A learning-based approach for evaluating scene recognizability of a view

It is important to understand which view is better recognizing and reconstructing a scene for many robotic applications, especially in a cluttered environment, where objects interact and may occlude one another in all views. In this paper, we introduce a novel, learning-based approach to evaluate scene recognizability from a view based on the quality and quantity of recognized objects, the recognition uncertainty, and the background recognizability, rather than the visibility. Our study shows that increasing visibility does not guarantee better recognizability of objects. The introduced view evaluator can better characterize which view is more useful for the purpose of autonomous object recognition and scene reconstruction. The approach is validated through experiments, and the effects of many factors to scene recognizability are discussed based on the experimental results.