The Sequential Processing of Report Detections in Multitarget Data Association

Most data association algorithms are structured recursively. Such an approach directly reduces the computational load as well as allowing one to prune low scoring hypotheses at each stage. On the other hand, typical sensor models include the possibility of muliple detections. This type of model is theoretically desirable inasmuch as it implies independence of the target states; however, if a sensor report contains more than a few measurements, excessive pruning prior to inputting that report may be required in order to make room for new hypothese. This paper develops a recursive Bayesian algorithm for the processing of entire reports and then derives an equivalent model in which individual report measurements are input sequentially.