A Finite State Machine Approach to Algorithmic Lateral Inhibition for Real-Time Motion Detection ^†

Area of Science:

• Computational Neuroscience
• Computer Engineering

Background:

• Finite state machines (FSMs) are well-defined computational models.
• Artificial neural networks (ANNs) are effective for problem-solving.
• Algorithmic lateral inhibition (ALI) is a neurally-inspired method for motion detection.

Purpose of the Study:

• To describe and apply control knowledge of the ALI method using FSMs.
• To implement ALI for motion detection on a Xilinx FPGA.
• To evaluate the performance of the FPGA-based ALI implementation.

Main Methods:

• Utilized finite state machines to represent the control knowledge of the ALI method.
• Implemented the ALI algorithm on a 16-nm Kintex UltraScale+ Xilinx FPGA.
• Tested the implementation on various motion detection datasets.

Main Results:

• Achieved accurate object tracking performance with a high F-score of 0.86 on complex sequences.
• Demonstrated real-time processing capabilities, surpassing previous ALI implementations.
• Outperformed both a complete ALI algorithm and a simplified 'accumulative computation' version.

Conclusions:

• FPGA implementation of ALI using FSMs enables accurate and real-time motion detection.
• This approach offers significant improvements over prior ALI methods in terms of speed and performance.
• Xilinx FPGAs provide the necessary precision for running neural algorithms with sensor technologies.