Emerging Techniques in Vision-based Indoor Localization

As to the human computer interface for the visually impaired people, there are two models a vision-based indoor navigation system can be used. First, if a visually impaired person wants to know the current location, it calls the application of a smartphone or wearable device and finds out the current location. Second, the system automatically sends notification to the users if they reach a specific location even though the users havent realize they have arrived this zone. Indoor localization is to use one or more sensors such as cameras, magnetic sensors, IMUs, RGBD sensors, etc., to automatically determine the location of a robot or a person in real time in an indoor environment. This literature review mainly focuses on the indoor localization solutions with visual sensors, including omnidirectional cameras and RGB-D sensors, for assisting visually impaired people with performing indoor navigation tasks. Some of the papers surveyed may be originally for other applications (such as robot navigation), some even on localization in urban environments, but as long as the methods can be readily used for portable visual navigation, they will be included. The review includes the following four parts. Part 1 discusses the general approaches to achieving indoor localization, and in particular, a special attention is given to omnidirectional vision based localization. Part 2 discusses 2D-based localization solutions with omnidirectional vision as well as regular cameras. In particular, image-retrieval-based approaches are discussed. Part 3 discusses emerging solutions with 3D approaches. Special attention is paid to localization with images from the Web, with omnidirectional images, and with RGB-D images. Part 4 discusses the challenges in achieving accurate, robust and real time navigation using portable devices, and then introduces a few emerging technologies to solve the problems: mobile/cloud computing and GPU acceleration.