Vehicle speed measurement method using monocular cameras

Area of Science:

• Computer Vision
• Intelligent Transportation Systems
• Machine Learning

Background:

• Traditional vehicle speed measurement methods often face challenges with accuracy and computational efficiency.
• Existing camera imaging models involve complex coordinate transformations and pixel de-distortion processes.
• Accurate real-time vehicle speed data is crucial for intelligent transportation and road safety.

Purpose of the Study:

• To develop a fast and accurate vehicle speed measurement method using a monocular camera.
• To simplify the camera imaging model and reduce computational complexity for positioning.
• To integrate advanced deep learning algorithms for precise multi-target speed estimation.

Main Methods:

• A new camera imaging model is proposed, optimizing calibration for variable focal lengths and simplifying coordinate transformations.
• A two-dimensional positioning algorithm is developed, reducing a 3D problem to a 2D model for enhanced efficiency.
• The method integrates You Only Look Once version 7 (YOLOv7) and Deep Simple Online and Realtime Tracking (DeepSORT) for optimized network performance.

Main Results:

• The proposed method achieves high accuracy in vehicle speed measurement, with minimum and average accuracies reaching 95.1% and 97.6% respectively.
• Significant advantages in both speed measurement accuracy and computational efficiency are demonstrated compared to existing methods.
• The frame-by-frame measurement conditions yield reliable and precise speed data.

Conclusions:

• The novel method offers a significant advancement in monocular camera-based vehicle speed measurement.
• The simplified imaging model and integrated deep learning algorithms contribute to superior accuracy and efficiency.
• This research is expected to greatly benefit intelligent transportation systems and road safety management.