CREATION: Computational constRained travEl aid for objecT detection in outdoor eNvironment

Abstract

In the era of technological advancements, where humans are leveraging it in every sector, it will be highly unfair for the visually impaired to be left off. Hence, this research of developing a compact computational constrained travel aid to help them easily tackle the daily practice is a necessity. As there are approximately 1.3 billion visually impaired people, having a low-cost solution for them is an important requirement. To match this condition, we have used raspberry pi 3 and pi 4 tagged with the pi camera and sonar sensor, all of which are budget-friendly and serves our purpose. This system is attached to a cane and notifies the user about the nature and range of object with the help of sonar sensor and images captured from the pi camera. We have used the lighter versions of You Only Look Once (YOLO) and Single Shot Multibox Detector (SSD) model that has been deployed on both raspberry pi 3 and pi 4. The performance of both versions of raspberry with different models has been tested in an outdoor scenario. It has been observed that raspberry pi 4 is twice faster than pi 3 and the overall performance of Tiny YOLOv3 (Model2) is best in comparison to other models on both pi 3 and pi 4. © 2019 IEEE.