Hybrid SLAM and object recognition on an embedded platform

Abstract

Simultaneous Localization and Mapping (SLAM) is a key component of modern autonomous robots. It provides a similar visualization and localization capability, that is easily perceived by a human, to an autonomous robot for it to function in an unfamiliar environment. However, a traditional SLAM system only creates a map that has no descriptive points of interest that may be useful for improved localization. In this project, a SLAM system is combined with a Text Detection and Recognition algorithm to provide a more descriptive visualization of the world. This composite system is designed and tested on the Jetson Xavier NX embedded platform. The ORB SLAM 2 algorithm was chosen for the SLAM system for its robustness and versatility. Then, the Efficient and Accurate Scene Text Detector (EAST) algorithm coupled with a Convolutional Recurrent Neural Network (CRNN) Scene Text Recognition was used to provide an efficient natural scene text detection and recognition.