An Incremental DPMM-Based Method for Trajectory Clustering, Modeling, and Retrieval

Abstract

Trajectory analysis is the basis for many applications, such as indexing of motion events in videos, activity recognition, and surveillance. In this paper, the Dirichlet process mixture model (DPMM) is applied to trajectory clustering, modeling, and retrieval. We propose an incremental version of a DPMM-based clustering algorithm and apply it to cluster trajectories. An appropriate number of trajectory clusters is determined automatically. When trajectories belonging to new clusters arrive, the new clusters can be identified online and added to the model without any retraining using the previous data. A time-sensitive Dirichlet process mixture model (tDPMM) is applied to each trajectory cluster for learning the trajectory pattern which represents the time-series characteristics of the trajectories in the cluster. Then, a parameterized index is constructed for each cluster. A novel likelihood estimation algorithm for the tDPMM is proposed, and a trajectory-based video retrieval model is developed. The tDPMM-based probabilistic matching method and the DPMM-based model growing method are combined to make the retrieval model scalable and adaptable. Experimental comparisons with state-of-the-art algorithms demonstrate the effectiveness of our algorithm.

High Level Event Detection based on Spatial Occupancy and Interpersonal Relationships

“…A threestage process was used: learning interesting nodes by Gaussian Mixture Modeling, connecting routes using trajectory clustering, and encoding spatio-temporal activities using Hidden Markov Models (HMMs). Hu and colleagues [16] presented an incremental DPMM (Dirichlet Process Mixture Model) to cluster, model and retrieve trajectories. Each trajactory extracted by trackers is represented in the frequency domain, and clustered using an incremental DPMM that learns the number of clusters and can be updated on-the-fly (temporal changes in each trajectory can be detected as well, by using smaller tracks to build each trajectory).…”

Understanding vehicular traffic behavior from video: a survey of unsupervised approaches

“…Hu et al 20 Trajectory tDPMM Dirichlet process mixture model is used for unsupervised clustering and modified to handle temporal structure and ordering inherent in a trajectory sequence.…”

“…Very recent work by Hu et al 20 used the Dirichlet process mixture model (DPMM) to cluster, model, and retrieve trajectories in an incremental fashion. Further, the time-varying information contained in a trajectory is modeled using the time-sensitive DPMM (tDPMM) over subtrajectories.…”

“…14,20,47 In order to evaluate a clustering result, a one-to-one mapping between ground truth labels and those returned by the clustering algorithm is found. 14,20,47 In order to evaluate a clustering result, a one-to-one mapping between ground truth labels and those returned by the clustering algorithm is found.…”

“…10,20 This functionality enables a traffic engineer to find occurrences of a particular behavior, e.g., illegal U-turn or jaywalking, from a large trajectory database for closer inspection. Given the large amounts of visual data that are collected from traffic cameras, it will become important to succinctly index activities and events for further analysis.…”