A Target Detection System for Mobile Robot Based On Single Shot Multibox Detector Neural Network

Abstract

Recent years, neural networks are used widely in the field of image processing. Neural networks have made great achievements in image classification, target detection, and semantic segmentation since 2012. In this paper, a target detection system for mobile robot is proposed based on the Single Shot Multibox Detector (SSD) neural network. The SSD neural network is trained to learn the features of input images using labeled dataset. The network model will converge using gradient descent algorithm during training. Then, the network becomes robust enough and is able to locate targets in images. In our research, we find that the performance of network is so perfect that it can hardly be done using traditional image processing algorithms. Especially, SSD network is most suitable for our system. The SSD network shows great adaptability, with 76.2% mAP (mean average precision) in the test dataset during our experiments. Detection algorithm based on SSD network has a great performance during experiments.