Trajectory Analysis and Semantic Region Modeling Using Nonparametric Hierarchical Bayesian Models

Abstract

We propose a novel framework of using a nonparametric Bayesian model, called Dual Hierarchical Dirichlet Processes (Dual-HDP) (Wang et al. in IEEE Trans. Pattern Anal. Mach. Intell. 31:539-555, 2009), for unsupervised trajectory analysis and semantic region modeling in surveillance settings. In our approach, trajectories are treated as documents and observations of an object on a trajectory are treated as words in a document. Trajectories are clustered into different activities. Abnormal trajectories are detected as samples with low likelihoods. The semantic regions, which are subsets of paths commonly taken by objects and are related to activities in the scene, are also modeled. Under Dual-HDP, both the number of activity categories and the number of semantic regions are automatically learnt from data. In this paper, we further extend Dual-HDP to a Dynamic Dual-HDP model which allows dynamic update of activity models and online detection of normal/abnormal activities. Experiments Part of this work was published in Wang et al. (2008).are evaluated on a simulated data set and two real data sets, which include 8, 478 radar tracks collected from a maritime port and 40,453 visual tracks collected from a parking lot.

Learning and Classification of Car Trajectories in Road Video by String Kernels

“…The proposed method has been validated on the MIT trajectories dataset (Wang et al, 2011), a standard and freely available dataset composed by 40.453 trajectories obtained from a parking lot scene within five days. Starting from the entire dataset D, a subset D * of trajectories belonging to vehicles (10.335) has been manually extracted by an expert (see Figure 3a) and the proposed system has been evaluated.…”

Coherent Filtering: Detecting Coherent Motions from Crowd Clutters

“…Meanwhile, the high-level semantic analysis in crowded scenes focuses on modeling scene structures and recognizing crowd behaviors. Wang et al [8,14] and Zhou et al [10] used hierarchial topic models to learn the models of semantic regions from the co-occurrence of optical flow and tracks. Zhou et al [9] proposed a mixture model of dynamic pedestrian-agents to learn the collective crowd behavior patterns in crowded scenes.…”

Motion Pattern Study and Analysis from Video Monitoring Trajectory

“…W. Hu [11] also introduced the fuzzy Self-Organizing Neural Network based method and make the learning process much more efficient. C. Stauffer [12] and X. Wang [13] put forward CoOccurrence Decomposition, where trajectories were viewed as a bag of words where similar bags contain similar words. A co-occurrence matrix was formed from training data and decomposed to build document subjects (routes).…”

“…The Baum-Welch algorithm is used for estimating final parameters of HMM, and the convergence threshold of our HMM is 1×10 −4 , which can be computed as Formula (13), where Pr i represents the maximum likelihood estimation in this HMM. To simplify the computational complexity, the iteration of training HMM is set as 30 in our work showed in Fig.…”