Neural Networks for Automatic Posture Recognition in Ambient-Assisted Living

Abstract

Human Action Recognition (HAR) is a rapidly evolving field impacting numerous domains, among which is Ambient Assisted Living (AAL). In such a context, the aim of HAR is meeting the needs of frail individuals, whether elderly and/or disabled and promoting autonomous, safe and secure living. To this goal, we propose a monitoring system detecting dangerous situations by classifying human postures through Artificial Intelligence (AI) solutions. The developed algorithm works on a set of features computed from the skeleton data provided by four Kinect One systems simultaneously recording the scene from different angles and identifying the posture of the subject in an ecological context within each recorded frame. Here, we compare the recognition abilities of Multi-Layer Perceptron (MLP) and Long-Short Term Memory (LSTM) Sequence networks. Starting from the set of previously selected features we performed a further feature selection based on an SVM algorithm for the optimization of the MLP network and used a genetic algorithm for selecting the features for the LSTM sequence model. We then optimized the architecture and hyperparameters of both models before comparing their performances. The best MLP model (3 hidden layers and a Softmax output layer) achieved 78.4%, while the best LSTM (2 bidirectional LSTM layers, 2 dropout and a fully connected layer) reached 85.7%. The analysis of the performances on individual classes highlights the better suitability of the LSTM approach.

SMGR-BS: Stacking Multiple Gated Recurrent Butterfly Search Model-Based Innovative AAL for Aging and Disabled Individuals

“…A key contribution of AAL is to increase the survival rate of the elderly in selected environments through personalized health-monitoring devices, communication technologies, and information. It encourages research into more flexible lifestyles that evolve into creative ways of aging and looking at how care is given (Guerra et al, 2022). This can be thought-provoking research, and searching for multiple ADLs and self-categorization can be a huge obstacle.…”

Recurrent Network Solutions for Human Posture Recognition Based on Kinect Skeletal Data

“…Different feature selection methods, as well as different machine and deep learning architectures, were proposed for the classification block in previous studies by our group [28,29]. The most promising solution was proposed in Guerra et al [30], where a genetic algorithm was applied to select eight kinematic features and a sequence-to-sequence model was trained to identify five classes. Three classes correspond to the three postures frequently adopted by a person during daily activities: standing, sitting, and lying down; one represents an unconventional daily posture, labeled "dangerous-sitting", and groups all postures that in some way manifest a malaise or fainting, such as a seated person slumped or lying backward; and the last class groups all the transitions between two consecutive postures (for example, between sitting and lying postures and vice-versa).…”

“…Therefore, for the efficiency of the home monitoring system, the specificity of the classifying model for such a class is extremely important for reducing the number of false negatives. The model, characterized by two Bidirectional Long Short-Term Memory layers, alternated by two dropout layers and, as last layer, a fully connected layer (2BLSTM2D), reached an overall accuracy of 85.7% and a percentage of about 85% and 95% regarding the specificity and sensitivity metrics of the dangerous-sitting posture [30]. Here, aiming to take advantage of the temporal dependency of the inputs to further improve the accuracy of the classification block and, in particular, its specificity for the dangerous-sitting class, a new deep RNN architecture based on GRU networks with a sequence-to-last approach was trained and tested to identify the five classes described above.…”

High Speed and Accuracy of Animation 3D Pose Recognition Based on an Improved Deep Convolution Neural Network

“…Feedforward constraints facilitate implementation, optimization, and learning. Therefore, feedforward neural networks are among the most mature and widely used neural networks [24][25][26][27]. However, when processing image data, traditional feedforward neural networks require a large number of neurons to read the pixel information of the input image, making them unsuitable for processing high-pixel images.…”