Real-time tracking of multiple objects using adaptive correlation filters with complex constraints

Abstract

A real-time system for classification and tracking of multiple moving objects is proposed. The system employs a bank of composite correlation filters with complex constraints implemented in parallel on a graphics processing unit. When a scene frame is captured, the system splits the frame into several fragments on the base of a modeling kinematic prediction of target's locations. The fragments are processed with a bank of adaptive filters. The filters are synthesized with the help of an iterative algorithm, which optimizes discrimination capability for each target. Using complex constraints in the filter design, multiple objects in the input frame can be detected and classified by analyzing the intensity and phase distributions on the output complex correlation plane for each fragment. The performance of the proposed system in terms of tracking accuracy, classification efficiency and time expenses is tested and discussed with synthetic and real input-scene sequences. The results are compared with those of common techniques based on correlation filtering.