A 3D Human Posture Approach for Activity Recognition Based on Depth Camera
Human activity recognition plays an important role in the context of Ambient Assisted Living (AAL), providing useful tools to improve people quality of life. This work presents an activity recognition algorithm based on the extraction of skeleton joints f
- PDF / 392,779 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 90 Downloads / 191 Views
Abstract. Human activity recognition plays an important role in the context of Ambient Assisted Living (AAL), providing useful tools to improve people quality of life. This work presents an activity recognition algorithm based on the extraction of skeleton joints from a depth camera. The system describes an activity using a set of few and basic postures extracted by means of the X-means clustering algorithm. A multi-class Support Vector Machine, trained with the Sequential Minimal Optimization is employed to perform the classification. The system is evaluated on two public datasets for activity recognition which have different skeleton models, the CAD-60 with 15 joints and the TST with 25 joints. The proposed approach achieves precision/recall performances of 99.8 % on CAD-60 and 97.2 %/91.7 % on TST. The results are promising for an applied use in the context of AAL. Keywords: Activity monitoring systems · Human activity recognition · Depth camera · RGB-D camera · Ambient Assisted Living · Assistive technologies
1
Introduction
Human activity recognition is one of the most important areas of computer vision research today. It can be described as the spatiotemporal evolutions of different body postures and its main goal is to automatically detect human activities analyzing data from various types of devices (e.g. color cameras or range sensors). Regarding the assistive technologies, the possibility of application is really wide, in particular, they can include surveillance and monitoring systems, and a large range of applications involving human-machine interactions [1]. Although the recognition of human actions is very important for many real applications, it is still a challenging problem. In the past, the research has mainly focused on recognizing activities from video sequences by means of color cameras. However, capturing articulated human motion from monocular video sensors results in a considerable loss of information [2]. These solutions are often constrained in terms of computational efficiency and robustness to illumination changes [3]. Another approach is to use 3D data from marker-based motion capture or stereo camera systems, i.e. capturing 2D image sequences from multiple views to reconstruct 3D information [4]. Nowadays, the use of depth cameras has become very c Springer International Publishing Switzerland 2016 G. Hua and H. J´ egou (Eds.): ECCV 2016 Workshops, Part II, LNCS 9914, pp. 432–447, 2016. DOI: 10.1007/978-3-319-48881-3 30
A 3D Human Posture Approach
433
popular, because the technological progress has made available devices that are cost effective providing also 3D data at suitable resolution rate. Recently, this kind of sensors has led several new works on activity recognition from depth data [5,6]. These inexpensive devices, such as Microsoft Kinect or Asus Xtion, allow capturing both color and depth information. Moreover, specific tracker software can efficiently detect the human skeleton directly from the depth maps [7]. These features can be exploited in order to develop effective soluti
Data Loading...
