Transfer reinforcement learning-based road object detection in next generation IoT domain

Abstract

The landscape of fifth generation (5G) and beyond 5G (B5G)-enabled Internet of Things(IoT) is expected to seamlessly and ubiquitously connect everything, which includes 5G, cloud computing, artificial intelligence and other cutting-edge technologies to realize truly intelligent applications in smart cities. In this paper, we present an important key technology for smart city, which is a road target recognition algorithm for smart city applications and designs a set of corresponding programs to assist automatic drivers, pedestrians and visually impaired people in road safety, or to manage city infrastructure. The system can connect robots in cars, wearable devices and body area network in pedestrians or blind people. A target recognition algorithm based on scene fusion is designed to recognize the specific target in the road environment, and transfer reinforcement learning method is used to improve the accuracy and real-time performance of target recognition. The system provides them with travel assistance, identify dangerous or useful objects for them through high-performance target recognition services. It can collect the road visual scene data by road cameras and transmit it to edge devices for training model. The model is collaborated trained in the edge devices and aggregated by the cloud. Based on the transfer reinforcement learning method, the vision-based road target recognition has been implemented, and the accurate and reliable target recognition can be realized. Many details of experiments verify the effectiveness of our technology.