Evaluation of Inertial Sensor-Based Pre-Impact Fall Detection Algorithms Using Public Dataset

TinyFallNet: A Lightweight Pre-Impact Fall Detection Model

Falls represent a significant health concern for the elderly. While studies on deep learning-based preimpact fall detection have been conducted to mitigate fall-related injuries, additional efforts are needed for embedding in microcomputer units (MCUs). In this study, ConvLSTM, the state-of-the-art model, was benchmarked, and we attempted to lightweight it by leveraging features from image-classification models VGGNet and ResNet while maintaining performance for wearable airbags. The models were developed and evaluated using data from young subjects in the KFall public dataset based on an inertial measurement unit (IMU), leading to the proposal of TinyFallNet based on ResNet. Despite exhibiting higher accuracy (97.37% < 98.00%) than the benchmarked ConvLSTM, the proposed model requires lower memory (1.58 MB > 0.70 MB). Additionally, data on the elderly from the fall data of the FARSEEING dataset and activities of daily living (ADLs) data of the KFall dataset were analyzed for algorithm validation. This study demonstrated the applicability of image-classification models to preimpact fall detection using IMU and showed that additional tuning for lightweighting is possible due to the different data types. This research is expected to contribute to the lightweighting of deep learning models based on IMU and the development of applications based on IMU data.

Comparison of four machine learning algorithms for a pre-impact fall detection system

In recent years, real-time health monitoring using wearable sensors has been an active area of research. This paper presents an efficient and low-cost fall detection system based on a pair of shoes equipped with inertial sensors and plantar pressure sensors. In addition, four machine learning algorithms (KNN, SVM, RF, and BP neural network) are compared in terms of their detection performance and suitability for pre-impact fall detection. The results show that KNN and BP neural network outperformed the other two algorithms, where KNN had 98.8% sensitivity, 99.8% specificity, and 99.7% accuracy, and BP neural network had 100% sensitivity, 99.8% specificity, and 99.9% accuracy. KNN outperformed BP neural network in terms of fitting ability, and their lead times were both 460.95 ms. The system can provide sufficient intervention time for the wearer in the pre-impact phase and together with the touchdown fall protection device can effectively prevent fall injuries.

Fall Detection Algorithm Based on Inertial Sensor and Hierarchical Decision

With the aging of the human body and the reduction in its physiological capacities, falls have become a huge threat to individuals' physical and mental health, leading to serious bodily damage to the elderly and financial pressure on their families. As a result, it is vital to design a fall detection algorithm that monitors the state of human activity. This work designs a human fall detection algorithm based on hierarchical decision making. First, this work proposes a dimensionality reduction approach based on feature importance analysis (FIA), which optimizes the feature space via feature importance. This procedure reduces the dimension of features greatly and reduces the time spent by the model in the training phase. Second, this work proposes a hierarchical decision-making algorithm with an XGBoost model. The algorithm is divided into three levels. The first level uses the threshold approach to make a preliminary assessment of the data and only transfers the fall type data to the next level. The second level is an XGBoost-based classification algorithm to analyze again the type of data which remained from the first level. The third level employs a comparison method to determine the direction of the falling. Finally, the fall detection algorithm proposed in this paper has an accuracy of 98.19%, a sensitivity of 97.50%, and a specificity of 98.63%. The classification accuracy of the fall direction reaches 93.44%, and the algorithm can efficiently determine the fall direction.

Gait Trajectory Prediction on an Embedded Microcontroller Using Deep Learning

Achieving a normal gait trajectory for an amputee's active prosthesis is challenging due to its kinematic complexity. Accordingly, lower limb gait trajectory kinematics and gait phase segmentation are essential parameters in controlling an active prosthesis. Recently, the most practiced algorithm in gait trajectory generation is the neural network. Deploying such a complex Artificial Neural Network (ANN) algorithm on an embedded system requires performing the calculations on an external computational device; however, this approach lacks mobility and reliability. In this paper, more simple and reliable ANNs are investigated to be deployed on a single low-cost Microcontroller (MC) and hence provide system mobility. Two neural network configurations were studied: Multi-Layered Perceptron (MLP) and Convolutional Neural Network (CNN); the models were trained on shank and foot IMU data. The data were collected from four subjects and tested on a fifth to predict the trajectory of 200 ms ahead. The prediction was made for two cases: with and without providing the current phase of the gait. Then, the models were deployed on a low-cost microcontroller (ESP32). It was found that with fewer data (excluding the current gait phase), CNN achieved a better correlation coefficient of 0.973 when compared to 0.945 for MLP; when including the current phase, both network configurations achieved better correlation coefficients of nearly 0.98. However, when comparing the execution time required for the prediction on the intended MC, MLP was much faster than CNN, with an execution time of 2.4 ms and 142 ms, respectively. In summary, it was found that when training data are scarce, CNN is more efficient within the acceptable execution time, while MLP achieves relative accuracy with low execution time with enough data.

Smart Wearables with Sensor Fusion for Fall Detection in Firefighting

During the past decade, falling has been one of the top three causes of death amongst firefighters in China. Even though there are many studies on fall-detection systems (FDSs), the majority use a single motion sensor. Furthermore, few existing studies have considered the impact sensor placement and positioning have on fall-detection performance; most are targeted toward fall detection of the elderly. Unfortunately, floor cracks and unstable building structures in the fireground increase the difficulty of detecting the fall of a firefighter. In particular, the movement activities of firefighters are more varied; hence, distinguishing fall-like activities from actual falls is a significant challenge. This study proposed a smart wearable FDS for firefighter fall detection by integrating motion sensors into the firefighter's personal protective clothing on the chest, elbows, wrists, thighs, and ankles. The firefighter's fall activities are detected by the proposed multisensory recurrent neural network, and the performances of different combinations of inertial measurement units (IMUs) on different body parts were also investigated. The results indicated that the sensor fusion of IMUs from all five proposed body parts achieved performances of 94.10%, 92.25%, and 94.59% in accuracy, sensitivity, and specificity, respectively.

A Hierarchical Approach to Activity Recognition and Fall Detection Using Wavelets and Adaptive Pooling

Human activity recognition has been a key study topic in the development of cyber physical systems and assisted living applications. In particular, inertial sensor based systems have become increasingly popular because they do not restrict users’ movement and are also relatively simple to implement compared to other approaches. In this paper, we present a hierarchical classification framework based on wavelets and adaptive pooling for activity recognition and fall detection predicting fall direction and severity. To accomplish this, windowed segments were extracted from each recording of inertial measurements from the SisFall dataset. A combination of wavelet based feature extraction and adaptive pooling was used before a classification framework was applied to determine the output class. Furthermore, tests were performed to determine the best observation window size and the sensor modality to use. Based on the experiments the best window size was found to be 3 s and the best sensor modality was found to be a combination of accelerometer and gyroscope measurements. These were used to perform activity recognition and fall detection with a resulting weighted F1 score of 94.67%. This framework is novel in terms of the approach to the human activity recognition and fall detection problem as it provides a scheme that is computationally less intensive while providing promising results and therefore can contribute to edge deployment of such systems.

Prediction of gait trajectories based on the Long Short Term Memory neural networks

The forecasting of lower limb trajectories can improve the operation of assistive devices and minimise the risk of tripping and balance loss. The aim of this work was to examine four Long Short Term Memory (LSTM) neural network architectures (Vanilla, Stacked, Bidirectional and Autoencoders) in predicting the future trajectories of lower limb kinematics, i.e. Angular Velocity (AV) and Linear Acceleration (LA). Kinematics data of foot, shank and thigh (LA and AV) were collected from 13 male and 3 female participants (28 ± 4 years old, 1.72 ± 0.07 m in height, 66 ± 10 kg in mass) who walked for 10 minutes at preferred walking speed (4.34 ± 0.43 km.h-1) and at an imposed speed (5km.h-1, 15.4% ± 7.6% faster) on a 0% gradient treadmill. The sliding window technique was adopted for training and testing the LSTM models with total kinematics time-series data of 10,500 strides. Results based on leave-one-out cross validation, suggested that the LSTM autoencoders is the top predictor of the lower limb kinematics trajectories (i.e. up to 0.1s). The normalised mean squared error was evaluated on trajectory predictions at each time-step and it obtained 2.82-5.31% for the LSTM autoencoders. The ability to predict future lower limb motions may have a wide range of applications including the design and control of bionics allowing improved human-machine interface and mitigating the risk of falls and balance loss.

Effects of Combined Balance and Strength Training on Measures of Balance and Muscle Strength in Older Women With a History of Falls

OBJECTIVE

We investigated the effects of combined balance and strength training on measures of balance and muscle strength in older women with a history of falls.

METHODS

Twenty-seven older women aged 70.4 ± 4.1 years (age range: 65 to 75 years) were randomly allocated to either an intervention (IG, n = 12) or an active control (CG, n = 15) group. The IG completed 8 weeks combined balance and strength training program with three sessions per week including visual biofeedback using force plates. The CG received physical therapy and gait training at a rehabilitation center. Training volumes were similar between the groups. Pre and post training, tests were applied for the assessment of muscle strength (weight-bearing squat [WBS] by measuring the percentage of body mass borne by each leg at different knee flexions [0°, 30°, 60°, and 90°], sit-to-stand test [STS]), and balance. Balance tests used the modified clinical test of sensory interaction (mCTSIB) with eyes closed (EC) and opened (EO), on stable (firm) and unstable (foam) surfaces as well as spatial parameters of gait such as step width and length (cm) and walking speed (cm/s).

RESULTS

Significant group × time interactions were found for different degrees of knee flexion during WBS (0.0001 < p < 0.013, 0.441 < d < 0.762). Post hoc tests revealed significant pre-to-post improvements for both legs and for all degrees of flexion (0.0001 < p < 0.002, 0.697 < d < 1.875) for IG compared to CG. Significant group × time interactions were found for firm EO, foam EO, firm EC, and foam EC (0.006 < p < 0.029; 0.302 < d < 0.518). Post hoc tests showed significant pre-to-post improvements for both legs and for all degrees of oscillations (0.0001 < p < 0.004, 0.753 < d < 2.097) for IG compared to CG. This study indicates that combined balance and strength training improved percentage distribution of body weight between legs at different conditions of knee flexion (0°, 30°, 60°, and 90°) and also decreased the sway oscillation on a firm surface with eyes closed, and on foam surface (with eyes opened or closed) in the IG.

CONCLUSION

The higher positive effects of training seen in standing balance tests, compared with dynamic tests, suggests that balance training exercises including lateral, forward, and backward exercises improved static balance to a greater extent in older women.

A Smart Robotic Walker With Intelligent Close-Proximity Interaction Capabilities for Elderly Mobility Safety

The elderly population has rapidly increased in past years, bringing huge demands for elderly serving devices, especially for those with mobility impairment. Present assistant walkers designed for elderly users are primitive with limited user interactivity and intelligence. We propose a novel smart robotic walker that targets a convenient-to-use indoor walking aid for the elderly. The walker supports multiple modes of interactions through voice, gait or haptic touch, and allows intelligent control via learning-based methods to achieve mobility safety. Our design enables a flexible, initiative and reliable walker due to the following: (1) we take a hybrid approach by combining the conventional mobile robotic platform with the existing rollator design, to achieve a novel robotic system that fulfills expected functionalities; (2) our walker tracks users in front by detecting lower limb gait, while providing close-proximity walking safety support; (3) our walker can detect human intentions and predict emergency events, e.g., falling, by monitoring force pressure on a specially designed soft-robotic interface on the handle; (4) our walker performs reinforcement learning-based sound source localization to locate and navigate to the user based on his/her voice signals. Experiment results demonstrate the sturdy mechanical structure, the reliability of multiple novel interactions, and the efficiency of the intelligent control algorithms implemented. The demonstration video is available at: https://sites.google.com/view/smart-walker-hku.

Pre-Impact Fall Detection with CNN-Based Class Activation Mapping Method

In this paper, we report our improvement on the prediction accuracy of pre-impact fall detection by applying a learning-based method on the real-time data from an IMU (inertial measurement unit)-sensor mounted on the waist, making it possible to achieve a high accuracy on a wearable device with the extracted features. Using the fixed threshold method is difficult for achieving satisfactory detection accuracy, due to various characteristics and behaviors in the movement of different individuals. In contrast, one could realize high-accuracy detection with machine learning-based methods, but it is difficult to apply them in the wearable devices due to the high hardware requirement. Our method merges the two methods above. We build a convolutional neural network (CNN) with a class activation mapping (CAM) method, which could highlight the class-specific region in the data and obtain a hot map of the fall data. After training on the MobiAct dataset, the model could achieve high-accuracy detection (95.55%) and obtain the region with high contributions to the classification. Then, we manually extract effective features and characteristics of this region and form our special threshold method, achieving pre-impact fall detection in real-world data. Consequently, our method achieves accuracy of 95.33% and a detection time of within 400 ms.

Enhanced Algorithm for the Detection of Preimpact Fall for Wearable Airbags

Fall-related injury is a common cause of mortality among the elderly. Hip fractures are especially dangerous and can even be fatal. In this study, a threshold-based preimpact fall detection algorithm was developed for wearable airbags that minimize the impact of falls on the user's body. Acceleration sum vector magnitude (SVM), angular velocity SVM, and vertical angle, calculated using inertial data captured from an inertial measurement unit were used to develop the algorithm. To calculate the vertical angle accurately, a complementary filter with a proportional integral controller was used to minimize integration errors and the effect of external impacts. In total, 30 healthy young men were recruited to simulate 6 types of falls and 14 activities of daily life. The developed algorithm achieved 100% sensitivity, 97.54% specificity, 98.33% accuracy, and an average lead time (i.e., the time between the fall detection and the collision) of 280.25 ± 10.29 ms with our experimental data, whereas it achieved 96.1% sensitivity, 90.5% specificity, and 92.4% accuracy with the SisFall public dataset. This paper demonstrates that the algorithm achieved a high accuracy using our experimental data, which included some highly dynamic motions that had not been tested previously.

A Novel Hybrid Deep Neural Network to Predict Pre-impact Fall for Older People Based on Wearable Inertial Sensors

Falls in the elderly is a major public health concern due to its high prevalence, serious consequences and heavy burden on the society. Many falls in older people happen within a very short time, which makes it difficult to predict a fall before it occurs and then to provide protection for the person who is falling. The primary objective of this study was to develop deep neural networks for predicting a fall during its initiation and descending but before the body impacts to the ground so that a safety mechanism can be enabled to prevent fall-related injuries. We divided the falling process into three stages (non-fall, pre-impact fall and fall) and developed deep neutral networks to perform three-class classification. Three deep learning models, convolutional neural network (CNN), long short term memory (LSTM), and a novel hybrid model integrating both convolution and long short term memory (ConvLSTM) were proposed and evaluated on a large public dataset of various falls and activities of daily living (ADL) acquired with wearable inertial sensors (accelerometer and gyroscope). Fivefold cross validation results showed that the hybrid ConvLSTM model had mean sensitivities of 93.15, 93.78, and 96.00% for non-fall, pre-impact fall and fall, respectively, which were higher than both LSTM (except the fall class) and CNN models. ConvLSTM model also showed higher specificities for all three classes (96.59, 94.49, and 98.69%) than LSTM and CNN models. In addition, latency test on a microcontroller unit showed that ConvLSTM model had a short latency of 1.06 ms, which was much lower than LSTM model (3.15 ms) and comparable with CNN model (0.77 ms). High prediction accuracy (especially for pre-impact fall) and low latency on the microboard indicated that the proposed hybrid ConvLSTM model outperformed both LSTM and CNN models. These findings suggest that our proposed novel hybrid ConvLSTM model has great potential to be embedded into wearable inertial sensor-based systems to predict pre-impact fall in real-time so that protective devices could be triggered in time to prevent fall-related injuries for older people.