Mobile Robot Object Recognition in The Internet of Things based on Fog Computing
Meixia Fu, Songlin Sun, Kaili Ni, Xiaoying Hou · 2019
Mobile robot object recognition has attracted significant attention in the internet of things recently, in which there are many challenging tasks, such as the objects, the communication networks and the computer system. It still needs a large of computation, communication and storage capability for the whole system. In this paper, we propose a scheme of mobile robot object recognition in IOT and use edge nodes to process the data from robot vision instead of cloud computing. Besides, we adopt YOLOv3 as the main algorithm in the edge nodes to process the video data. The video from the camera on the robot is transmitted to the fog node by a Wi-Fi router. The advantages of using edge nodes for computation local robot clusters are reliability, real-time capability and flexibility compared with the cloud. In our experiment, we efficiently train the computer model using COCO database deployed to GPU. The robot using the proposed model could recognize the objects in real-time. We achieve mAP of 31.0% and response time of 52ms that illustrates the state-of-art performance of the proposed framework.