Performance evaluation of binary descriptors for mobile robots

Mobile robot technology has been one of the most promising topics recently. As the basis of mobile robots, object recognition has important research significance to perceive environments. Float-point descriptors like SIFT are widely used for object recognition, but need large storage and computational costs. To overcome the limitations of float-point descriptors applied to mobile robots, binary descriptors are put forward. Since the property of object recognition algorithms can be affected by the selection of descriptors, evaluation of different descriptors to compare performance of them is necessary. However, evaluation of up-to-date binary descriptors for mobile robots is lack. In this paper, an evaluation of several algorithms based on binary descriptors i.e. BRIEF, ORB, BRISK and FREAK on a mobile robot object recognition dataset is provided. To find an efficient descriptor for mobile robots, some typical performance metrics are used to analyze the results of evaluation. And an improved object recognition strategy is presented. It can enhance the performance of binary descriptors for object recognition of mobile robots. This paper can provide a practical reference for researchers in this field.