The importance of traffic signs cannot be overstated when it comes to road safety. The necessity for rapid and precise Traffic Sign classifier remains a challenge due to the complexity of traffic signs shapes and forms. In this paper, a real-time detector is presented for the German Traffic Sign Recognition Benchmark (GTSRB). GTSRB has 43 different classes with various shapes, forms, and colours. Their similarity is useful for object localisation but not for sign classification. In this article, a real-time detector for GTSRB is created using an upgraded compact YOLO-V4 Technique and implemented on the new NVIDIA Jetson Nano. To find and detect GTSRB pictures, a compact and efficient classifier is introduced. For the first time, this paper compares the detection and categorization of traffic signs using YOLO-V3 and 4, both regular and tiny.
Because most of real-time identification algorithms require a lot of processing power, the suggested compact classifier, which is based on the new YOLO-V4 Tiny, can recognize all 43 traffic signals with an average accuracy of 95.44% percent and a YOLO model size of just 9 MB. The GTSRB test dataset was used to validate this approach, which was then tested on the new Jetson Nano. In comparison to existing algorithms such as CNN, YOLO-V3, YOLO-V4, and Faster R-CNN, the suggested technique may successfully save more computational power and processing time.
Implementation of a Compact Traffic Signs Recognition System Using a New Squeezed YOLO
Int. J. ITS Res.
International Journal of Intelligent Transportation Systems Research ; 20 , 2 ; 466-482
2022-08-01
17 pages
Article (Journal)
Electronic Resource
English
YOLO , Traffic sign , Self-driving Car , Artificial intelligence , Real-time detection , Deep learning , NVIDIA Jetson Nano Engineering , Electrical Engineering , Automotive Engineering , Robotics and Automation , Computer Imaging, Vision, Pattern Recognition and Graphics , Civil Engineering , User Interfaces and Human Computer Interaction