3-dimensional object recognition using mirrors and deep learning

Abstract

Current object recognition model can only perceive an object from a single perspective. This may hamper its performance since some objects have a more distinct feature on one side and not the other. Therefore, past researchers rely on other methods to increase the performance from a single perspective viewpoint with additional equipment. However, this method is cost ineffective as it requires the usage of high- grade equipment and time to retrain the deep learning model. In this paper, we proposed the usage of a mirror configuration for an object recognition model to increase the chances of an object to be detected using existing trained object recognition model. It was found that the object recognition model with mirrors had an increased sensitivity of 9 %. Furthermore, the optimum mirror setup was established.