Incorporating fully sparse topic models for abnormality detection in traffic videos

Abstract

Automatic analysis and understanding of typical activities and identification of abnormal events in crowded traffic scenes is a fundamental task for traffic video surveillance. In this paper, we address the problem of abnormality detection based on an unsupervised learning approach with Fully Sparse Topic Models (FSTM). The method uses a set of visual features and automatically discovers the activity patterns occurring in complicated scenes. We show how the discovered patterns can be used to detect abnormal events. Furthermore, we compare FSTM with other topic models based on various measures. Experimental results and comparisons on two traffic datasets demonstrate that our approach outperforms other methods in finding meaningful activity patterns and discovers the abnormal events accurately.

Low‐rank structured sparse representation and reduced dictionary learning‐based abnormity detection

“…Hospedales et al [46] introduce a Markov clustering topic model which combines the benefits of DBNs and PTMs. Sparse topical coding [36] and fully sparse topic model [37] are used to find the motion patterns and model the interactions between the objects.…”

“…To overcome the difficulties mentioned above, low‐level features such as pixel‐level motion and appearance have been widely applied [1, 9, 13, 15, 19, 20, 23, 26, 29, 33, 35–37, 43–47, 52]. Optical flow is one of the most popular features for motion pattern, which can be used to describe the position and motion direction of each local patch and generate bag‐of‐words representation [15, 33, 35–37, 46].…”

“…To overcome the difficulties mentioned above, low‐level features such as pixel‐level motion and appearance have been widely applied [1, 9, 13, 15, 19, 20, 23, 26, 29, 33, 35–37, 43–47, 52]. Optical flow is one of the most popular features for motion pattern, which can be used to describe the position and motion direction of each local patch and generate bag‐of‐words representation [15, 33, 35–37, 46]. Histograms of local optical flow [13], a histogram of gradient and optical flow [20], and a multi‐scale histogram of optical flow [23, 47] can encode the flow into a histogram and generate rotation and translation invariant representation.…”

Beyond bag-of-words: An improved Sparse Topical Coding for learning motion patterns in traffic scenes

“…Topic Models are the basis of most recent approaches, targeting complex scene analysis [6][7][8][9][10][11][12][13][14]. Probabilistic Topic models (PTMs) such as Probabilistic Latent Semantic Analysis (PLSA) [15], Latent Dirichlet Allocation (LDA) [16] and Hierarchical Dirichlet Process (HDP) [17] were first introduced to discover latent topics in large text corpora and then utilized by researchers for video analysis.…”

A Regional Topic Model Using Hybrid Stochastic Variational Gibbs Sampling for Real-Time Video Mining

“…Jeong et al [26] proposed a topic model for detecting anomalous trajectories of people or vehicles in surveillance-video images. Kaviani et al [27] addressed the problem of abnormality detection based on a fully sparse topic models (FSTM). Isupova et al [28] proposed a novel dynamic Bayesian nonparametric topic model and its Batch and online Gibbs samplers for anomaly detection in video.…”