Moving Object Detection Based on Optical Flow Estimation and a Gaussian Mixture Model for Advanced Driver Assistance...

Area of Science:

• Computer Vision
• Embedded Systems Engineering
• Automotive Technology

Background:

• Traditional moving object detection (MOD) methods struggle in dynamic environments with moving cameras.
• Ego-motion in advanced driver assistance systems (ADAS) introduces complex motion patterns, hindering object and background classification.
• Existing MOD systems lack efficiency and robustness for real-world, non-stationary scenarios.

Purpose of the Study:

• To develop an efficient moving object detection (MOD) algorithm capable of handling moving camera scenarios.
• To design and implement a hardware solution for real-time processing of the proposed MOD algorithm.
• To evaluate the performance and efficiency of the hardware-accelerated MOD system.

Main Methods:

• Proposed an efficient moving object detection (MOD) algorithm tailored for dynamic camera motion.
• Designed the MOD algorithm using Hardware Description Language (HDL) for hardware implementation.
• Evaluated real-time performance using a Field-Programmable Gate Array (FPGA) based test system.

Main Results:

• The proposed MOD algorithm demonstrated superior detection performance compared to existing systems.
• The hardware implementation achieved real-time processing at 30 frames per second (fps) at 200 MHz.
• The design utilized 13.2K logic slices, 104 DSP48s, and 163 BRAM, indicating efficient resource utilization.

Conclusions:

• The developed efficient MOD algorithm effectively addresses challenges posed by moving camera environments.
• The FPGA-based hardware implementation provides a viable solution for real-time object detection in ADAS.
• The system offers improved detection accuracy and real-time capabilities for safety-critical automotive applications.