Multi-Object Detection Using Single Shot Multibox Detector MobileNet for People with Visual Impairments

This research aims to assist people with visual impairments live their daily lives using intelligent technology based on computer vision. In the early stages, this research focused on detecting and estimating the barrier distance for the blind where the object of the barrier is poles and the motorcycles. The input data is obtained from a smartphone camera hung around the respondent's neck using POVIE while walking towards the object for various distances. Dynamic movement from users is the challenge in this research: processing data from a moving camera. The data is divided into training and testing data. The detection methods used are Single Shot Multibox Detector (SSD) and Mobilenet. Meanwhile, the Pinhole Model algorithm is used to estimate the distance between the obstacle object and the position of the blind person. The output of this application is sound using the text-to-speech library on Android. The best motorcycle detection system achieved accuracy of 100%, and for pole, detection obtained an accuracy of 98.66%.