Fully Convolutional Neural Network for Vehicle Speed and Emergency-Brake Prediction.

Area of Science:

• Computer Vision
• Machine Learning
• Autonomous Systems

Background:

• Accurate ego-vehicle state prediction is crucial for autonomous driving safety.
• Monocular camera-based methods are vital for driver assistance and accident prevention.
• Predicting vehicle velocity and braking events enhances road safety.

Purpose of the Study:

• To implement a convolutional neural network (CNN) for predicting vehicle velocity, braking, and emergency braking.
• To utilize sequential image data and velocity information for enhanced prediction accuracy.
• To contribute to the development of perception-based techniques in autonomous vehicles.

Main Methods:

• Developed a CNN model trained on sequences of 20 images and corresponding velocity data.
• Input data comprised image sequences from a moving vehicle in traffic scenarios.
• Evaluated model performance using regression and classification metrics.

Main Results:

• The CNN model demonstrated effectiveness in predicting vehicle velocity and braking events.
• Comparative analysis showed competitive or improved performance against existing recurrent neural network (RNN) methods.
• Achieved improved prediction accuracy for velocity, braking behavior, and emergency braking.

Conclusions:

• The proposed CNN model offers a significant contribution to improving road safety in autonomous vehicles.
• The research provides valuable insights for perception-based techniques in autonomous driving.
• Enhanced prediction capabilities are key to preventing speed-related accidents and ensuring safe navigation.