A lightweight network for traffic sign recognition based on multi-scale feature and attention mechanism

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transportation Systems

Background:

• Traffic sign recognition is crucial for automated driving and intelligent transportation systems.
• Existing methods often require significant computational resources.
• Automated detection and recognition of traffic signs enhance road safety and traffic efficiency.

Purpose of the Study:

• To develop a lightweight and efficient model for traffic sign recognition.
• To improve the accuracy and robustness of traffic sign detection systems.
• To propose a novel convolutional neural network architecture named ConvNeSe.

Main Methods:

• Utilizing Depthwise Separable Convolution and Inverted Residuals for feature extraction.
• Implementing multi-scale feature extraction and fusion techniques.
• Incorporating Squeeze and Excitation (SE) Blocks to enhance feature representation.

Main Results:

• Achieved a high accuracy of 99.85% on the German Traffic Sign Recognition Benchmark Database (GTSRB).
• Demonstrated robust performance through ablation experiments.
• The proposed ConvNeSe model is lightweight and computationally efficient.

Conclusions:

• ConvNeSe offers a highly accurate and robust solution for traffic sign recognition.
• The model's lightweight design makes it suitable for real-time intelligent transportation applications.
• Further research can explore deployment in diverse environmental conditions and advanced traffic scenarios.