Histogram probabilistic multi-hypothesis tracker with colour attributes

Abstract

The histogram probabilistic multi-hypothesis tracking (H-PMHT) algorithm is a multi-target track-before-detect method that has been demonstrated to be scalable and efficient and shows good performance on video imagery. However, its application has been limited to intensity-only imagery. In practice, colour and information, such as texture, are often available and can be very helpful when targets are close together. This study compares two alternative methods that incorporate colour information in the H-PMHT: an existing method referred to as spectral H-PMHT and a new approach that treats the colour information as attribute data. The methods are compared with intensity-only H-PMHT on simulations and benchmark data from the VS-PETS03 archive.