Digital Signal Processing Reference
In-Depth Information
Camera Calibration of the Stereo-Vision System
with Large Field of View Based on Parallel Particle
Swarm Optimization
Guangming Zhang, Yuming Chen, and Yuhao Yuan
School of Automation & Electrical Engineering,
Nanjing University of Technology, Nanjing, 211816
Abstract.
Binocular vision calibration with Large Field of View (LFV) is a
multi-parameter nonlinear complex problem. This paper
utiliz
es particle
swarm (PSO) algorithm to the multi-parameters optimization of binocular vi-
sion calibration. Considering radial distortion and tangential distortion of the
camera model, this paper extends the standard particle Swarm Optimization
(PSO) and proposes Parallel PSO Algorithmsthis method improves the abili-
ty of searching the global optimal solution and the searching speed. Taking into
account camera radial and tangential distortion, we achieve the global optimum
of interior and exterior camera's parameters with this method in a large field of
binocular vision system calibration. Experiments show that this method is
simple, high precision, good stability. The interior and exterior camera's para-
meters are searched the global optimum solution, the result of the experiment
indicated, the method is simple with high accuracy and good stability.
Keywords:
Calibration of stereo-vision System, Parallel Particle Swarm Opti-
mization, Large field of view.
1
Introduction
Binocular vision calibration is the most important aspect of stereo vision Inspection.
The calibration parameters of the camera model are divided into internal and external
parameters[1].According to the camera image model, binocular vision calibration is
divided into the linear and non-linear calibration. The linear calibration idealizes the
camera model, and solves the parameters with the linear equations, it is simple and
fast, but its accuracy is poor. Considering the distortion parameters, non-linear cali-
bration presents a nonlinear optimization method, such as the Levenberg_Marquardt
(LM) optimization algorithm, Roger Tsai RAC two-step method [2]. However the
optimized parameters are too many, the conventional algorithms compute complexity
in the binocular stereo vision. The parameters must be set to a more appropriate initial
value, so that it can get a better optimized solution [3]. With the increase of the para-
meters, the traditional algorithms often cannot find the optimal solution. The particle
swarm optimization algorithm is a viable method of nonlinear optimization algorithm,
it do not estimate the characteristics of the initial value, and achieves the good results
