Hough Transform Implementation For Event-Based Systems: Concepts and Challenges.

Area of Science:

• Computer Vision
• Neuromorphic Engineering
• Image Processing

Background:

• Conventional Hough Transform (HT) is effective for frame-based systems but not directly applicable to event-based cameras.
• Event-based cameras (e.g., Dynamic Vision Sensors) offer high temporal resolution but output change-based data, not color.
• Adapting existing algorithms is crucial for leveraging event-based camera capabilities.

Purpose of the Study:

• To systematically adapt the conventional Hough Transform for event-based vision systems.
• To implement the adapted 3D Hough Transform using Spiking Neural Networks (SNNs).
• To explore efficient SNN implementation techniques for hardware realization.

Main Methods:

• Extension of conventional HT to a 3D Hough Transform.
• Adaptation of the 3D HT algorithm for event-based data streams.
• Implementation of the 3D HT using Spiking Neural Networks (SNNs) for FPGA deployment.

Main Results:

• Successful adaptation of HT for event-based camera data.
• Feasible hardware implementation of 3D HT using SNNs on FPGAs, minimizing CPU and memory requirements.
• Discussion of optimization techniques for SNNs regarding neuron count, accuracy, and resolution.

Conclusions:

• The proposed SNN-based 3D HT effectively processes event-based camera data.
• This approach facilitates efficient, low-power hardware implementation for real-time applications.
• The work aims to reduce the performance gap between event-based and frame-based vision systems.