Comparative Study of Object Recognition Algorithms for Effective Electronic Travel Aids

Abstract

Object recognition involves detecting the probability of presence and class of one or more objects in a given image. In this paper we carried out a comparison between different state of art approaches and concluded which approach is more user-friendly, cost-effective and produces more accurate results for object recognition. We deployed various algorithms like You Only Look Once (YOLO), Single Shot Multibox Detector (SSD), Faster RCNN (Region Convolution Neural Network) on diverse Convolution neural network models (CNNs) namely Mobile Net, Inception, Res Net that act as base networks when used with detection algorithms. Subsequently, using varying development boards like Jetson Nano and Raspberry Pi 3 B+ the speed, latency and accuracy values were evaluated for different combinations of base networks and detection algorithms. © 2021, The Author(s), under exclusive license to Springer Nature Switzerland AG.