Detection and Classification of Human Movements in Video Scenes
A novel approach for the detection and classification of human movements in videos scenes is presented in this paper. It consists in detecting, segmenting and tracking foreground objects in video scenes to further classify their movements as conventional
- PDF / 1,327,613 Bytes
- 14 Pages / 430 x 660 pts Page_size
- 64 Downloads / 223 Views
ract. A novel approach for the detection and classification of human movements in videos scenes is presented in this paper. It consists in detecting, segmenting and tracking foreground objects in video scenes to further classify their movements as conventional or non-conventional. From each tracked object in the scene, features such as position, speed, changes in direction and temporal consistency of the bounding box dimension are extracted. These features make up feature vectors that are stored together with labels that categorize the movement and which are assigned by human supervisors. At the classification step, an instancebased learning algorithm is used to classify the object movement as conventional or non-conventional. For this aim, feature vectors computed from objects in motion are matched against reference feature vectors previously labeled. Experimental results on video clips from two different databases (Parking Lot and CAVIAR) have shown that the proposed approach is able to detect non-conventional human movements in video scenes with accuracies between 77% and 82%. Keywords: Human Movement Classification, Computer Vision, Security.
1
Introduction
The classification of events in video scenes is a relative new research area in computer science and it has been growing more and more due to the broad applicability in real-life. One of the main reasons is the growing interest and use of video-based security systems, known as CCTV. However, the majority of the CCTV systems currently available in the market have limited functionality which comprises capture, storing and visualization of video gathered from one or more cameras. Some CCTV systems already include motion detection algorithms and are able to constrain the recording of videos only when variations in the scene foreground are detected. The main utility of such systems is the recording of conventional and non-conventional events for further consultation and analysis. D. Mery and L. Rueda (Eds.): PSIVT 2007, LNCS 4872, pp. 678–691, 2007. c Springer-Verlag Berlin Heidelberg 2007
Detection and Classification of Human Movements in Video Scenes
679
In other words, such systems do not have any embedded intelligence which is able to provide a classification of the events. They do not have mechanisms to warn operators when non-conventional events are occurring. Such an attribute would be very helpful to prevent and detect in an active fashion the occurrence of non-conventional events. Besides the need of a more efficient tool in the security area, the detection of non-conventional events in video scenes could be used in other contexts, such as: to detect when an elderly people has an accident inside his/her house [1,2], non-conventional activities in an office, transit infractions [3]. Therefore, a nonconventional event can be viewed as an action that does not belong to the context. The research in this area has been focused on two main streams: state-space modeling and template matching [4]. In the former, most of the approaches employ Markov Process and state
Data Loading...