An Evaluation of RetinaNet on Indoor Object Detection for Blind and Visually Impaired Persons Assistance Navigation

摘要

Indoor object detection presents a computer vision task that deals with the detection of specific indoor classes. This task attracts a lot of attention, especially in the last few years. The strong interest related to this field can be explained by the big importance of this task for indoor assistance navigation for visually impaired people and also by the phenomenal development of the deep convolutional neural networks (Deep CNN). In this paper, an effort is made to perform a new indoor object detector using the deep convolutional neural network-based framework. The framework is built based on the deep convolutional neural network “RetinaNet”. Evaluation is done by using various backbones as ResNet, DenseNet, and VGGNet in order to improve detection performances and processing time. We obtained very encouraging results coming up to 84.61% mAP as detection precision.