MSU Technologies

Random Finite Set based Tracking for 2D/3D Object Detection

VALUE PROPOSITION

Multi object tracking is critical in enabling an autonomous vehicle to perceive and maneuver its environment. Three common flaws in multi-object tracking detection is (1) uncertainty in the number of objects, (2) uncertainty regarding when and where the objects may appear and disappear and (3) uncertainty in objects states. A Random Finite Set (RFS) based method is paired with Poisson Multi-Bernoulli Mixture (PMBM) filtering to both reduce frequency of flaws and improve efficiency while saving in computational cost.

DESCRIPTION OF TECHNOLOGY

Random Finite Set tracking is the process of tracking multiple objects by employing the use of a Poisson Multi-Bernoulli Mixture which models detected and undetected objects using two probability distributions. This process predicts object location. Objects would be categorized as newly detected objects, previously detected objects or clutter (i.e. false positives). The system provides for improved multi-object tracking by reducing the amount of data for processing based on object identifiers, continuation of movement and filtering using probabilities determined during filtering.

BENEFITS

An upgraded approach to object tracking by employing the use of an RFS- based detection method.
It applies a Poisson multi-Bernoulli mixture (PMBM) filter in conjunction with 3D LiDAR data to more accurately depict objects in both 2D and 3D.
Improved efficiency would greatly benefit autonomous vehicle applications.
Can be integrated into current platforms.

APPLICATIONS