Our focus in this research is on the use of deep learning approaches for human activity recognition (HAR) scenario, in which inputs are multichannel time series signals acquired from a set of body-worn inertial sensors and outputs are predefined human activities. Here, we present a feature learning method that deploys convolutional neural networks (CNN) to automate feature learning from the raw inputs in a systematic way. The influence of various important hyper-parameters such as number of convolutional layers and kernel size on the performance of CNN was monitored. Experimental results indicate that CNNs achieved significant speed-up in computing and deciding the final class and marginal improvement in overall classification accuracy compared to the baseline models such as Support Vector Machines and Multi-layer perceptron networks.

/pdf/human-activity-recognition-with-inertial-sensors-using-a-e3bg1umi3r.pdf

Human activity recognition with inertial sensors using a deep learning approach

The essential human gait parameters are briefly reviewed, followed by a detailed review of the state of the art in deep learning for the human gait analysis. The modalities for capturing the gait data are grouped according to the sensing technology: video sequences, wearable sensors, and floor sensors, as well as the publicly available datasets. The established artificial neural network architectures for deep learning are reviewed for each group, and their performance are compared with particular emphasis on the spatiotemporal character of gait data and the motivation for multi-sensor, multi-modality fusion. It is shown that by most of the essential metrics, deep learning convolutional neural networks typically outperform shallow learning models. In the light of the discussed character of gait data, this is attributed to the possibility to extract the gait features automatically in deep learning as opposed to the shallow learning from the handcrafted gait features.

/pdf/deep-learning-for-monitoring-of-human-gait-a-review-3jt8u270qk.pdf

Deep Learning for Monitoring of Human Gait: A Review

We demonstrate accurate spatio-temporal gait data classification from raw tomography sensor data without the need to reconstruct images. This is based on a simple yet efficient machine learning methodology based on a convolutional neural network architecture for learning spatio-temporal features, automatically end-to-end from raw sensor data. In a case study on a floor pressure tomography sensor, experimental results show an effective gait pattern classification F -score performance of 97.88 $\pm$ 1.70%. It is shown that the automatic extraction of classification features from raw data leads to a substantially better performance, compared to features derived by shallow machine learning models that use the reconstructed images as input, implying that for the purpose of automatic decision-making it is possible to eliminate the image reconstruction step. This approach is portable across a range of industrial tasks that involve tomography sensors. The proposed learning architecture is computationally efficient, has a low number of parameters and is able to achieve reliable classification F -score performance from a limited set of experimental samples. We also introduce a floor sensor dataset of 892 samples, encompassing experiments of 10 manners of walking and 3 cognitive-oriented tasks to yield a total of 13 types of gait patterns.

Deep Neural Networks for Learning Spatio-Temporal Features From Tomography Sensors

Accurate, reliable indoor localization or positioning is the key enabler for location-based services. Indoor localization can be broadly classified into two distinct categories. Active localization entails tracking a tag attached to or carried by the target. Passive localization, on the other hand, involves positioning a device-free or untagged target. While passive or device-free localization is comparatively more difficult to achieve, it is the preferred option for many applications. Vision-based techniques can accurately localize an untagged target. However, privacy is a significant concern and thus have limited usability for many applications, especially in noncommercial and residential settings. Passive localization using radio frequency (RF) or wireless sensing is an unobtrusive option and has seen extensive research efforts in recent years leading to a saturated research field but no consensus solution. Researchers have been investigating alternative solutions that can facilitate robust passive localization. The rapid proliferation of Internet of Things (IoT) is bringing ubiquitous networked devices and ambient sensors into modern buildings. The consequential pervasive ambient intelligence and signals of opportunity can enable unobtrusive device-free positioning. This article presents a comprehensive review of non-RF solutions covering visible light-, infrared-, physical excitation-and electric field sensing-based techniques. Limitations of the state-of-the-art and potential future research directions are also outlined.

Device-Free Localization: A Review of Non-RF Techniques for Unobtrusive Indoor Positioning

In this paper, we assess the capability of a unique unobtrusive footprint imaging sensor system, based on plastic optical fiber technology, to allow efficient gait analysis from time domain sensor data by pattern recognition techniques. Trial gait classification experiments are executed as ten manners of walking, affecting the amplitude and frequency characteristics of the temporal signals. The data analysis involves the design of five temporal features, subsequently analyzed in 14 different machine learning models, representing linear, non-linear, ensemble, and deep learning models. The model performance is presented as cross-validated accuracy scores for the best model-feature combinations, along with the optimal hyper-parameters for each of them. The best classification performance was observed for a random forest model with the adjacent mean feature, yielding a mean validation score of 90.84% ± 2.46%. We conclude that the floor sensor system is capable of detecting changes in gait by means of pattern recognition techniques applied in the time domain. This suggests that the footprint imaging sensor system is suitable for gait analysis applications ranging from healthcare to security.

Temporal Pattern Recognition in Gait Activities Recorded With a Footprint Imaging Sensor System

We report on the photonic variant of the previously introduced Guided-Path Tomography (GPT), by demonstrating a system for footstep imaging using Plastic Optical Fiber (POF) sensors. The 1 m × 2 m sensor head is manufactured by attaching 80 POF sensors on a standard commercial carpet underlay. The sensing principle relies on the sensitivity of POF to bending, quantified by measuring light transmission. The Photonic GPT (PGPT) system, comprising the sensor head with processing hardware and software, covered by a mass-production general-purpose carpet top, successfully performs footstep imaging and correctly displays the position and footfall of a person walking on the carpet in real time. We also present the implementation of fast footprint “center of mass” calculations, suitable for recording gait and footfall. A split-screen movie, showing the frame-by-frame camera-captured action next to the reproduced footprints is included in the electronic version of this paper.

Christine Brown-Wilson

Papers

Intelligent Carpet System, Based on Photonic Guided-Path Tomography, for Gait and Balance Monitoring in Home Environments