Multitarget Tracking Library

Recognizing and interpreting the motion of objects in image sequences is an essential task in a number of applications, such as security, surveillance, autonomous vehicles, etc. In many instances, the objects to be tracked have no known distinguishing features that would allow feature (or token) tracking optical flow or motion estimation. Therefore, the targets can only be identified and tracked by their measured positions and derived motion parameters.

The Multitarget Tracking Library contains a set of functions developed to track multiple targets in image sequences. The library contains high-performance algorithms with small memory footprints, enabling its use in resource constrained environments, like smart cameras performing surveillance, and other embedded devices. It provides datatypes, classes and functions for multitarget tracking applications with a special emphasis on tracking multiple targets in images sequence processing.

Features

The library provides methods for the estimation of the target positions based on their previous behavior and current measurements using:

• fixed-gain state estimation filters (alpha-beta-gamma steady-state Kalman filters),
  
• the implementation of the IMM (interacting multiple model), and
  
• two very fast methods for the association of the current measurements with live tracks (JVC and extended nearest-neighbor methods).