Object detection for an autonomous mobile robot based on deep neural networks

Naing Min Tun, Gavrilov Alexander Igorevich, Pyae Phyo Paing, Nyan Linn Tun, Thet Aung Thu

Object detection is the most important visual task for an autonomous mobile robot. Computer vision, which includes images and videos, can serve as a cheaper sensor for detecting objects than others. In this paper, we used a small object dataset and conducted end-to-end own multiclass object detection using deep neural networks. We used a pre-trained VG 16 model using two new fully-connected subnets, not only for feature extraction, but also for the calculations of bounding box and object classification. Finally, the effectiveness of the proposed model was evaluated on test images using the accuracy measurement metric (mAP).

Keywords:object detection, convolutional neural networks, deep learning.