摘要
为了快速、精准地获取视觉测量系统中的位移信息,研究分析了基于模板匹配的全局搜索算法、三步搜索算法、新三步搜索算法,以及菱形搜索算法对位移信息的提取精度和速度的影响,并比较分析了曲面拟合法和梯度法两种亚像素定位算法的性能。结果表明:四种位移提取算法的计算精度一样,均在[-0.5像素,0.5像素]之间;全局搜索算法最为耗时;三步搜索算法适于较大位移的提取;新三步搜索算法具有自适应性,相对于三步搜索算法,更适于小位移的运动估计;对于小位移为主的运动估计,菱形搜索算法计算效率最高,最为合适;在所选用位移提取算法的基础上,曲面拟合法额外增加的计算时间远小于梯度法,更适用于实际视觉测量系统中位移的高精度提取。分析结果对基于视觉的实际运动估计具有一定的参考价值。
In order to obtain the displacement information in the vision measurement system quickly and accurately,the influence of the full search algorithm,the three step search algorithm,the new three step search algorithm,and the diamond search algorithm on the accuracy and speed of displacement extraction were compared and analyzed in this paper.The performance of the two sub-pixel localization algorithms,the surface fitting method and the gradient method were also compared.The results showed that the calculation errors of the four search algorithms are the same,within[-0.5 pixel,0.5 pixel];the full search algorithm is the most time-consuming;the three-step search algorithm is suitable for the extraction of large displacement;the new three step search algorithm is adaptive,compared with the three step search algorithm,it is more suitable for motion estimation of small displacement;for motion estimation of small displace ment,the diamond search algorithm is most suitable;based on the used displacement extraction algo rithm,the additionally added calculation time of the surface fitting method is much smaller,which is more suitable for high-precision displacement extraction in the actual vision measurement system.The comparative analysis results in this study have certain reference value for the algorithm selection for actual motion estimation.
作者
朱春丽
雷秀军
申远
ZHU Chun-li;LEI Xiu-jun;SHEN Yuan(Anhui Province Key Laboratory of Simulation and Design for Electronic Information System,Hefei Normal University,Hefei,230601,Anhui;Hefei Fuhuang Agile Device Co.Ltd.,Hefei,230031,Anhui)
出处
《蚌埠学院学报》
2021年第5期37-41,53,共6页
Journal of Bengbu University
基金
安徽省自然科学基金面上项目(1908085ME135)
安徽省科技重大专项(812126883015)。
关键词
视觉测量
模板匹配
快速位移提取
visual measurement
template matching
fast displacement extraction