chair to gather data related to common human postures (sitting and standing up). These common actions have a significant impact on various motor capabilities, including gait parameters, fall risk, and information on sarcopenia. The quality of these actions lacks an absolute measurement, and currently, there is no qualitative and objective metric for it. To address this, the designed analyzer introduces variables like Smoothness and Percussion to provide more information and objectify measurements in the assessment of stand-up/sit-down actions. Both the analyzer and the proposed variables offer additional information that can objectify assessments depending on the clinical eye of the physicians.