Multitarget-tracking optical processing system

Area of Science:

• Computer Vision
• Signal Processing
• Optical Engineering

Background:

• Multitarget tracking systems estimate the state (position and velocity) of multiple objects using sensor measurements.
• Integrating analog optical processing offers potential for improved computational efficiency and speed in complex tracking tasks.
• Data association, linking current measurements to existing tracks, is a critical challenge in multitarget tracking.

Purpose of the Study:

• To evaluate the efficacy of analog optical processing within a comprehensive multitarget tracking system.
• To investigate the role of a neural network-based data-association subsystem.
• To demonstrate the system's capability in handling measurement noise and data errors.

Main Methods:

• Development of a full multitarget tracking system incorporating analog optical subsystems.
• Implementation of a neural network for the data-association task, linking measurements to target estimates.
• Utilization of optical detection and Hough-transform subsystems for measurement refinement and error correction.

Main Results:

• The neural network data-association subsystem, combined with a fixed-coefficient estimation filter, achieved effective measurement-to-estimate associations.
• Optical detection successfully reduced frame-to-frame jitter in measurements.
• The Hough-transform subsystem demonstrated utility in correcting data dropout and errors within the tracking and association processes.

Conclusions:

• Analog optical processing is a valuable component for enhancing the performance of full multitarget tracking systems.
• The proposed neural network approach for data association is efficient and effective.
• The integrated system shows robustness in handling measurement inaccuracies and data dropouts.