Global optimization for coupled detection and data association in multiple object tracking

Abstract

We present a novel framework for tracking multiple objects imaged from one or more static cameras, where the problems of object detection and data association are expressed by a single objective function. Particularly, we combine a sparsity-driven detector with the network-flow data association technique. The framework follows the Lagrange dual decomposition strategy, taking advantage of the often complementary nature of the two subproblems. Our coupling formulation avoids the problem of error propagation from which traditional “detection-tracking approaches” to multiple object tracking suffer. We also eschew common heuristics such as “non-maximum suppression” of hypotheses by modeling the joint image likelihood as opposed to applying independent likelihood assumptions. Our coupling algorithm is guaranteed to converge and can resolve the ambiguities in track maintenance due to frequent occlusion and indistinguishable appearance between objects. Furthermore, our method does not have severe scalability issues but can process hundreds of frames at the same time. Our experiments involve challenging, notably distinct datasets and demonstrate that our method can achieve results comparable to or better than those of state-of-art approaches.