DyPERS: Dynamic Personal Enhanced Reality System

Nuria Oliver Tony Jebara Bernt Schiele Alex Pentland

nuria@microsoft.com jebara@media.mit.edu bernt@media.mit.edu sandy@media.mit.edu


DyPERS is a 'Dynamic Personal Enhanced Reality System' which uses augmented reality and computer vision to overlay video and audio clips relevant for the user on top on real real objects that the user is paying attention to. The system is wearable and adaptively learns an audio and video memory and what everyday objects to associate it with and to evoke or playback in the future.


DyPERS is a 'Dynamic Personal Enhanced Reality System' which uses augmented reality and computer vision to overlay video and audio clips relevant for the user on top on real real objects that the user is paying attention to. The user wears a HUD (Heads-Up Display) with a small mounted ELMO CCD QN401E color camera on board (as in the Stochasticks system) and a wireless microphone. A generic and trainable object recognition system processes images from the camera as the user turns his head to view an object of interest. It then automatically highlights important objects as previously specified by the user. The user shows the system objects of interest ahead of time and then will associate video and audio clips that the user records to those objects. DyPERS can be considered a videographic memory.

System Architecture

The three main components of DyPERS are:

  1. Audio visual memory, which accumulates personal memories and associates them to objects
  2. Generic trainable object recognition system using computer vision as input
  3. Wearable system with audio visual input/output capabilities and interface.

Audio Visual Memory

Audio and Video are recorded and played back in real time as the following images show:

Visual Learning

The generic object recognition system uses computer vision. It is invariant to scaling, translation, rotation, small lighting changes and deformations of the object so that it can usually be recognized in different situations (see figure below). Some of its features are:

  1. Learns user-relevant visual cues
  2. Builds a statistical representation of the objects
  3. Groups sample images into objects and several objects into a audio-visual association

Recognition of the tie


The system is full wearable but needs offline processing which is done via wireless links. Its important hardware components are:

  1. ELMO CCD QN401E color camera
  2. Wireless 3-button mouse
  3. Wireless microphone
  4. Glasstron heads-up display and headphones
  5. SGI O2
  6. Wavecom Jr transmitter/receiver units


The interface paradigm is Record and Associate: just by using two buttons of a wireless mouse the user selects when to record some video and audio in real time (see Figure below). A third button (garbage) is for negative feedback to signal to the system that the association it learned is incorrect and to delete it.


Some examples of objects or situations that DyPERS can recognize and augment are:

  1. Clock: Every time DyPERS recognizes a wall clock or the watch of the user it displays a video with the user's schedule for the day.
  2. Demo poster: Just by looking at a poster of a demo of the lab, DyPERS will show a short video and audio with the demo associated to the poster.
  3. Multilingual Teacher: The user could record the name of several objects in several languages. During recognition DyPERS would speak the name of the learnt objects in the different languages, teaching the user how to say them.
  4. Stuffed Animal: A story about the animal could be recorded in such a way that everytime the kid looks at the animal DyPERS triggers the story associated to it.
  5. Augmented Storyteller: The parents could associate the pictures on each page of their children's book with the story about them. Later the kid would listen to the story just by looking at the pages of the book.
  6. KeyPad Door Lock: By recording the keypad door combination with an image of the keypad, DyPERS would remind the user of the right combination when the user would look at the keypad.
  7. Business Card: A video and audio clip of our conversation with an important person would be associated with his or her business card. Then everytime the user would look at the business card the video and audio would appear and remind the user about whom the business card belongs to.
  8. Origami: DyPERS could teach the user how to create different origami objects by playing video and audio about how to make them.
  9. Specific Machinery: Instructions about how to change pieces of some appliance, or how to use it would be associated with the appliance itself and played back when looking at the object. For example, instructions on how to change the ink cartridges of the printer and the printer.
  10. CD Cover or Poster of a Movie: The CD cover of some music or the poster of a movie would be associated with some clips of the music contained on the CD or a small preview of the movie. When the user would look at the CD cover or at the poster DyPERS would play the audio and video associated with it, given to the user a good cue of what type of music or movie is.
  11. Blind People: Important objects could be associated with some audio information in such a way that DyPERS could describe what visually impaired persons are seeing in a personalized and private way.
  12. Medicine: The user could record the directions of the doctor about how to take a specific medicine with the box/containter of it. When necessary just by looking at the medicine DyPERS would play the instructions automatically to the user.
  13. Name/Logo of a specific store, associated with the nearest location of it, its schedule and which items the user would be interested in purchasing there.
  14. Art objects (images, paintings, sculptures) and some explanationsn about them, the location and other relevant information.

The output consists of superimposesd video and audio on the real video images.

The system is dynamic, personal and trainable.



  1. "DyPERS: Dynamic Personal Enhanced Reality System" . Bernt Schiele, Nuria Oliver, Tony Jebara and Alex Pentland. Intl. Conference on Computer Vision Systems, ICVS'99 Jan1999. Gran Canaria, Spain.
  2. "Sensory Augmented Computing: Wearing the Museum's Guide" . Bernt Schiele, Tony Jebara and Nuria Oliver. IEEE Micro Journal. 2001.
  3. "Stochasticks:Augmenting the Billiards Experience with Probabilistic Vision and Wearable Computers" . Tony Jebara, Cyrus Eyster, Josh Weaver, Thad Starner and Alex Pentland. Proc. of the First Intl. Symposium on Wearable Computers. Oct 1997. Cambridge, MA.
  4. "Object Recognition using multidimensional receptive field histograms." Bernt Schiele. PhD thesis. July 97. I.N.P. Grenoble. France