Innovative Strategies for Efficient Traffic Sign Prediction Using Dense-CNNet Classification

With the rise of intelligent driving technologies, traffic sign identification has gained growing research interest. Traditional methods relying on shape or color features can recognize broad categories such as warnings and prohibitions but suffer from limited accuracy and are sensitive to light and weather changes. Though computationally efficient and suitable for real-time use, these methods offer limited recognition capabilities. This study proposes a robust solution combining Non-Local Means (NLM) for noise reduction, a modified UNet for detailed feature extraction, and DenseCNN for accurate classification. To further enhance performance, Rock Hyrax Swarm Optimization (RHSO) is applied for hyperparameter tuning. This synergistic integration significantly boosts detection accuracy. Evaluations on standard datasets show that the proposed method outperforms state-of-the-art techniques, achieving 99.64% accuracy on the TT100K dataset and 99.6% on the GTSDB dataset, establishing its effectiveness and reliability in varied real-world conditions.