基于改进ORB算法定点化研究与FPGA加速在智能驾驶中的应用

Application of Fixed-Point Research and FPGA Acceleration Based on Impoved ORB Algorithm in Automatic Driving

特征点检测是计算机视觉研究领域一项重要研究问题。虽然存在许多特征点检测方法,但特征点检测与匹配仍无法达到实时效果。改进ORB特征点检测算法,首先对图像进行直方图均衡,提高图像对比度,使图像更清晰。然后对图像进行高斯滤波,消除图像噪声。在原有ORB算法的基础上,在匹配点视差为5像素的左右邻域内矫正匹配点的位置。如果匹配点所处金字塔层数相差两层及以上,则特征点误匹配。并对该算法进行定点化研究。同时利用FPGA实现该算法的硬件加速,使之能实时地应用到智能驾驶系统中。

Feature point detection is an important research problem in computer vision research field. Although there are many related methods for feature point detection, feature point detection and matching still cannot achieve real-time effect. Makes improvement on the ORB algo- rithm, first of all, uses histogram equalization to improve the image contrast, then uses Gaussian Filter to eliminate the image noise. On the basis of the original ORB algorithm, the matching points are corrected in the left and right neighborhood which size is 5 pixels around the matching points. If the number of layers in Pyramid of matching points is two or more, then the feature points are wrongly matched. And does a fixed-point research about this algorithm. At the same time, the hardware acceleration of the algorithm is realized by using FPGA, so that it can he applied to the automatic driving system in real time.