Generic Dynamic Environment Perception Using Smart Mobile Devices.

Area of Science:

• Computer Vision
• Robotics
• Automotive Safety

Background:

• Driving environments are complex, requiring advanced driver-assistance systems (ADAS) for safety.
• Monocular vision systems offer advantages for mobile devices due to embedded cameras and increasing computational power.
• Existing mobile ADAS often have limitations in handling diverse obstacle scenarios.

Purpose of the Study:

• To develop a generic, shape-independent, real-time obstacle detection technique for mobile devices.
• To create a robust system capable of operating in complex urban traffic environments.
• To achieve measurement accuracy comparable to established stereovision systems.

Main Methods:

• Utilizes a particle-based occupancy grid for dynamic, free-form 3D environmental representation.
• Processes real-time camera images, removing perspective effects to create a bird's-eye view.
• Segments the bird's-eye view to identify candidate obstacle areas for updating the occupancy grid.
• Groups tracked cells into cuboid representations of obstacles, including position, size, orientation, and speed.

Main Results:

• The system reliably detects most obstacles in urban traffic scenarios.
• Achieves measurement accuracy comparable to stereovision systems.
• Demonstrates effectiveness for generic, shape-independent obstacle detection.

Conclusions:

• The proposed particle-based occupancy grid technique offers an effective solution for real-time obstacle detection on mobile devices.
• The system enhances traffic safety by providing reliable obstacle information.
• It presents a viable and accurate alternative to more complex sensing systems like stereovision.