What is Super Vision?
Super Vision enables not only information retrieval from the visible scene around the blind user, but can also provide super human vision for people with or without biological vision, through multi-modal, multi-sensory, AI and a separate powerful computing unit.
Below are few demo videos for different real-time Blind Assistant features such as:
Object-of-Interest detection and tracking
Family/Friend presence notification
Scene Description
Person Description
Obstacle detection and navigation.
* The AI system here is named “Bella” and is activated to receive voice command by calling her name. The command phrase ends by saying “Thanks”. Altenatively, a command can also be given via pressing a specific button on the side of the frame, in case users cannot or wish to not talk to the system.
* A “blind” user only experiences the audio in the videos. The point-of-view (POV) videos from the glasses camera, and the annotated texts are only displayed for the purpose of demonstration to the viewers to understand the situation, environment and challenges.
all demo videos are based on real recording and real-time processing
Object-of-Interest
detection and tracking
The “blind” user asks the system to find an item of interet, in this case a TV remote. The system starts searching as the user looks and/or walks around. It notifies the user as soon as the item is spotted with its relative direction with respect to the user and then guides the user to it via Spatial Audio.
The Spatial Audio guide is beeping sound with perceived direction being the location of the found item and beeping periodicity relative to the distance of the item to the user. The closer the user gets, the faster the sound beeps.
Family/Friends presence notification
In this feature, the system actively looks for faces in the sight of the camera and constantly compares them with a images database of the user’s friend and family, which must have already been created or imported from his/her phone. The dataase includes one or multiple casual images of people of interest for the user with their name.
As a familiar face similar to one in the user database is identified, the user is notified with the name of the person. This helps the blind users to be aware of the people in their surrounding for interaction and greeting just like an individual with normal vision.
Scene Description
The “blind” user asks the system to describe the scene he/she is looking at. The system gives a brief description of the scene with importnat and obvious items in it. The user can later ask for more detailed description if needed.
This feature can be used not only for a scene but also for a specific item that should be close to the camera and take most of the frame, in order to get details about that specific item.
This feature can be very helpful for detection and description of landmarks and touristic places. It can be useful even for non-blind users.
Person Description
Similar to Scene Description, but for Person. The user asks the system to describe the person potentially present in front of the user. The system replies back with the gender, age category (adult or child) and the clothing information including the decomposition of each clothing part in an isolated image, which then can be used for reverse image search across the internet for more details such as the brand, price and shopping store to buy from.
This is a useful feature for interaction with people that are typically met for the first time and to verify them before starting a chat, or situations where the user is interested in the details of other person appearance and clothing even for non-blind users. E.g. Finding and buying a T-shirt you are interested in without asking or having access to the model or info.
Obstacle detection and navigation
In this feature, the system actively looks for obstacle on the way of the walking user and notifies him/her in case an obstacle is detected. The distance is reported to the user via speech as well as a beeping sound.
The alternative safe escape direction is also identified and suggested to the user for diversion. A successful diversion from a potential obstacle is also reported to the user.
Any approaching person with its rough incoming direction is identified and reported to the user for situational awareness. Similar to White Canes or Guide Dogs, this feature helps the blind users to navigate safely through a route and to avoid collision with fixed or moving obstacles.