Enhancing road safety with a power-aware approach to traffic sign detection and recognition in driving assistance systems based on machine learning using optimization techniques

OBJECTIVES: This study aims to improve road safety by creating a more effective deep learning model for traffic sign detection and recognition in driving assistance systems. The model, called Optimized Convo Sequential Recurrent Neuro Ant Hierarchical Net (OCSRN-AHN), combines convolutional and sequential methods with optimization strategies. This enhances recognition accuracy and processing speed in different environmental conditions. METHODS: A Kaggle dataset of 58 traffic sign classes was used to train and evaluate the model. The OCSRN-AHN model was trained using Python-based deep learning tools. This involved combining Convolutional Neural Network (CNN) and Sequential Recurrent Neural Network (SRNN) with Ant Hierarchical Network (AHN) optimization to fine-tune parameters. We evaluated model performance using metrics such as accuracy, precision, recall, F1-score, mean average precision (mAP), and mAP@0.5 (mean average precision at an intersection over a union threshold of 0.5). RESULTS: The OCSRN-AHN model achieved 99% accuracy, 99% precision, 99% recall, and an F1-score of 98%. Additionally, mAP@0.5 reached 98%, and the overall mAP was 98%. It performed better than existing models, including PFANet, YOLO v3, and YOLO v7t, in both detection accuracy and robustness for small, medium, and large traffic signs. CONCLUSIONS: The OCSRN-AHN framework not only achieved superior accuracy but also exhibited a more robust detection performance than all existing methods. Besides the higher detection accuracy, the proposed OCSRN-AHN has a reduced model complexity (3.2 M parameters), lower computational complexity (4.8 FLOPs), shorter inference time (21.4 ms), and lower estimated power consumption (0.39 J per image). These findings suggest the appropriateness of its use in real-time scenarios.