Object identification to assist visually challenged

Abstract

Our recent survey on problems faced by the visually challenged suggested self-reliant movement in urban spaces as a major challenge. In this paper, the authors propose a novel way of assisting the visually challenged to identify various public transport means and also help get onboard with little or no assistance from others. We use an integrated system of a mobile phone connected wirelessly via Bluetooth to Arduino controlled array of uniquely placed ultrasonic sensors complemented with vibro motor for haptic feedback. The system detects obstacles in all four directions and helps navigate through crowded spaces. We employ image-based recognition based on the visual information obtained from a mobile phone camera to detect vehicles like bus, car, truck, two-wheelers, auto-rickshaw (three-wheeler) as well as objects. The results are converted to audio feedback via the mobile device. We experiment with detectors like Multivariate Generalized Gaussian Mixture Model (MGGMM) based on features from Histogram of Oriented Gradients (HOG). Results indicate 96.89% accuracy.