This paper presents a framework based on multi-layer bi-LSTM network (bidirectional Long Short-Term Memory) for multimodal sensor fusion to sense and classify daily activities’ patterns and high-risk events such as falls. The data collected in this work are continuous activity streams from FMCW radar and three wearable inertial sensors on the wrist, waist, and ankle. Each activity has a variable duration in the data stream so that the transitions between activities can happen at random times within the stream, without resorting to conventional fixed-duration snapshots. The proposed bi-LSTM implements soft feature fusion between wearable sensors and radar data, as well as two robust hard-fusion methods using the confusion matrices of both sensors. A novel hybrid fusion scheme is then proposed to combine soft and hard fusion to push the classification performances to approximately 96% accuracy in identifying continuous activities and fall events. These fusion schemes implemented with the proposed bi-LSTM network are compared with conventional sliding window approach, and all are validated with realistic “leaving one participant out” (L1PO) method (i.e. testing subjects unknown to the classifier). The developed hybrid-fusion approach is capable of stabilizing the classification performance among different participants in terms of reducing accuracy variance of up to 18.1% and increasing minimum, worst-case accuracy up to 16.2%.

Recognition of human movements with radar for ambient activity monitoring is a developed area of research that yet presents outstanding challenges to address. In real environments, activities and movements are performed with seamless motion, with continuous transitions between activities of different duration and a large range of dynamic motions, compared with discrete activities of fixed-time lengths which are typically analysed in the literature. This paper proposes a novel approach based on recurrent LSTM and Bi-LSTM network architectures for continuous activity monitoring and classification. This approach uses radar data in the form of a continuous temporal sequence of micro-Doppler or range-time information, differently from from other conventional approaches based on convolutional networks that interpret the radar data as images. Experimental radar data involving 15 participants and different sequences of 6 actions are used to validate the proposed approach. It is demonstrated that using the Dopplerdomain data together with the Bi-LSTM network and an optimal learning rate can achieve over 90% mean accuracy, whereas range-domain data only achieved approximately 76%. The details of the network architectures, insights in their behaviour as a function of key hyper-parameters such as the learning rate, and a discussion on their performance across are provided in the paper.

This paper presents different information fusion approaches to classify human gait patterns and falls in a radar sensors network. The human gaits classified in this work are both individual and sequential, continuous gait collected by a FMCW radar and three UWB pulse radar placed at different spatial locations. Sequential gaits are those containing multiple gait styles performed one after the other, with natural transitions in between, including fall events developing from walking gait in some cases. The proposed information fusion approaches operate at signal and decision level. For the signal level combination, a simple trilateration algorithm is implemented on the range data from the 3 UWB radar sensors, achieving good classification results with the proposed Bi-LSTM (Bidirectional LSTM neural network) as classifier, without exploiting conventional micro-Doppler information. For the decision level fusion, the classification results of individual radars using the Bi-LSTM network are combined with a robust Naive Bayes Combiner (NBC), and this showed subsequent improvement compared to the single radar case thanks to multi-perspective views of the subjects. Compared to conventional SVM and Random Forest classifiers, the proposed approach yields +20% and +17% improvement in the classification accuracy of individual gaits for the range-only trilateration method and NBC decision fusion method, respectively. When classifying sequential gaits, the overall accuracy for the two proposed methods reaches 93% and 90%, with validation via a ’leaving one participant out’ approach to test the robustness with subjects unknown to the network.

Growing life expectancy and increasing incidence of multiple chronic health conditions are significant societal challenges. Different technologies have been proposed to address these issues, to detect critical events such as stroke or falls, and to monitor automatically human activities for health condition inference and anomalies detection. This paper aims to investigate two types of sensing technologies proposed for assisted living: wearable and radar sensors. First, different feature selection methods are validated and compared in terms of accuracy and computational loads. Then, information fusion is applied to enhance activity classification accuracy combining the two sensors. Improvements in classification accuracy of approximately 12% using feature level fusion is achieved with both Support Vector Machine and K Nearest Neighbor classifiers. Decision-level fusion schemes are also investigated, yielding classification accuracy in the order of 97-98%.

With the increased life expectancy and rise in health conditions related to aging, there is a need for new technologies that can routinely monitor vulnerable people, identify their daily pattern of activities and any anomaly or critical events such as falls. This paper aims to evaluate magnetic and radar sensors as suitable technologies for remote health monitoring purpose, both individually and fusing their information. After experiments and collecting data from 20 volunteers, numerical features has been extracted in both time and frequency domains. In order to analyse and verify the validation of fusion method for different classifiers, a Support Vector Machine with a quadratic kernel, and an Artificial Neural Network with one and multiple hidden layers have been implemented. Furthermore, for both classifiers, feature selection has been performed to obtain salient features. Using this technique along with fusion, both classifiers can detect 10 different activities with an accuracy rate of approximately 96%. In cases where the user is unknown to the classifier, an accuracy of approximately 92% is maintained.

This paper presents a hierarchical sensor fusion approach for human micro-gesture recognition by combining an Ultra Wide Band (UWB) Doppler radar and wearable pressure sensors. First, the wrist-worn pressure sensor array (PSA) and Doppler radar are used to respectively identify static and dynamic gestures through a Quadratic-kernel SVM (Support Vector Machine) classifier. Then, a robust wrapper method is applied on the features from both sensors to search the optimal combination. Subsequently, two hierarchical approaches where one sensor acts as “enhancer” of the other are explored. In the first case, scores from Doppler radar related to the confidence level of its classifier and the prediction label corresponding to the posterior probabilities are utilized to maximize the static hand gestures classification performance by hierarchical combination with PSA data. In the second case, the PSA acts as an “enhancer” for radar to improve the dynamic gesture recognition. In this regard, different weights of the “enhancer” sensor in the fusion process have been evaluated and compared in terms of classification accuracy. A realistic cross-validation method is chosen to test one unknown participant with the model trained by data from others, demonstrating that this hierarchical fusion approach for static and dynamic gestures yields approximately 15% improvement in classification accuracy in the best cases.