Centralised and Decentralised Sensor Fusion-Based Emergency Brake Assist

Area of Science:

• Automotive Engineering
• Sensor Fusion
• Artificial Intelligence

Background:

• Advanced Driver Assistance Systems (ADAS) increasingly use multi-sensor architectures for improved reliability.
• Single-sensor systems face limitations in accuracy and consistency across diverse conditions.
• Data fusion from multiple sensors is crucial for robust environmental perception in automotive applications.

Purpose of the Study:

• To highlight the importance of efficient multi-sensor data fusion in ADAS.
• To propose and compare centralized and decentralized sensor fusion architectures for an Emergency Brake Assist (EBA) system.
• To evaluate the performance trade-offs between these fusion architectures.

Main Methods:

• Developed and analyzed centralized and decentralized sensor fusion architectures for EBA.
• Utilized Light Detection and Ranging (LiDAR) and camera sensors for data input.
• Evaluated system performance based on execution speed, accuracy, and computational cost.

Main Results:

• Both centralized and decentralized fusion methods achieved acceptable frame rates (~20 fps) on an Intel i5 Ubuntu system.
• Decentralized fusion demonstrated higher accuracy in EBA performance.
• Centralized fusion offered a higher frame rate and lower computational cost, albeit with less accuracy.

Conclusions:

• Decentralized sensor fusion provides superior accuracy for EBA but incurs a higher computational burden.
• Centralized sensor fusion offers a balance of speed and lower computational cost, suitable for certain ADAS applications.
• The choice between centralized and decentralized fusion depends on the specific requirements for accuracy, speed, and computational resources in ADAS.