Postural classification using Kinect

Abstract

This research focuses on the comparison of posture recognition, using a data mining classification approach on the skeleton data stream obtained from Kinect camera. We classified four standard postures including Stand, Sit, Sit on floor and Lie Down. We compared six classifiers, namely, decision tree, neural network, naïve Bayes, support vector machine, logistic regression and random forest in order to find a suitable classifier. Our best results can correctly classify the postures with 97.88% accuracy, 97.40% sensitivity, and 0.991 ROC area under curve using Max-Min normalization with a decision tree classifier on four transformed attributes. Our future work will use the knowledge obtained to classify a wider range of postures of the elderly while watching television, to be a part of a bigger effort to monitor and study elderly behavior at home.

Automatic Pose Recognition for Monitoring Dangerous Situations in Ambient-Assisted Living

“…Using properly located cameras, the inhabitant can be recorded while free to perform the normal actions of daily life without limitations and without having to be in anyway involved, e.g., having to remember to wear a device or to charge it. The cameras used for AAL purposes are commonly depth cameras, such as Asus Xtion (Taipei, Taiwan), Intel RealSense (Santa Clara, United States), Orbbec Astra (Troy, United States) and Microsoft Kinect (Redmond, United States) (Ben Hadj Mohamed et al, 2013;Han et al, 2013;Gasparrini et al, 2014;Mastorakis and Makris, 2014;Pannurat et al, 2014;Visutarrom et al, 2014Visutarrom et al, , 2015Damaševičius et al, 2016;Calin and Coroiu, 2018). Thanks to many approaches based on RGB sequences, depth images or their combination, these sensors are able to provide detailed information about 3D human motion (Wang et al, 2014;Kim et al, 2017).…”

“…This makes ML particularly suitable to analyze data coming from smart house sensors in order to recognize falls or to detect a dangerous situation during daily life activities. Machine learning algorithms such as Naïve Bayes classifiers (NBC), K-nearest neighbor (KNN), support vector machines (SVM), hidden Markov models (HMM), and artificial neural networks (ANN), random forest (RF), decision tree (DT), and logistic regression (LR) (Begg and Hassan, 2006;Crandall and Cook, 2010;Hussein et al, 2014;Visutarrom et al, 2014;Wang et al, 2014;Amiribesheli et al, 2015;Jalal et al, 2015) are the most popular algorithms used in sensor-and vision-based activity recognition. K-nearest neighbor is widely used in reallife scenarios since it is non-parametric, meaning that it does not make any assumptions about the underlying distribution of the data.…”

“…The best classifier was found to be the MLP network (100% of accuracy vs. 99.75% of SVM,93.19% of DT and 81.94% of NBC). Visutarrom et al (2014) went deeper into this topic comparing six different ML classifiers and two different sets of features (geometric vs. skeletal joints features). Four poses (standing, sitting, sitting on floor, and lying down) of a subject watching television in front of the Kinect device were classified.…”

“…In visionbased action recognition, the common approach is to extract image features from video data and to issue a corresponding action class label (Poppe, 2010;Babiker et al, 2018). Nevertheless, when skeleton representation of the human body is used, the most privileged discriminative features are the raw data coming from the skeletal tracking (joint spatial coordinates) (Patsadu et al, 2012;Youness and Abdelhak, 2016) or some indices expressing geometric relations between certain body points, such as: the vertical distance from hip joint to room floor (Visutarrom et al, 2014(Visutarrom et al, , 2015, the distance between the right toe and the plane spanned by the left ankle, the left hip and the foot for a fixed pose (Müller et al, 2005) the distance between two joints, two body segments, or a joint and a body segment (Yang and Tian, 2014), the relative angle between two segments within the body kinematic chain (Müller et al, 2005) and finally, the size of the 3D bounding box enclosing the body skeleton (Bevilacqua et al, 2014). Geometric features are synthetic in the sense that they express a single geometric aspect making them particularly robust to spatial variations that are not correlated with the aspect of interest (Müller et al, 2005).…”

A Hybrid Approach to Human Posture Classification During TV Watching

“…Unlike most other researchers, we use Kinect to classify human postures while watching television. In an earlier work, a system for simple postural detection and classification of elderly people while watching television was developed 12 . The experiments included four standard postures of stand, sit, sit on floor and lie down.…”