Motion-constrained, multiple-hypothesis, target- tracking technique

Abstract

In one embodiment, an infrared camera generates thermal images and a computer generates track trees for volant targets (e.g., bats) detected in the images using an HKUF technique and detection filtering. The computer extends each track tree by adding new branches for each target detected in a subsequent image, generates motion metrics for each new branch, applies biologically informed motion constraints to the motion metrics to delete branches from the track trees, selects a track in each terminated track tree as a most likely track for the corresponding detected target using a track-scoring system based on a combination of track tortuosity and Kalman prediction errors, automatically generates a counting box based on density plots generated using the selected tracks, and calculates a census of the volant targets based on the selected tracks relative to the counting box. The technique handles targets that may be occluded for one or more images.