Boosting the multiple aircraft online tracking performance via enriching the associated data with fused targets features

Abstract

ABSTRACT Multi aircraft tracking from an aerial view is a backbone for several military and civilian applications. Recent tracking by detection approaches was utilized to accomplish such multiple target tracking (MTT) tasks as Simple Online and Real-time Tracking (SORT). SORT is a strong and fast MTT, that employs a Kalman filter for the target motion parameters and the Hungarian method for the data association. But it discards the target appearance for resolving the association problem to preserve the real-time execution which results in increasing the number of (IDS) Identity Switches and decreasing the tracking accuracy. In this work, the target appearance information is incorporated alongside its geometry to leverage the tracking accuracy and reduce the tracklet fragments due to the high number of IDS. The target shape and contextually based feature of Histogram orientation of Gradient (HOG) are combined with its color histogram to enrich the Hungarian association with the appearance information. A recent-released multi-aircraft data set is utilized to examine the proposed improvement through a comparative experiment that reveals the MTT performance-boosting while keeping the real-time execution. The proposed method reduces the tracked targets IDsw by 60.97% that improves the tracker overall accuracy by 8.6% compared to the SORT tracker.