Indoor objects detection and recognition for an ICT mobility assistance of visually impaired people

摘要

Indoor object detection in real scene presents a challenging computer vision task; it is also a key component of an ICT autonomous displacement assistance of Visually Impaired People (VTP). To handle this challenge, a DCNN (Deep Convolutional Neural Networks) for indoor object detection and a new indoor dataset are proposed. The novel DCNN design is based on a pre-trained DCNN called YOLO v3. In order to train and test the proposed DCNN, a new dataset for indoor objects was created. The images of the new dataset present large variety of objects, of indoor illuminations and of indoor architectural structures potentially unsafe for a VIP independent mobility. The dataset contains about 8000 images and presents 16 indoor object categories. Experimental results prove the high performance of the proposed indoor object detection as its recognition rate (a mean average precision) is 73,19%.