Multi-Object Tracking Based on Soft Assignment of Detection Responses

Abstract

We introduce a new detection-based method that is able to track multiple objects from a single camera. The method is built upon an approach that combines Kalman filtering and the Expectation Maximization (EM) algorithm. The benefit of this approach is that soft assignment of the detections to corresponding objects can be performed automatically using their a posteriori probabilities. This is a general approach for detectionbased multi-object tracking, and there are various ways to detect the objects. In this paper, we demonstrate the applicability of the approach for tracking multiple pedestrians and faces using a basic cascade detector.