基于局部自适应核回归的仪表定位方法

Instrument Localization Method Based on Locally Adaptive Regression Kernels

计算机视觉技术不断发展,利用巡检机器人对钻井平台、水电工程等复杂工业环境下的各种仪表设备进行定期检查成为可能,然而这些功能的实现依赖仪表的精准定位。本文提出一种基于局部自适应核回归(Locally adaptive regression kernels,LARK)的方法进行仪表的快速定位。LARK算法无需训练,可以快速搜索感兴趣的视觉对象,并且不需要进行过多的预处理,提高了定位的效率。通过提取查询图像的显著特征,在目标图像中寻找所有可能相似的对象,然后用非极大值抑制法保留最强相似点,实现目标对象的定位。实验选用从不同角度拍摄的具有不同放缩比例的仪表图像作为实验所需数据。实验结果表明,该算法定位准确度高,可以很好地满足工业环境下仪表的定位要求。

Abstract： With the development of computer vision technology, it is possible to inspect the instruments regularly through the inspection robot in some complex industrial environments. However, the realiza- tion of the inspection depends on the precise localization of instruments. Here, a method based on locally adaptive regression kernels （LARK） is introduced to achieve quick localization of instruments. LARK can quickly search the visual object without training or too much preprocessing, which greatly improve the ef- ficiency of the instruments localization. Firstly, the salient features are extracted and analogous objects are searched and with the target image to find all possible similar matches. Then nonmaxima suppression is employed to localize the target object. The instruments images with different scaling taken from differ- ent angles are used in experiments. Experimental results suggest that the algorithm is accurate and can meet the requirements of instrument localization in industrial environments.