Sensing and Detection of Traffic Signs Using CNNs: An Assessment on Their Performance.

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Traffic sign detection is crucial for autonomous driving and driver assistance systems.
• Various deep learning models like ResNet, Vgg, SqueezeNet, and DenseNet have been applied, but their comparative performance is debated.
• Fair comparison requires identical conditions, including database structure, training epochs, coding language, and system calls.

Purpose of the Study:

• To evaluate and compare the performance of multiple deep learning models for traffic sign detection.
• To identify the most effective models for traffic sign recognition under standardized conditions.
• To introduce novel model applications like AlexNet and XresNet 50 for traffic sign detection.

Main Methods:

• Implemented and trained various Convolutional Neural Network (CNN) models: ResNet (18, 34, 50), DenseNet (121, 169, 201), Vgg (16_bn, 19_bn), AlexNet, and SqueezeNet (1_0, 1_1).
• Ensured all models were trained under identical conditions, including the same database structure and number of epochs.
• Evaluated models based on key performance metrics: training loss, validation loss, accuracy, error rate, and processing time.

Main Results:

• Vgg 16_bn, Vgg19_bn, and AlexNet models exhibited superior performance compared to other evaluated models.
• These top-performing models showed favorable results in terms of accuracy and error rates.
• The experimental setup allowed for a valid comparison of different deep learning architectures for traffic sign detection.

Conclusions:

• Vgg 16_bn, Vgg19_bn, and AlexNet are recommended as the most effective deep learning models for traffic sign detection tasks.
• These models show significant potential for enhancing autonomous driving and driver safety systems.
• Further research can build upon these findings to optimize traffic sign recognition technologies.