Deep Learning-Enabled MANET Architecture for Real-Time Traffic Sign Detection

Abstract

In this paper, the real-time traffic sign detection of an Intelligent Transportation Systems (ITS) using the Mobile Ad Hoc Network (MANET) that is enabled by deep learning. The proposed solution (fusion of lightweight deep learning inference and decentralized MANET-based communication) overcomes the disadvantages of the centralized and cloud-dependent approaches (high latency and the need to run on stable infrastructure) to operate. The integration will allow the vehicles to sense and exchange the information about the traffic signs to the neighboring nodes in a low-latency decision-making process of the dynamic vehicular systems. Dataset of all traffic signs intensive enough to contain vast traffic sign bunch was collected using the German Traffic Sign Recognition Benchmark (GTSRB) and tailored roadside photos. More difficult real-world conditions used data augmentation method such as rotations, noise addition, brightness changes, and occlusion masking. This detection model had to be optimized in terms of pruning, quantization, and knowledge distillation so that, due to the combination of both optimization strategies, it will be applicable on an embedded system like NVIDIA Jetson Nano, and Raspberry Pi. The MANET layer was tested in the way of using the network simulation tools to analyze various routing protocols, movements of mobility and network density. Another testbed which simulated real world applications was also used to ensure that with realistic conditions, the detection and communication pipelines can be integrated. The suggested architecture is scalable, tolerant of node failures, and flexible to use different bandwidths, as well as to different mobility patterns. Combining perceptions based on deep learning with decentralized MANET communication, the work presents and achieves a practical and infrastructure-free method of cooperative vehicular sensing that may be used in mixed traffic of autonomous and human-driven vehicles, further road safety, and deployment of next-generation ITS.