1 3 Artif Life Robotics DOI 10.1007/s10015-014-0195-4 ORIGINAL ARTICLE Temporal analysis for fast motion detection in a crowd Panca Mudjirahardjo · Joo Kooi Tan · Hyoungseop Kim · Seiji Ishikawa Received: 8 April 2014 / Accepted: 4 November 2014 © ISAROB 2015 1 Introduction In recent years, abnormal motion detection has attracted great research attention in computer vision. Most of cur- rent surveillance systems only provide reactive security by enabling the analysis of events after the event has already occurred. What is really needed by the security commu- nity is proactive security to help prevent future attacks. The abnormal motion is a motion which differs from many people’s motion. In this research, running is considered as abnormal motion, while walking is normal motion. Many approaches on video event analysis are based on the object trajectories extracted from the video [1–7]. The abnormal events can be detected through a prior learning of normal events [2, 3] or without a learning process by ana- lyzing the trajectory result directly [1, 4–7]. An abnormal event is categorized into two classes: a global abnormal event (GAE) and a local abnormal event (LAE) [8]. The abnormality of an image is judged in the former, whereas that of a local object is judged and localized in the latter. According to this categorization, [2, 3, 7] deal with GAE, whereas [1, 4–6] concentrate in LAE. This paper deals with LAE under the cluttered (cloud) background. Jiang et al. [1] used spatial and temporal context and performed frequency-based analysis to detect anomalous video events. The normal observation is modeled by a hid- den Markov model. This research detected the anomalous car trajectory on the road from top view. Kiryati et al. [2] recognized an abnormal human behavior from high camera view. Before the detection phase, they included the training phase for normal condition. Baranwal et al. [3] detected an abnormal indoor motion in a static background environment. They trained various motions using radial basis functions networks. Park et al. [4] used clustering of motion based on similarity measurement of a feature space. They detected an Abstract We present a fast motion detection technique in a crowd as an abnormal motion based on optical flow and a motion history image (MHI). Since a camera view is usu- ally not in perpendicular with motion direction, the velocity of motion is not uniform spatially. Instead of object detec- tion directly from an image, we separate an image into several blocks. In this paper, we propose a novel method to analyze a motion using MHI representation, called a shift space and a shift histogram. Together with a velocity histogram, the method can detect fast motion in a crowd, realizing local abnormal event detection. The performance of the proposed method is experimentally illustrated and evaluated. Keywords Abnormal motion · Lucas–Kanade tracker · Motion history image · Velocity histogram · Shift histogram This work was presented in part at the 19th International Symposium on Artificial Life and Robotics, Beppu, Oita, January 22–24, 2014. P. Mudjirahardjo (*) · J. K. Tan · H. Kim · S. Ishikawa Department of Control Engineering, Kyushu Institute of Technology, 1-1 Sensuichou, Tobata-ku, Kitakyushu, Fukuoka 804-8550, Japan e-mail: panca@ss10.cntl.kyutech.ac.jp J. K. Tan e-mail: etheltan@ss10.cntl.kyutech.ac.jp H. Kim e-mail: kimhs@cntl.kyutech.ac.jp S. Ishikawa e-mail: ishikawa@ss10.cntl.kyutech.ac.jp