Generating pseudo-3D representations from conventional maps

  image gallery  

  HapticMap application


Generation of a pseudo-3D representation of a conventional map in order to convert the encoded information into haptic-aural format so as to be perceivable by visually impaired people requires at an initial step to define the characteristics of the virtual map.

For our case we define that:
1) Streets are separated from buildings using larger height for buildings in the 3D representation
2) Names should lie inside the corresponding street
3) Cross-roads are identified using haptic texturing

The pseudo-3D map generation is constructed using simple operations like morphological filtering, state of the art algorithms for image processing and of-the-shelf systems for Optical Character Recognition (OCR) and Text To Speech (TTS) synthesis.

The input to the system is a conventional 2D map.
The output is a pseudo-3D representation of the map with indexing information regarding the names of the roads, i.e. buildings are represented with larger height than streets so that the difference can be easily perceived through a haptic device. The name of each street is available and is reproduced whenever the user crosses it. Moreover, special effects like haptic texturing and friction are utilized to distinguish cross-roads and special buildings like churches, etc. from the main map components that are the streets and the major buildings.

  thesis details  

presented by : Konstantinos Kostopoulos
Aristotle University Of Thessaloniki, Polytechnic Faculty : Department Of Electrical & Computer Engineering
Supervisor : Michael-Gerasimos Strintzis



In the following, the steps of the algorithm are briefly described:

Input: Still image of the 2D map

Generation of 3D model:
A.1) Morphological filtering 1: Apply Dilate filter, thresholding and dithering to the input map so as to discriminate the streets from the major buildings.
A.2) Recursively smooth the resulting map so as to delete fragmented small areas (Figure a).
A.3) Generate the 3D model (Figure d).
A.4) Identify cross-roads and special areas (e.g. churches, round-about, etc.).

Street names recognition:
A.5) Morphological filtering 2: Apply Erode filter , and decrease color depth to discriminate the image areas with street names from the rest of the map (Figure b).
A.6) Estimate the direction of the street names using linear least-squares. Match the names and their directions with the segmented map of step A.2.
A.7) Delete names that do not correspond to streets (Figure c).
A.8) Perform OCR to the remaining street names.

During run-time:
A.9) Collision detection and haptic rendering.
A.10) Simulate haptic texturing and friction for the cross-roads and the special areas of A.4.
A.11) Synthesize text from speech to provide aural feedback of the street names while the user navigates into the map using the haptic device.

Output: 3D representation of the 2D map including information about the names of the street which is displayed using text to speech synthesis.