"Serving individuals with special needs through research and innovation in affordable assistive technologies"
You are Here : Home > Research > Student Project



Automation of Tactile Graphics
Is to design an end-to-end system which, given an image, generates a vector graph which can be subsequently used to produce tactile graphics. The current domain of images are the IX standard Science & Mathematics Textbooks.

Nipun Gupta, Abhishek Bansal, Ankit Singh
Sign Board Detection Module for MAVI PROJECT
The project is aimed to provide mobility assistance to the visually impaired by providing them with the information on the pedestrian sign boards. The central idea is to detect the sign board and extract the information by employing Optical character recognition techniques and also by validating its correctness using GPS. Using the colour as the distinguishing parameter , along with the use of shape (an approximated rectangle)the preliminary stage of detection is accomplished . The further processing is done for the identification of text which is whit in colour .Once this id done the detection of the sign board is done .This is passed on to the OCR stage for the information extraction.Simultaneously the data is also sent to the server for the purposes of validation , updating the information etc., The project involves the use of computer vision and is currently developed in opencv.

Dedepeya Pappireddy, Hassen Basha Khanubhai. KK Yoosuf, Munib Fazal
Camera as an Assistive Device - Apps
Project includes using camera and computational abilities of the smart phone to aid blind users. Objective: -Image recognition framework using android smart phone -Framework is extensible -Processing structured documents Operations performed: -Front-back detection -Identifying the principle item from the camera frame -Cropping the background -Alignment of item along screen edges App work flow: -Specify context: This includes two operations that needs to be carried out by a sighted person. a. First, upload reference images of the document on the MoodStocks uploader for desktop. (frontside, frontside rotated 180 degrees, similar for backside) b. Second, click photo of the document using the app and specify the position of the fields by drawing rectangles around them. Context recognition: Context recognition is done using MoodStocks API. This is done to find which one of the saved contexts should be used to read the stored positions of the fields specified earlier. Alert user for backside of document (Front-Back detection). Upside-down detection is also done and taken care of without unnecessarily informing the user about it. Image capture and processing: a. Select color of the document to detect by tapping on b. Approx. contours with polygons (2% arc length) c. Ensure that the polygon is a quadrilateral, has all the document. d. Auto Capture when the sharpness value is above a e. Transform skewed image to rectangular shape and Text recognition (OCR): With the cropped photo of the document to be read (with background cropped out) and also the relative positions of the fields to be read, we perform OCR only within rectangular regions using Tesseract OCR for android. The identified text is then fed to Text-to-speech engine.

Jasmeet Singh & Nikhil Verma
Bus Route Portal
Visually impaired users have trouble locating where they are when travelling through a bus. So an app is being developed which would assist them which would tell where they are and other such allied information using GPS. The project is to design and implement a web portal which provides information regarding various bus routes in a given city. The web portal should be accessible to the visually impaired. Should provide necessary information to the android app "Travigator". Should provide multi language support, so that travellers who don't understand the local languages can use it too.

P. Sameer Chakravarthy