PttAcc: Pipeline-Based Taylor Expansion Fitting Arctangent Angle Hardware Accelerator Design for Descriptor Duty in ORB-SLAM System

Abstract

In simultaneous localization and mapping (SLAM), the oriented FAST and rotated BRIEF (ORB) algorithm is key for feature detection and description, but its computational demands, particularly in arctangent angle computation, hinder real-time use on embedded systems. To solve this, we introduce PttAcc, a hardware accelerator using pipeline architecture and the Taylor expansion fitting algorithm. Our simulations show PttAcc significantly accelerates arctangent computation while maintaining accuracy, reducing total area by 82.56% and power consumption by 63.49% compared to coordinate rotation digital computer (CORDIC) components. It achieves a performance acceleration ratio of 15.8 and speeds up descriptor tasks by up to 4.3 times.