Feature extraction for robust physical activity recognition

Abstract

IntroductionWith information obtained from sensors, computer based system can make more intelligent actions by adapting their behavior to the context conditions. These days, thanks to the development of multi-sensor networks, related research areas have increased rapidly. Among those areas, Human Activity Recognition (HAR) based on wearable sensors (accelerometer, gyroscope, magnetometer, etc.) has recently received lots of attention due to its large number of promising applications. One of the most interesting HAR applications is ubiquitous identification of physical activity. As we know, over-weighting is a general human problem as a result of a physical inactivity habit. A Lancet publication [1] estimates that physical inactivity causes 9% of all premature deaths worldwide. By monitoring physical activity, we can help people to learn the calories they consumed or gained during the day in a much more precise way and Abstract This paper presents the development of a Human Activity Recognition (HAR) system that uses a network of nine inertial measurement units situated in different body parts. Every unit provides 3D (3-dimension) acceleration, 3D angular velocity, 3D magnetic field orientation, and 4D quaternions. This system identifies 33 different physical activities (walking, running, cycling, lateral elevation of arms, etc.). The system is composed of two main modules: a feature extractor for obtaining the most relevant characteristics from the inertial signals every second, and a machine learning algorithm for classifying between the different activities. This paper focuses on the feature extractor module, evaluating several types of features and proposing different normalization approaches. This paper also analyses the performance of every sensor included in the inertial measurement units. The main experiments have been done using a public available dataset named REALDISP Activity Recognition dataset. This dataset includes recordings from 17 subjects performing 33 different activities in three different scenarios. Final results demonstrate that the proposed HAR system significantly improves the classification accuracy compared to previous works on this dataset. For the best configuration, the system accuracy is 99.1%. This system has been also evaluated with the OPPORTUNITY dataset obtaining competitive results.

A Multisensory Approach for Remote Health Monitoring of Older People

“…In terms of data processing, additional features can be added to the pool of those considered (e.g. "jerk" [41] or wavelet-based [19] features for wearable sensors, or additional representation domains for the radar data [42]), and additional feature selection methods and metrics for information fusion investigated. The application of deep learning methods may also be considered, in particular the challenge of using deep networks with small amount of experimental data available, for example through transfer learning approaches or through the generation of suitable simulation data.…”

Advanced Machine Learning for Gesture Learning and Recognition Based on Intelligent Big Data of Heterogeneous Sensors

“…Moreover, active developments in gesture recognition techniques have led to the recent developments in virtual reality (VR) technology [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16]. The gesture recognition accuracy, which is a key metric to evaluate the effectiveness of gesture recognition systems, has witnessed steady improvements using input data from multiple sensors [17][18][19][20][21][22]. Moreover, researchers have studied visual recognition and learning systems for more intuitive learning of gesture recognition.…”

“…First, in the gesture registration stage, the proposed method obtains gesture data from the multi-sensor fusion. Using two or more sensors, various types of gestures can be learnt and precise and accurate gesture recognition can be realized [17][18][19][20][21][22]. Moreover, the proposed method allows the end user to limit the range of gestures to be registered for recognizing specific body parts, thereby improving the recognition accuracy.…”

“…Table 2 represents available joints acquired by each sensor. 11,4 ), Thumb proximal (s 11,5 ), Thumb metacarpal (s 11,6 ) Index distal (s 11,7 ), Index intermediate (s 11,8 ), Index proximal (s 11,9 ), Index metacarpal (s 11,10 ) Middle distal (s 11,11 ), Middle intermediate (s 11,12 ), Middle proximal (s 11,13 ), Middle metacarpal (s 11,14 ) Ring distal (s 11,15 ), Ring intermediate (s 11,16 ), Ring proximal (s 11,17 ), Ring metacarpal (s 11,18 ) Pinky distal (s 11,19 ), Pinky intermediate (s 11,20 ), Pinky proximal (s 11,21 When the gesture recording module records and stores a gesture, the gesture at time t, g t , includes B * t and S t at time t, g t = B * t , S t . Although the proposed method stores S t , S * t ⊂ S t ⊂S, during the gesture registration stage, the learning model learns and recognizes only S * t during the gesture learning and recognition stage in order to reduce the processing time required.…”

Human Action Recognition Using Deep Learning Methods on Limited Sensory Data

“…However, this is a classification problem and we used supervised learning methods. The UCI HAR dataset which is used as benchmark dataset is also used in several studies [14] [15]. In these works, mostly classical machine learning algorithms such as Support Vector Machines (SVM) [16], k Nearest Neighbour, Linear Discriminant Analysis (LDA) [17], Multilayer Perceptron(MLP) [18] and deep learning methods are used on this dataset.…”