ASSET-2 - A Scene Segmenter Establishing Tracking

ASSET-2 Introduction and Overview

	This is ASSET-2; A Scene Segmenter Establishing Tracking - Version 2. ASSET-2 is a complete real-time vision system for segmentation and tracking of independently moving objects. ASSET-2 requires no camera calibration or motion registration, and performs all calculations in the image plane. Information gained from the motion of two dimensional features is integrated with one dimensional boundaries. ASSET-2 achieves automatic instantiation of tracking for each new object detected, and any number of objects can be tracked. Occlusion of one object by another is correctly handled. The temporal integration of segmentation information in such a robust and flexible way is a step forward in the area of image sequence understanding. For a pictorial overview, press here.
	ASSET-2 starts by tracking 2D image features as they move across the picture over time, to get a sparse image flow field. The flow field is then segmented into clusters which have internally consistent flow variation (using a first-order - six parameter - fit to the flow field) and which have different flow to each other and to the background. The green shapes above are examples of initial segmentation. The cluster shapes are filtered over time to give improved accuracy and robustness.
	ASSET-2 has been implemented as a real-time system on PowerPC/Transputer image processing equipment. See here for details of using ASSET-2 for traffic and pedestrian analysis.

Tracking a single large object (image sequence):
A harrier taken with a shaky hand-held camcorder:
The result of tracking a vehicle at night. The video was taken using an infra-red camera:
The result of tracking vehicles from a swaying traffic-monitoring video camera (complete image):
The result of tracking vehicles from a video camera over a motorway; real vehicle speeds are calculated from image speeds and positions, once the simple real-world calibration has been carried out (complete image):
Further traffic and pedestrian analysis.

Papers on ASSET-2 have been published in EAAI, ICCV95, PAMI and JRTI; for a complete report, either download the PostScript (asset2.ps.gz - 1 Meg / asset2.ps - 3.5 Megs) or view the html.