Object Matching Across Multiple Non-overlapping Fields of View Using Fuzzy Logic
Loke Yuan Ren, Pankaj Kumar, Weimin Huang · Acta Automatica Sinica · 2006
An approach based on fuzzy logic for matching both articulated and non-articulated objects across multiple non-overlapping field of views (FoVs) from multiple cameras is proposed. We call it fuzzy logic matching algorithm (FLMA). The approach uses the information of object motion, shape and camera topology for matching objects across camera views. The motion and shape information of targets are obtained by tracking them using a combination of ConDensation and CAMShift tracking algorithms. The information of camera topology is obtained and used by calculating the projective transformation of each view with the common ground plane. The algorithm is suitable for tracking non-rigid objects with both linear and non-linear motion. We show videos of tracking objects across multiple cameras based on FLMA. From our experiments, the system is able to correctly match the targets across views with a high accuracy. Multi-target tracking across multiple cameras with non-overlapping field of views (FoVs) is a challenging task. The clues are the behaviors of the targets before they disappear in an FoV and after they reappear in another FoV. An approach based on fuzzy logic, fuzzy logic matching algorithm (FLMA) is proposed to find the correspondence of multiple targets in multi-camera non-overlapping FoV network. Such a system is useful for analysis of behaviors under multi-camera system, which can be applied for surveilling large area such as highways, shopping malls, office buildingsetc. An overview of the system is showed in Fig.1. It consists of several intra-camera tracking systems and an inter-camera tracking system. Intra-camera tracking system is used for tracking targets in the FoV of each camera. The first few frames of the video is used to build the background model, that is used to detect the foreground objects. These foreground objects are initialized and tracked by using ConDensation (1) and CAMShift (2) tracking algorithms. Condensation algorithm is used for estimating the states of the targets and CAMShift is used for data association. We have used ConDensation algorithm for multiple target tracking in single camera FoV, where the data association is done using the search window approximated by the CAMshift filter for each target. It takes care of errors in tracking which may arise due to the splitting of targets, a common phenomena in detection of articulated objects. This has lead to significant improvement in tracking of articulated objects compared to previous work (3) , which used Kalman filter for tracking rigid targets with linear motion. The present work uses particle filter and CAMshift filter to track the targets even when they split into more than one measurement. Splitting of targets is especially true for articulated targets, with non linear motion. Once a target leaves the FoV, inter-camera tracking system takes over the tracking task. In the inter-camera tracking system, the FoV of each camera is projected onto a common ground plane view (CGPV) by using projective transformation, which provides the position and motion of the targets in CGPV. By manually matching the points in the common ground plane with those in the image plane we compute the perspective transform of each view on to the common ground plane view (CGPV). A targets position and motion in CGPV and its shape information are used to match a new initialized target with the targets in the missing target list. Missing target list is a list that stores the states of the targets that have left the FoVs of other cameras in the camera network. Thus the objective of the inter-camera tracking system is to determine whether a new target in the FoV of a camera is a new target in the whole camera network or a missing target from other views. To address the challenging task of matching targets across views with no overlap, we have developed a system based on fuzzy logic to match targets across the FoVs of the different cameras. The system proposed here is able to track