Intelligent Offline Multi Object Recognition Walking Stick for The Blind

Abstract

Vision is one of the most important characteristics of a human that aid their day to day activities. Loss of vision however affects the ability of humans to freely navigate their environment and recognized objects along their path. Existing object recognition systems for the blind are mostly cloud based and its perfor- mance depends on reliable internet access. This makes them unsuitable in places with unreliable internet. Therefore, in this paper, a multi-object recognition intelli- gent walking stick for the blind that is completely independent of the internet was developed. The system is divided into three units, detection, recognition and com- munication units. The detection unit make use of an ultrasonic sensor and a buzzer, for informing the user of an impending obstacle. The recognition system makes use of a camera for capturing images with Convolutionary Neural Network architecture and Mobile Network Single Shot Multi-Box Detector (MobileNet SSD) for detecting objects in images. The communication unit transmits the rec- ognised objects through voice to the user in English Language. The entire system was deployed in a Raspberry Pi microcontroller due to its processing power. The result obtained from testing of the device on the field showed that the recognition unit achieved an average sensitivity, specificity, precision and accuracy of 87.26%, 67.45%, 89.07%, 82.50% respectively. This shows that the system is reli- able and can be used in recognizing objects for the blind