Coarse-to-Fine Contrast Maximization for Energy-Efficient Motion Estimation in Edge-Deployed Event-Based SLAM

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Event-based vision sensors offer high temporal resolution and low power, ideal for edge robotics and SLAM.
• Contrast Maximization (CMAX) is a direct geometric method for ego-motion estimation using event data.
• Conventional CMAX is computationally intensive due to processing full event sets and high-resolution images repeatedly.

Purpose of the Study:

• To develop a computationally efficient variant of CMAX for event-based motion estimation.
• To reduce the computational and energy costs of CMAX without sacrificing accuracy.
• To enable real-time SLAM on resource-constrained edge platforms.

Main Methods:

• Proposed Coarse-to-Fine Contrast Maximization (CCMAX), a computation-aware CMAX approach.
• Implemented progressive IWE resolution increase and coarse-grid event subsampling in early stages.
• Retained full-resolution refinement in the final optimization stage.

Main Results:

• CCMAX achieved accuracy comparable to full-resolution CMAX on standard benchmarks.
• Reduced floating-point operations (FLOPs) by up to 42%.
• Demonstrated up to 87% lower energy consumption on a RISC-V edge SoC.

Conclusions:

• CCMAX offers a significant reduction in computational cost and energy consumption for event-based motion estimation.
• The proposed method is suitable for real-time edge SLAM on power- and resource-constrained platforms.
• CCMAX presents an energy-efficient front-end for advanced robotics applications.