Asymmetric Residual Neural Network for Accurate Human Activity Recognition

Abstract

Human Activity Recognition (HAR) using deep neural network has become a hot topic in human-computer interaction. Machine can effectively identify human naturalistic activities by learning from a large collection of sensor data. Activity recognition is not only an interesting research problem, but also has many real-world practical applications. Based on the success of residual networks in achieving a high level of aesthetic representation of the automatic learning, we propose a novel Asymmetric Residual Network, named ARN. ARN is implemented using two identical path frameworks consisting of (1) a short time window, which is used to capture spatial features, and (2) a long time window, which is used to capture fine temporal features. The long time window path can be made very lightweight by reducing its channel capacity, yet still being able to learn useful temporal representations for activity recognition. In this paper, we mainly focus on proposing a new model to improve the accuracy of HAR. In order to demonstrate the effectiveness of ARN model, we carried out extensive experiments on benchmark datasets (i.e., OPPORTUNITY, UniMiB-SHAR) and compared with some conventional and state-of-the-art learning-based methods. Then, we discuss the influence of networks parameters on performance to provide insights about its optimization. Results from our experiments show that ARN is effective in recognizing human activities via wearable datasets.

A Comparative Study: Toward an Effective Convolutional Neural Network Architecture for Sensor-Based Human Activity Recognition

“…Dong et al [13] proposed an inception module combined with an HCF. Long et al [14] proposed methods to learn large-and small-scale networks separately, and finally connected them. The core of the methods is the introduction of two sizes of residual blocks.…”

Easy Ensemble: Simple Deep Ensemble Learning for Sensor-Based Human Activity Recognition

“…Group convolution, as defined, divides the input channel into N equal groups and performs an operation to convolute independent parameters into each group. Therefore, Z is equivalent to the output of combining Z (1) , Z (2) , Z (3) through N individual convolution layers.…”

“…In group normalization, combined input tensor X = [X (1) , X (2) , • • • , X (N ) ] (the shape of each X (i) is as same as X) is normalized to output Z = [Z (1) , Z (2) , • • • , Z (N ) ] as follows:…”

“…It is difficult for humans to interpret the context of raw sensor values and human behavior differs among individuals; therefore, automatic feature extraction by deep representation learning is used to build HAR models. To develop an accurate HAR model, researchers have investigated various model architectures [1,2,3] using deep learning in the image recognition field.…”

A Comparative Performance Analysis of Computational Intelligence Techniques to Solve the Asymmetric Travelling Salesman Problem

“…A critical review of literature on the Travelling Salesman Problem reveals that there are lots of studies on the symmetric Travelling Salesman Problem over the past several decades. However, it is rather ridiculous that there exists a paucity of literature on the asymmetric TSP [ 50 ]. This is rather puzzling because most day-to-day human activities are, indeed, asymmetric.…”