Overview

This paper presents a novel method to extract dominant motion patterns (MPs) and the main entry/exit areas from a surveillance video. The method first computes motion histograms for each pixel and then converts it into orientation distribution functions (ODFs). Given these ODFs, a novel particle meta-tracking procedure is launched which produces meta-tracks, i.e. particle trajectories. As opposed to conventional tracking which focuses on individual moving objects, meta-tracking uses particles to follow the dominant flow of the traffic. In a last step, a novel method is used to simultaneously identify the main entry/exit areas and recover the predominant MPs.

The meta-tracking procedure is a unique way to connect low-level motion features to long-range MPs. This kind of tracking is inspired by brain fiber tractography which has long been used to find dominant connections in the brain.

Our method is fast, simple to implement, and works both on sparse and extremely crowded scenes. It also works on highly structured scenes (highways, traffic-light corners, etc) as well as on chaotic scenes containing moving objects crossing each other and following different directions. Our method locates short-range and long-range MPs of arbitrary shape with high accuracy.

Pipeline

Our method is a 4-step pipeline which are:

1. computes a motion histogram for each pixel
2. converts motion histograms into ODFs
3. performs meta-tracking
4. clusters meta-tracks to recover MPs as well as the entry/exit points

Experimental Results

Meta-tracking Results

Our meta-tracking method works well on videos with complex layout and unstructured dynamics. It takes on average 30 secs to track 1000 particles on a 2.8 GHz laptop with Matlab code.

Clustering Results

We use a bottom-up hierarchical clustering method to recover the predominant motion patterns of a video sequence. Some results are presented below.

Videos and Matlab code

A Matlab implementation of our work can be downloaded here (1.3 Gb). This code loads precomputed ODF, computes metatracks and performs clustering to recover motion patterns. Precomputed ODF for 4 video sequences are provided.

The complete video dataset used in the paper can be downloaded here (2.4 Gb).

Contact

• Pierre-Marc Jodoin (Université de Sherbrooke), pierre-marc.jodoin@usherbrooke.ca
• Yannick Benezeth (Université de Bourgogne), yannick.benezeth@u-bourgogne.fr
• Yi Wang (Université de Sherbrooke), yi.wang@usherbrooke.ca