Traffic Sign Recognition Using Multi-Task Deep Learning for Self-Driving Vehicles.

Area of Science:

• Computer Vision and Machine Learning
• Intelligent Transportation Systems

Background:

• Advancements in driverless transport necessitate reliable traffic signal recognition.
• Computer vision and deep learning, particularly convolutional neural networks, offer powerful tools for this task.

Purpose of the Study:

• To precisely detect and recognize traffic signs and signals using computer vision and deep learning.
• To improve upon single-task learning models by implementing a multi-task learning approach.

Main Methods:

• Employed multi-task learning algorithms sharing convolutional layer parameters.
• Utilized pre-trained architectures including InceptionResNetV2 and DenseNet201.
• Integrated a region of interest module to enhance traffic sign extraction.

Main Results:

• Achieved a high accuracy of 99.07% after training the entire network.
• The multi-task learning model demonstrated superior performance compared to single-task models.
• Successfully validated the model in real-time on Riyadh highways.

Conclusions:

• The developed multi-task learning model with a region of interest module significantly enhances traffic signal detection accuracy.
• This approach is crucial for ensuring the safety and efficiency of future driverless transportation systems.