摘要
ORB(Oriented FAST and Rotated BRIEF)算法提取特征点时使用固定阈值,在不同场景或光照突变环境中会出现特征点匹配数目较少、匹配正确率较低、适应性较差的问题,针对这一情况提出了一种全局自适应阈值与局部自适应阈值结合的方式改进算法。首先根据图像的灰度分布判断亮度变化情况,针对整体亮度变化与局部亮度变化两种情况分别采用全局或半全局自适应阈值方案,并与局部自适应阈值相结合,通过建立评价函数得到阈值计算公式中自适应参数值,完成FAST(Features from Accelerated Segment Test)特征点提取,计算描述子,借助KNN(K-nearest Neighbor)算法完成特征点对筛选。实验结果表明,改进后的ORB算法在整体光照变化时特征匹配点对数量平均提升129.9%,正确率平均提升15.94%,局部光照变化时特征匹配点对数量平均提升149.4%,正确率平均提升3.53%。
The ORB(Oriented FAST and Rotated BRIEF) algorithm uses fixed threshold to extract feature points, which will lead to the problems of small number of feature points matching, low matching accuracy and poor adaptability in different scenes or abrupt illumination environments. In view of this situation, an improved algorithm based on the combination of global adaptive threshold and local adaptive threshold is proposed. First of all, the brightness change is judged according to the gray distribution of the image, and the global or semi-global adaptive threshold scheme is adopted for the overall luminance change and the local luminance change, respectively, and combined with the local adaptive threshold, the adaptive parameters in the threshold calculation formula are obtained by establishing the evaluation function, the FAST(Features from Accelerated Segment Test) feature points are extracted, and the descriptors are calculated. With the help of KNN(K-nearest Neighbor) algorithm to complete the feature point pair screening. The experimental results show that the improved ORB algorithm increases the average number of feature matching point pairs by 129.9%, and the average correct rate by 15.94%;The average number of feature matching point pairs by 149.4% and the average correct rate by 3.53% when the overall illumination changes.
作者
焦嵩鸣
钟宇飞
姚鑫
丁辉
白健鹏
JIAO Songming;ZHONG Yufei;YAO Xin;DING Hui;BAI Jianpeng(Department of Automation,North China Electric Power University,Baoding Hebei 071033,China)
出处
《激光杂志》
CAS
北大核心
2022年第6期85-90,共6页
Laser Journal
基金
河北省自然科学基金青年科学基金(No.F2021502008)。
关键词
光照强度突变
ORB算法
自适应阈值
评价函数
sudden illumination variations
ORB algorithm
adaptive threshold
evaluation function
