Moving Object Detection and Tracking by Event Frame from Neuromorphic Vision Sensors

Area of Science:

• Computer Vision
• Robotics
• Neuromorphic Engineering

Background:

• Standard cameras struggle with fast motion and illumination changes, impacting object detection and tracking accuracy.
• Event cameras offer a low-latency, high-dynamic-range alternative by capturing scene changes asynchronously.
• Existing methods often face challenges with missed detections and inaccurate distance measurements in dynamic environments.

Purpose of the Study:

• To develop an improved system for moving object detection and tracking using bio-inspired event cameras.
• To enhance the accuracy of object detection by combining event frames with standard camera frames.
• To refine object tracking and distance estimation using event-based data and advanced algorithms.

Main Methods:

• A hybrid object detection approach combining event frames (probability-based) and standard frames (color-based).
• A detection-based object tracking method utilizing event frames and an improved kernel correlation filter to minimize missed detections.
• A distance measurement technique employing event frame-based tracking and similar triangle theory for precise object-to-camera distance estimation.

Main Results:

• The proposed methods demonstrate significant effectiveness in moving object detection and tracking tasks.
• The combined frame approach improved detection reliability under adverse conditions.
• Enhanced accuracy in tracking and distance estimation was achieved using the event-based system.

Conclusions:

• Event cameras provide a robust solution for object detection and tracking, overcoming limitations of traditional cameras.
• The developed hybrid detection and event-based tracking system offers superior performance in dynamic visual scenes.
• This research contributes to advancements in real-time computer vision applications requiring reliable motion analysis.