Beam adjustment method hardware accelerator based on Zynq FPGA

Abstract

The invention discloses a beam adjustment method hardware accelerator based on a Zynq FPGA. Shuer elimination, jacobian matrix updating and cost function calculation parts in a beam adjustment methodare realized at a PL end of Zynq through circuit design, and other parts are realized at a PS end of Zynq through software programming; an accelerator PL end splits two hardware circuits according toan algorithm, independent hardware is used for Shuer elimination, and Jacobian matrix updating and cost function calculation share the same hardware; wherein different computing units are formed in aninternal circuit of the accelerator, data transmission is carried out between the computing units by adopting a dual-port RAM (Random Access Memory) on an FPGA (Field Programmable Gate Array) chip, afinite-state machine is adopted in each computing unit, and read-in, read-out and storage of each step of data are controlled through state conversion; the Jacobian matrix updating and cost functioncalculating module calculates a Jacobian matrix by adopting a mode of combining analytic differential and automatic differential and calculates a cost function at the same time. BA calculation energyconsumption can be greatly reduced.

Description

technical field [0001] The invention relates to the technical field of hardware accelerators, in particular to a Zynq FPGA-based beam adjustment hardware accelerator. Background technique [0002] The bundle adjustment method (Bundle Adjustment, BA) obtains the optimal visual reconstruction effect by optimizing the camera parameters and the position of the three-dimensional space feature points at the same time. Optimum means to find the optimal parameter estimate by minimizing the quantization model, and simultaneous optimization of the two means that both the three-dimensional space feature points and the parameters of the camera are optimal. It is widely used in the reconstruction and simultaneous localization and mapping (SLAM) of 3D stereoscopic scenes. The goal of BA is to find optimal parameter estimates, i.e. point locations and camera parameters in 3D space, by minimizing the reprojection error. The error is defined as the L2 norm of the difference between the obs...

AI Technical Summary

Problems solved by technology

[0004] But at present, whether it is in the local BA calculation of the robot in SLAM or in the offline SfM visual reconstruction application, performance and power consumption are still the bottlenecks affecting BA applications.

Images