ESO-Det: An Efficient Small Object Detector for Real-Time UAV Perception.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Object detection in Unmanned Aerial Vehicle (UAV) imagery is crucial for various sensing applications.
• Challenges include detecting small objects with scale variations and background interference.
• Limited onboard computing power necessitates efficient algorithms for real-time processing.

Purpose of the Study:

• To propose an efficient object detection network, ESO-Det, for real-time UAV perception.
• To address the challenges of small object detection and computational constraints in aerial imagery.

Main Methods:

• Developed ESO-Det, featuring a Dense Cross-branch Complementary Module for integrating semantic and spatial information.
• Incorporated a Large-Kernel Context Integration Module to enhance multi-scale contextual aggregation.
• Utilized a Lightweight Selective Aggregation Module for efficient fusion of multi-scale features.

Main Results:

• ESO-Det demonstrated superior performance compared to existing methods in object detection tasks.
• The network maintained real-time processing capabilities, crucial for UAV applications.
• Achieved higher accuracy in identifying small objects within complex aerial scenes.

Conclusions:

• ESO-Det is a highly effective and efficient network for real-time object detection in UAV imagery.
• The proposed modules successfully address challenges related to small objects and computational efficiency.
• The method is well-suited for practical real-time UAV perception applications.