Multi-sensor target tracking using multiple hypothesis testing

Abstract

The invention is for a multi-sensor multiple hypotheses testing tracking system. The multi-hypothesis testing system associates measurements from multiple sensors with tracks. Measurements are incorporated using a Kalman Filter and the same filters are used to propagate the trajectories of the tracks.

1 . A system for tracking a plurality of moving vehicles that reduces a number of computations required, the system comprising:
 a plurality of sensors that form a plurality of scans by measuring positions of the plurality of moving vehicles over a plurality of time intervals, wherein each of the plurality of scans is a collection of measurements from the plurality of sensors taken in a particular time interval; and   a processor that is communicatively coupled to the plurality of sensors; wherein the processor:   receives, from the plurality of sensors, a first collection of measurements of a first scan measured during a first time interval, wherein each measurement of the first collection of measurements is a single measurement from one of the plurality of sensors,   generates a first group of tracks using a dynamic state transition model, wherein the first group of tracks includes a track for each of the first collection of measurements;   forms a first hypothesis for the first group of tracks, wherein the first hypothesis is a set of compatible tracks, wherein two or more tracks are compatible if each track describes a different target and do not share an identical measurement from any of the plurality of scans;   removes one or more particular tracks from the first group of tracks that are not included in the first hypothesis to form a first pruned group by assigning a score to each track group;   predicts using a Kalman filter, expected positions of each of the plurality of moving vehicles based on the first pruned group;   updates the dynamic state transition model to form an updated dynamic transition model based on a plurality of covariances calculated for the first pruned group,   receives, from the plurality of sensors, a second collection of measurements from a second scan during a second time interval,   forms a second hypothesis based on the updated dynamic transition model, and   tracks the plurality of moving vehicles based on the second hypothesis.   
     
     
         2 . The system of  claim 1 , wherein the plurality of sensors includes a camera. 
     
     
         3 . The system of  claim 1 , wherein the plurality of sensors includes a radar device. 
     
     
         4 . The system of  claim 1 , wherein the plurality of sensors includes a range finder. 
     
     
         5 . The system of  claim 1 , wherein the plurality of sensors is an acoustic range finder. 
     
     
         6 . The system of  claim 1 , wherein the Kalman filter is an unscented Kalman filter. 
     
     
         7 . The system of  claim 1 , wherein the Kalman filter is any combination of a Kalman filter, an unscented Kalman filter, and an extended Kalman filter. 
     
     
         8 . A method for tracking a plurality of moving vehicles that reduces a required number of computations, the method comprising:
 receiving, from a plurality of sensors, a first collection of measurements of a first scan measured during a first time interval, wherein the plurality of sensors form a plurality of scans by measuring positions of the plurality of moving vehicles over a plurality of time intervals, wherein each of the plurality of scans is a collection of measurements from the plurality of sensors taken in a particular time interval;   generating a first group of tracks using a dynamic state transition model, wherein the first group of tracks includes a track for each of the first collection of measurements;   forming a first hypothesis for the first group of tracks, wherein the first hypothesis is a set of compatible tracks, and two or more tracks are compatible if each track describes a different target and do not share an identical measurement from any of the plurality of scans;   removing one or more particular tracks from the first group of tracks that are not included in the first hypothesis to form a first pruned group by assigning a score to each track;   predicting, using a Kalman filter, expected positions of each of the plurality of moving vehicles based on the first pruned group;   updating the dynamic state transition model to form an updated dynamic transition model based on a plurality of covariances calculated for the first pruned group;   receiving, from the plurality of sensors, a second collection of measurements from a second scan during a second time interval;   forming a second hypothesis based on the updated dynamic transition model; and   tracking the plurality of moving vehicles based on the second hypothesis.   
     
     
         9 . The method of  claim 8 , wherein the second hypothesis is formed by appending one of the second collection of measurements to an existing track in the first pruned group by generating a new track that includes at least one of the first collection of measurements associated with the existing track on a condition that the one of the second collection of measurements describes an object described by the existing track. 
     
     
         10 . The method of  claim 8 , wherein the second hypothesis is formed by generating a new track, independent of all of first collection of measurements, using one of the second collection of measurements on a condition that the one of the second collection of measurements does not describe an object described by any existing track in the first hypothesis.