Parallel implementation of 3-D image processing.

摘要

A structured light based single-direction 3-D navigation system consists of a projector, a camera, and an infrared light filter. The system serves as a navigation sensor to help robot to perform the task of grabbing an object in front of the robot. It uses a unique algorithm to calculate 3-D wold coordinates of each point viewed by the camera from the pixel information obtained from the camera. The 3-D information of the wold coordinates generated by this camera system can be further applied in a variety of different applications, such as industrial quality inspection, mobile robot navigation, robot remote control, etc. As a navigation system, fast, accurate, real-time response is very important. Through experiments, the computation time of the existing algorithm of this navigation system is relative long. To improve the system, optimization needs to be applied. The motivation of this research paper is to optimize this existing algorithm to have significant fast computation speed and maintain the accuracy. There are nine functions executed in the existing algorithm. Compared the processing time of each function, decoding is the one takes more than half of the entire program's computation time. Therefore, in order to speed up the entire system, optimizing the decoding function is a necessary step. The existing decoding algorithm in the single-directional 3-D navigation system is designed and implemented using CPU. The computation arithmetic is a sequential serial computing algorithm. Nowadays, GPU's (Graphics processing unit) unique parallel architecture starts to be introduced to a wide range of studies to speed up computation time. However, how to modify and redesign the serial computing algorithm to efficiently utilize the GPU's parallel architecture is a challenge. In order to optimize the single-directional 3-D navigation system with faster processing speed and still maintain the same accuracy, re-design the serial decoding algorithm is necessary. In this paper, we are going to present a parallel decoding algorithm using CUDA-based GPU to accelerate the existing serial decoding algorithm more than 2000 times.