Research on the perception method of tiny objects in low-light and wide-field video

Area of Science:

• Computer Vision
• Machine Learning
• Surveillance Technology

Background:

• Existing object trackers struggle with low-light conditions, leading to decreased accuracy in night surveillance.
• Challenges include high resolution, complex backgrounds, and target-background similarity in Hawk-Eye videos.
• Previous methods fail to effectively track small objects in low-light, wide-field scenarios.

Purpose of the Study:

• To develop an advanced tracker for robust tiny object tracking in low-light and night surveillance.
• To address the limitations of current trackers in challenging visual environments.
• To introduce a novel approach integrating difference constraint methods into correlation filters.

Main Methods:

• Integration of the difference constraint method into a correlation filter (CF) tracker.
• Generation of a change-aware mask using inter-frame difference information.
• Implementation of a dual regression model with an inter-frame difference constraint term for dual filter learning.

Main Results:

• The proposed method demonstrates superior accuracy in low-light and night surveillance tracking.
• It surpasses state-of-the-art trackers on a newly constructed benchmark of 41 Hawk-Eye sequences.
• Real-time performance of 27 frames per second (fps) achieved on a single CPU.

Conclusions:

• The novel CF tracker significantly advances tiny object tracking capabilities in challenging low-light surveillance.
• The method provides a robust solution for Hawk-Eye surveillance videos captured at night.
• The approach offers a substantial improvement over existing trackers in accuracy and real-time performance.