Traffic Sign Prediction System using YOLOv3 Model with Performance Analysis

Real-time and accurate recognition of traffic sign is very imperative to the autonomy of driving and intelligent transportation systems. In different environmental conditions, traditional methods cannot be very practical, which narrows down their practical applicability. To address all these issues, a potent video-based detection and classification of traffic signs scheme have been introduced in this paper. The first thing is the process of median filtering to smooth salt-and-pepper noise in frames. This is followed by the use of the region-based segmentation to pre-extract the possible sign-retentions regions in every frame. YOLOv3 is the deep learning model that is the central component of the system and that showed a reasonably high level in both speed and accuracy, as well as the dynamic of detecting in a real-time scenario. The YOLOv3 is trained with the vast amount of data to adapt to the mobile driving conditions. The fact that the system is accurate (96.7 %), has precision of 96.3 %, recall of 94.5 %, and F1-score 95.4 % is proved with confusion matrix measures. The combination of image enhancement, region mining, and deep learning-based classification makes it easy to implement this method in advanced driver-assistance systems (ADAS), which is valuable to real-time decision-making processes and all-time safety of autonomous vehicle usage.