Identifying and tracking pedestrians based on sensor fusion and motion stability predictions

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Sensor Fusion

Background:

• Development of Advanced Driver Assistance Systems (ADAS) is hindered by unreliable sensors.
• Intelligent systems require reliable sensors and real-time algorithms for accurate driver alerts.

Purpose of the Study:

• To develop and test an application for detecting and predicting pedestrian movement to prevent imminent collisions.
• To enhance ADAS safety through intelligent sensor fusion and predictive algorithms.

Main Methods:

• Implemented a hybrid fusion approach combining stereo camera vision and laser scanner for accurate obstacle positioning.
• Utilized intelligent algorithms for pedestrian identification based on leg positions (laser) and disparity maps (vision).
• Employed statistical validation gates and confidence regions for robust pedestrian tracking and prediction.

Main Results:

• The system accurately measured obstacle positions using the hybrid sensor fusion.
• Pedestrians were correctly identified using distinct algorithms for each sensor subsystem.
• Successful experimental testing demonstrated effective tracking and prediction of pedestrian movement, including complex trajectories.

Conclusions:

• The developed intelligent sensor application successfully enhances pedestrian detection and prediction capabilities for ADAS.
• This approach offers a reliable solution for preventing vehicle-pedestrian collisions in real-world conditions.