Farm Vehicle Following Distance Estimation Using Deep Learning and Monocular Camera Images

Area of Science:

• Computer Vision
• Automotive Safety
• Machine Learning

Background:

• Rear-end collisions remain a significant safety concern.
• Accurate estimation of following distance is crucial for advanced driver-assistance systems (ADAS).
• Existing methods often rely on complex sensor suites.

Purpose of the Study:

• To develop and evaluate a vision-only method for estimating the distance to a following vehicle.
• To compare the performance of linear regression, pinhole model, and artificial neural network (ANN) for this task.
• To provide a foundation for understanding driver behavior and reducing traffic accidents.

Main Methods:

• Instrumented vehicles with real-time kinematic (RTK) GPS for ground truth.
• Recorded video footage of following vehicles (sedan and pickup truck).
• Applied a deep-learning framework for vehicle detection and compared three distance estimation models (linear regression, pinhole, ANN).

Main Results:

• The artificial neural network (ANN) demonstrated superior performance in following distance estimation compared to linear regression and the pinhole model.
• RTK GPS data served as accurate ground truth for validating the estimation methods.
• The deep learning approach successfully utilized monocular vision data for distance calculation.

Conclusions:

• Vision-based distance estimation using deep learning is a viable approach for automotive safety.
• ANNs offer a promising solution for accurate following distance estimation from low-resolution monocular cameras.
• This research contributes to developing technologies for reducing rear-end collisions and understanding driver behavior.