Sub-pixel detection of active targets

The problem of accurately detecting the location of active targets using low-cost machine vision systems is presented, with a requirement for location accuracies greater than 0.5 pixels. Methods of locating bloblike images of active targets are presented, including the calculation of the centroid of binarized versions of the blob, and the fitting of a parametric model to the blobs intensity distribution. Both methods are evaluated experimentally with simulated data and found to provide excellent blob localization. A physical experiment was conducted to evaluate the location accuracy, leading to the conclusion that the performance of the two methods is equivalent, with both providing the capability of locating active targets to an accuracy of 0.25 pixels