A Wearable Assistive Device for Safe Travel Using Transfer Learning and IoT for Visually Impaired People

There are a number of discomforts faced by visually impaired people every day in both indoor and outdoor surroundings. Assistive Technology for people with visual disabilities plays a vital role in their Independent living. Various systems have been developed to help them to live a better life even with the low or no vision. Visual Mobility plays a vital role in their Independent living such as arrival of buses, recognizing the Route Number from number plates, finding the doorstep in a train etc. This paper proposes an efficient approach for recognizing the Route details from a bus that helps in easy commuting in bus stations for the visually impaired. The You Only Look Once (YOLO v3) model can be used to detect the real time object bus and also segment the bus name board i.e. the Region of Interest (ROI) by transfer learning it with a custom dataset. To further improve the pattern recognition accuracy the numbers of anchoring boxes were increased from 3 to 5 in all the strides which provided more precise results. The route details present in the segmented ROI is converted into text using Tesseract tool that uses LSTM (Long Short Term Memory) engine for producing text from recognized characters in image and then regular expression is used to filter out only the bus number (alphanumeric or numeric values). The obtained bus number is converted as voice output (using text to speech library e-speak) along with bus details based on the bus number extracted from the number plate. An approach to detect the door step at railway compartments has also been proposed for visual mobility. During the travel, the device can be further used to know the current location and the distance remaining to reach the destination.