"Eztrack": A single-vehicle deterministic tracking algorithm

Carrano, C J

Lawrence Livermore National Laboratory

关键词: Cameras;    99 General And Miscellaneous//Mathematics, Computing, And Information Science;    Color;    Sampling;    42 Engineering;   

DOI  :  10.2172/924186
RP-ID  :  LLNL-TR-400667
RP-ID  :  W-7405-ENG-48
RP-ID  :  924186

美国|英语

来源: UNT Digital Library

【 摘 要 】

A variety of surveillance operations require the ability to track vehicles over a long period of time using sequences of images taken from a camera mounted on an airborne or similar platform. In order to be able to see and track a vehicle for any length of time, either a persistent surveillance imager is needed that can image wide fields of view over a long time-span or a highly maneuverable smaller field-of-view imager is needed that can follow the vehicle of interest. The algorithm described here was designed for the persistence surveillance case. In turns out that most vehicle tracking algorithms described in the literature[1,2,3,4] are designed for higher frame rates (> 5 FPS) and relatively short ground sampling distances (GSD) and resolutions ({approx} few cm to a couple tens of cm). But for our datasets, we are restricted to lower resolutions and GSD's ({ge}0.5 m) and limited frame-rates ({le}2.0 Hz). As a consequence, we designed our own simple approach in IDL which is a deterministic, motion-guided object tracker. The object tracking relies both on object features and path dynamics. The algorithm certainly has room for future improvements, but we have found it to be a useful tool in evaluating effects of frame-rate, resolution/GSD, and spectral content (eg. grayscale vs. color imaging ). A block diagram of the tracking approach is given in Figure 1. We describe each of the blocks of the diagram in the upcoming sections.