Presenter Information

Kyle Ingersoll

Location

The Leonardo Event Center

Start Date

5-12-2015 12:00 PM

Description

The Recursive-Random Sample Consensus (R-RANSAC) algorithm is a novel multiple target tracker designed to excel in tracking scenarios with high amounts of clutter measurements. R-RANSAC classifies each in-coming measurement as an inlier or an outlier; inliers are used to update existing tracks whereas are used to gen-erate new, hypothesis tracks using the standard RANSAC algorihtm. R-RANSAC is entirely autonomous in that it initiates, updates, and deletes tracks without user input. The tracking capabilities of R-RANSAC are extended by merging the algorithm with the Sequence Model (SM). The SM is a machine learner that learns sequences of identifiers. In the tracking context, the SM is used to learn sequences of target locations; in essence, it learns target trajectories and creates a probability distribution of future target locations. Simulation results demonstrate significant performance improvement when R-RANSAC is augmented with the SM, most noticeably in situation with high signal-to-noise ratio (SNR) and infrequent measurement updates

Share

COinS
 
May 12th, 12:00 PM

Multiple Target Tracking with Recursive-RANSAC and Machine Learning

The Leonardo Event Center

The Recursive-Random Sample Consensus (R-RANSAC) algorithm is a novel multiple target tracker designed to excel in tracking scenarios with high amounts of clutter measurements. R-RANSAC classifies each in-coming measurement as an inlier or an outlier; inliers are used to update existing tracks whereas are used to gen-erate new, hypothesis tracks using the standard RANSAC algorihtm. R-RANSAC is entirely autonomous in that it initiates, updates, and deletes tracks without user input. The tracking capabilities of R-RANSAC are extended by merging the algorithm with the Sequence Model (SM). The SM is a machine learner that learns sequences of identifiers. In the tracking context, the SM is used to learn sequences of target locations; in essence, it learns target trajectories and creates a probability distribution of future target locations. Simulation results demonstrate significant performance improvement when R-RANSAC is augmented with the SM, most noticeably in situation with high signal-to-noise ratio (SNR) and infrequent measurement updates