The starting point is the Innomake developed by Tec-Innovation, the intelligent shoe for detecting obstacles, which makes the personal mobility of visually impaired, blind and motor-impaired people in their everyday lives safer. The first product version, which is currently (1Q 2019) at the beginning of CE certification and is about to be launched on the market, uses its specially developed electronics based on ultrasonic sensors to detect obstacles of up to 4m while walking and warns of these in good time. This ultrasound-based technology can only indicate whether there is an obstacle in front of the user, but cannot determine which obstacle this is. Because of this, we have been working since 2016 together with Graz University of Technology on a 2D camera-based supplement to the first product version. Modern deep learning algorithms for the recognition of the image content can determine an obstacle-free area and also recognize and distinguish objects from foot-perspective. These algorithms trained on the PC can already be operated by us on a specially designed mobile system. The currently available isolated working systems - including the Innomake combining ultrasonic and camera technology - can only be improved by in-house development, which leads to a sluggish improvement of the recognition performance for all users. Since there are also no automated ways to report incorrect measurements by the customer, such incorrect measurements (e.g. an obstacle is not recognized in a certain situation) cannot be recorded systematically, and are therefore in practice usually a long, potentially dangerous companion. State-of-the-art deep learning algorithms, which are designed on the one hand to use large amounts of data sensibly and on the other hand to draw their own conclusions for their own improvement from these amounts of data, now offer a great opportunity in this area. If the learnings can be brought together from every path of each individual user, gaps in the recognition of situations could be continuously closed. The goal is to create a comprehensive network for secure mobility from which not only Innomake users, but all visually impaired people can benefit. Using the Innomake camera, GPS and the latest deep-learning image processing methods, the system is to be continuously improved through the everyday use of the user. Only in this way is it possible for an obstacle detection system to become secure, so that in future, systems will not only - as at present - remain a supplement to the over 70 year old cane.
|Effective start/end date||1/10/19 → 31/03/21|
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.