| Summary: | 碩士 === 國立中興大學 === 電機工程學系所 === 102 === In recent years, automotive electronics market is an important positive development of the semiconductor industry, there are more and more electronic products in a car. As electronic technology advances, active safety equipment with collisions avoidance is becoming main stream. In this thesis, we propose a technique how to avoid collisions in the image sequences captured by stereo camera. Moving objects are the most dangerous ones in many situations, and we implement a system that can detect independent moving objects through a stereo video platform.
Based on our proposed algorithm, we use the disparity information and the motion information (optical flow) through video preprocessing which video captured by a stereo camera to compute the 3D-position and 3D-motion of feature points of image. In order to detect the moving objects, it is necessary to compute the motion of moving observer (ego-motion). Since the variation of two ego-motions is small in 1/30 second, we use the former ego-motion to replace the current ego-motion. Then the moving objects and non-moving ones may be differentiated. After finding out the moving objects, we can estimate the position and motion next 1/30 second.
Due to long software processing time, the thesis attempts to speed up the processing by implementing hardware. We use the CORDIC architecture to implement the coordinate translation and coordinate rotation, and reduce the rotation steps and the hardware cost of CORDIC module by applying CORDIC architecture of fixed-angle rotations.
The spec of the proposed hardware architecture is the HDTV1080p applications at 346MHZ in TSMC 90 nm process.
|
|---|
