Track Stitching and Un-Switching Algorithms for Multiple Target Tracking

Abstract

Track breakages are common in target tracking due to highly maneuvering targets, association with false alarms or incorrect target-originated measurements, low detection probability, close target formations, large measurement errors, and long sampling intervals, among other causes. Existing track segment association (TSA) algorithms solve this breakage problem by predicting old track segments and retrodicting young track segments to a common time followed by two-dimensional (2-D) assignment. This approach presents two disadvantages. First, these algorithms predict or retrodict from the actual point of termination or beginning of their respective tracks: that is, they do not check if the cause of a track termination was incorrect association nor do they redress such an erroneous association. Second, these algorithms do not utilize the measurement information during the breakage period. But very often, track terminations occur due to incorrect measurement association. To solve the first aforementioned problem, a 2-D assignment-based TSA algorithm is proposed which releases incorrectly associated measurements by going backward and forward in time along old and young track segments, respectively, and then performing prediction and retrodiction. Further, to address both these shortcomings in existing TSA algorithms simultaneously, a novel multi-frame assignment-based TSA algorithm is proposed which estimates the track during the breakage period, utilizing both unassociated and released measurements simultaneously. In addition to the track breakages, other frequently encountering issue in multipletarget tracking is track switching/swapping. Track stitching or TSA algorithms have been proposed to stitch broken track segments deemed to have originated from the same target across time and to improve track continuity. On the other hand, measurements from closely-spaced multiple targets fall within their validation gates causing tracking errors that eventually lead to not just track breakage but also track swapping. Therefore, TSA alone is insufficient to improve the overall tracker performance as it considers only the broken tracks but not the continuous ones that might have swaps among themselves or with other broken tracks.To address track swapping issue, an algorithm, which detects and breaks possible track swaps before un-swapping using kinematic, classification, and amplitude information, is proposed. Track swap detection involves identifying the most likely instant of track swap without ground truth. Further, the proposed algorithm is extended to stitch broken track segments (as in the standard TSA case) as well as those track segments that are algorithmically broken due to detection of possible swaps. Moreover, all the proposed algorithms can handle target maneuvers subject to a single turn during the breakage period. In the proposed solution, model parameters such as starting time of the turn, ending time of the turn, and turn rate are obtained by maximizing the likelihood that a given measurement-tuple originated from the track couple under consideration. Simulation results demonstrate that the proposed algorithms are superior to existing ones in terms of improved track continuity and consistency in track identity maintenance.