Abstract:
Did you know that India is now the home to world's largest number of blind people? 41% of
the blind population are from India. And one of the real security concerns of visually impaired
persons is that whether they get collided to the stray animals on roads and streets or whether
the stray animals suddenly bump into their path and this may result in some accidents on roads
which may not be healthy for both. Cows are the queens of Indian streets, and we could find
them almost everywhere. And same goes with dogs. So we wanted a system that would alert
the blind people of any situation like this and help them avoid it. This project is a part of
the Mobility Assistance for Visually Impaired (MAVI) project and the task is to help visually
impaired navigate their way in city conditions. The person will be wearing a camera (on chest
or shoulder). The MAVI system would be able to find the location of the wearer navigate him
to a destination, know the surface conditions, read signs and scene text, detecting people and
animals around etc.The aim of this project is to propose a system that would detect the stray animals in a frame.Since detecting animals is a broader task, this part of the project aims to detect only the cows in Delhi streets and then we will analyze the test results and proceed further towards the bigger task. We have used Histogram of Oriented Gradients (HOG) features of images which has been used in Computer Vision for extracting essential features of objects and SVM as our machine learning classifier. As a first attempt, we focus on the problem of face and side body detection of cows. The main goal of this project in this semester is to determine the accuracy that HOG+SVM provides in cow detection , analyze it and whether it can be used for our bigger goal i.e. of detecting stray animals accurately in Delhi scenes. In the next semester we would be looking at other options of recognising these animals.