Real-Time Traffic Sign Recognition based AI Edge Computing
Authors/Creators
- 1. LARIT Laboratory, team Network, Telecommunication and intelligence, University Ibn Tofail, Faculty of Science, BP 242, Kenitra, Morocco
- 2. SIRC/LAGES, EHTP BP 8108, Casablanca, Morocco
Description
Abstract—Convolutional Neural Network (CNN) is a model
used in many disciplines to classify objects in video streaming,
clustering, pattern recognition, and prediction. With many systems
that deal with the classification problem, mobility is often
necessary. In this article, an implementation of Tiny-YOLOv3
(You Only Look Once) with CNN in Internet of Things (IoT)
devices is proposed to address the traffic sign classification
problem on the EDGE AI Sipeed MAIX platform. The main
feature of this module is the availability of the first RISC-V 64
AI module with K210-KPU inside, allowing high performance
with low power consumption. The proposed method uses the KMeans
algorithm to group our training set and find other boxes
of apparently excellent targets. The performances of the proposed
work are evaluated against the BTSD (Belgium Traffic Sign
Detection) dataset and the experiments show that the proposed
approach achieves an excellent compromise between speed and
precision. The system performs the detection recognition process
in approximately 112ms for each frame in (KPU) and 9 FPS in
the video stream.
Index Terms—Deep learning, Edge accelerator, Edge computing,
IOT, Tiny-YOLOv3, Traffic Sign Recognition, Edge AI,
Sipeed MAIX.