Event-Based Color Segmentation With a High Dynamic Range Sensor

Area of Science:

• Neuromorphic Engineering
• Computer Vision
• Sensor Technology

Background:

• Traditional cameras have limitations in temporal resolution and dynamic range.
• Event-based cameras offer high temporal resolution and low latency.
• Color information is crucial for many computer vision tasks but challenging for event-based sensors.

Purpose of the Study:

• To introduce a color asynchronous neuromorphic event-based camera.
• To develop a methodology for processing color output from the sensor.
• To perform color segmentation and tracking at the sensor's native temporal resolution.

Main Methods:

• A color vision sensor prototype combining three Asynchronous Time-based Image Sensors (ATIS) was developed.
• A computationally inexpensive color processing algorithm was devised to leverage absolute color information from the ATIS.
• The system's performance was evaluated using both controlled indoor and uncontrolled outdoor scenes.

Main Results:

• The developed system enables color segmentation and tracking at microsecond temporal resolution.
• The color processing algorithm is computationally efficient, benefiting from asynchronous acquisition.
• Outdoor tracking achieved a mean error ranging from two to twenty pixels compared to ground truth.

Conclusions:

• The novel color event-based camera and processing methodology significantly advance real-time color vision capabilities.
• Low-level processing benefits substantially from asynchronous data acquisition and high temporal resolution.
• The system demonstrates potential for applications requiring high-speed, accurate color-based object detection and tracking.