High precision target tracking with a compound-eye image sensor
Rubakumar Krishnasamy, Winnifred Wong, Edward Shen, Srdjan Pepic, Richard I. Hornsey, Paul J. Thomas · 2004
Rapidly moving objects are a challenge to track with a camera system, particularly when the range varies from near-field to far-field and there is a complex background. In the present work, the problem is restricted to a well-defined small object that is readily distinguished from the background scene. The emphasis is on the maintenance of an accurate trajectory estimation as the target moves from the near field to the far field (or vice versa) and traverses a large field of view. To address this and related scenarios, a prototype compound-eye image sensor, named "DragonflEYE", has been designed and fabricated. In essence, the compound-eye sensor uses a large number (10/sup 1/-10/sup 3/) of identical "eyelets" to cover a large angular field of view. The degree of overlap of the eyelet fields of view is an important system parameter that strongly influences the tracking problem. Selected highlights of the design and implementation of DragonflEYE are described. For the precision-tracking function, the following factors need to be assessed: coverage overlap, update rate, frame timing, subpixel calibration of multiple eyelet lenses and processing algorithms and approaches. Investigations with conventional cameras indicate that the instantaneous location precision can be 0.01 to 0.1 pixel for the centroid of a suitable 'spot object' in a favourable scenario. The calibration of multiple "eyelets" of the compound eye to the same level of precision is expected to be important for precision tracking over a wide field of view. Preliminary experimental results for calibration for tracking applications are presented and evaluated.