Autonomous Indoor Pathfinding Using Neural Network in Complex Scenes

Navigation to a specific destination indoors can be a challenge due to different reasons such as visual impairment, unknown environments, etc. There has been much work done to solve this issue such as indoor positioning systems, navigation using sensors and even using a robotic guide. In this paper, a novel and straightforward method of path planning (including object avoidance) is presented as a way of navigating to a desired location within a complex environment. The system proposed uses the combination of depth information from an RGB-D camera and the object information from a Neural Network based object identification technique, to efficiently calculate and plan a path in real-time, to a pre-specified destination. Persons to be helped are identified using object detection, and the most practical path to the desired destination is calculated. The path information would be sent to the handheld device of the person being helped in the suitable form of interface, such as visual, audio, etc. The surveillance type nature of the system enables it to help multiple persons in the same area. The model was tested in a controlled environment with one individual person being guided to nearby specified locations. While the testing showed promising results, strong conclusions are yet to be made with the current system.