Rule-Based Multiple Object Tracking for Traffic Surveillance Using Collaborative Background Extraction
In order to address the challenges of occlusions and background variations, we propose a novel and effective rule-based multiple object tracking system for traffic surveillance using a collaborative background extraction algorithm. The collaborative backg
- PDF / 848,051 Bytes
- 10 Pages / 430 x 660 pts Page_size
- 44 Downloads / 173 Views
Abstract. In order to address the challenges of occlusions and background variations, we propose a novel and effective rule-based multiple object tracking system for traffic surveillance using a collaborative background extraction algorithm. The collaborative background extraction algorithm collaboratively extracts a background from multiple independent extractions to remove spurious background pixels. The rule-based strategies are applied for thresholding, outlier removal, object consolidation, separating neighboring objects, and shadow removal. Empirical results show that our multiple object tracking system is highly accurate for traffic surveillance under occlusion conditions.
1 Introduction Multiple object tracking (MOT) is important for visual surveillance and event classification tasks [1]. However, due to challenges such as background variation, occlusion, and object appearance variation, MOT is generally difficult. In the case of traffic surveillance, background variations in terms of illumination variation, small motions in the environment, weather and shadow changes, occlusions in terms of vehicles overshadowed or blocked by neighboring vehicles, trees, or constructions, and vehicle appearance changes in terms of different sizes of the same vehicles in different video frames, contribute to inaccurate visual tracking. As traditional visual tracking methods, feature-based tracking detects features in a video frame and searches for the same features nearby in subsequent frames; Kalman filtering [2] uses a linear function of parameters with respect of time, and assumes white noise with a Gaussian distribution, however, the method with the Kalman filtering to predict states of objects can not be applied to objects in occlusion [3]; particle filtering [4] is appealing in MOT for its ability to have multiple hypotheses, however, its direct application for multiple object tracking is not feasible. For traffic surveillance videos that generally have stationary background, it is important to segment moving vehicles from the background either when viewing the scene from a fixed camera or after stabilization of the camera motion. With the assumption of a stationary camera, we can simply threshold the difference of intensities between the current image frame with the background image, I(x,y)-Ibg(x,y), to segment the moving objects from the background. However, due to background variations, this simple approach may not work well in general. In previous work, normal G. Bebis et al. (Eds.): ISVC 2007, Part II, LNCS 4842, pp. 469–478, 2007. © Springer-Verlag Berlin Heidelberg 2007
470
X. Su et al.
(Gaussian) distribution, linear prediction and adaptation [5], and hysteresis thresholding [6] have been investigated to model the background changes. We proposed a rule-based multiple object tracking system using a collaborative background extraction algorithm for the application of traffic surveillance, which is easy-to-implement and highly effective in handling occlusions in terms of removing outliers and shadows, consol
Data Loading...
