Globally Continuous and Non-Markovian Crowd Activity Analysis from Videos
Automatically recognizing activities in video is a classic problem in vision and helps to understand behaviors, describe scenes and detect anomalies. We propose an unsupervised method for such purposes. Given video data, we discover recurring activity pat
- PDF / 1,881,016 Bytes
- 18 Pages / 439.37 x 666.142 pts Page_size
- 23 Downloads / 202 Views
Disney Research Los Angeles, Glendale, USA 2 University of Leeds, Leeds, UK [email protected] 3 Trinity College Dublin, Dublin, Ireland [email protected]
Abstract. Automatically recognizing activities in video is a classic problem in vision and helps to understand behaviors, describe scenes and detect anomalies. We propose an unsupervised method for such purposes. Given video data, we discover recurring activity patterns that appear, peak, wane and disappear over time. By using non-parametric Bayesian methods, we learn coupled spatial and temporal patterns with minimum prior knowledge. To model the temporal changes of patterns, previous works compute Markovian progressions or locally continuous motifs whereas we model time in a globally continuous and non-Markovian way. Visually, the patterns depict flows of major activities. Temporally, each pattern has its own unique appearance-disappearance cycles. To compute compact pattern representations, we also propose a hybrid sampling method. By combining these patterns with detailed environment information, we interpret the semantics of activities and report anomalies. Also, our method fits data better and detects anomalies that were difficult to detect previously.
1
Introduction
Understanding crowd activities from videos has been a goal in many areas [1]. In computer vision, a number of subtopics have been studied extensively, including flow estimation [2], behavior tracking [3] and activity detection [4,5]. The main problem is essentially mining recurrent patterns over time from video data. In this work, we are particularly interested in mining recurrent spatio-temporal activity patterns, i.e., recurrent motions such as pedestrians walking or cars driving. Discovering these patterns can be useful for applications such as scene H. Wang—ORCID ID:orcid.org/0000-0002-2281-5679. This work is mostly done by the authors when they were with Disney Research Los Angeles. Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46454-1 32) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part V, LNCS 9909, pp. 527–544, 2016. DOI: 10.1007/978-3-319-46454-1 32
528
H. Wang and C. O’Sullivan
summarization, event counting or unusual activity detection. On a higher level, such patterns could be used to reduce the dimensionality of the scene description for other research questions. Pattern finding has been previously addressed [4,6,7], but only either for the spatial case, a Markovian progression or local motifs. To consider temporal information in a global non-Markovian fashion, we propose a Spatio-temporal Hierarchical Dirichlet Process (STHDP) model. STHDP leverages the power of Hierarchical Dirichlet Process (HDP) models to cluster location-velocity pairs and time simultaneously by introducing two mutually-influential HDPs. The results are presented as activity patterns and their time-varying presence (e.g. appear, peak, wane and di
Data Loading...
