Multi-exposure virtual photometer: A tool for evaluating the illumination robustness of feature detectors

Feature detection is a basic issue in computer vision, and the illumination robustness of the detector is an important evaluation indicator. However, no indicators that can directly and quantitatively evaluate the robustness of illumination have been found in the known evaluation methods. In this paper, we propose a novel evaluation method that can quantify the evaluation results. The proposed method constructs a multi-exposure virtual photometer, and finds the mapping relationship between feature points and photometric exposure based on the photometer. Further, experiments prove that the mapping relationship can be fitted by Gaussian function. Then, we designed a novel evaluation index based on the mapping relationship between features and photometric exposure. Extensive quantitative evaluation shows that our method can effectively reflect the illumination robustness of feature detectors. In particular, the quantitative display is more intuitive and facilitates the comparison of different detection methods.

Automated summary

In this paper, a novel evaluation method for feature detectors' illumination robustness is proposed, which can quantify the evaluation results effectively. By constructing a virtual photometer and finding the mapping relationship, the method has been proven to be successful in reflecting the illumination robustness of feature detectors.