Object trajectory estimation in the dashcam videos with ego-motion.

Area of Science:

• Computer Vision
• Forensic Science
• Physics

Background:

• Object trajectory estimation from dashcam videos is crucial for traffic accident analysis.
• Existing 2D methods often fail due to neglecting camera ego-motion and lacking physical grounding.
• Launched objects follow parabolic trajectories influenced by gravity and camera movement.

Purpose of the Study:

• To develop a practical and physically accurate object trajectory estimation method for dashcam videos.
• To address limitations of conventional 2D methods by incorporating ego-motion.
• To provide a tool for forensic analysts in accident investigations.

Main Methods:

• Derived a projection-based model from a 3D parabolic trajectory, including dashcam longitudinal translation.
• Utilized least-squares optimization for trajectory parameter estimation.
• Validated the model on real-world forensic case studies.

Main Results:

• The proposed model demonstrated a close fit to observed data in forensic case studies.
• Achieved an average Root Mean Square Error of 6.64 pixels, indicating high accuracy.
• The method provides a practical and physically grounded approach to trajectory estimation.

Conclusions:

• The developed method offers a significant improvement over conventional techniques for object trajectory estimation.
• This physically grounded approach enhances the interpretation of object motion in traffic accident investigations.
• The tool can assist forensic analysts by providing reliable trajectory data.