Environment-Adaptive Object Detection Framework for Autonomous Mobile Robots.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Object detection is crucial for mobile robot missions.
• Deep learning models, particularly Convolutional Neural Networks (CNNs), excel at object detection.
• Environmental context significantly impacts model performance, leading to mission failure when unaddressed.

Purpose of the Study:

• To propose a systematic solution for the class imbalance problem in robot object detection.
• To develop an environment-context-aware, multi-model object detection framework.
• To improve the reliability and efficiency of mobile robots in diverse operational environments.

Main Methods:

• Training multiple object detection models tailored to specific environmental contexts.
• Implementing a framework that effectively utilizes these models based on the robot's current environment.
• Evaluating the proposed framework's performance against traditional single-model approaches.

Main Results:

• The multi-model framework significantly overcomes the class imbalance problem.
• Average object detection accuracy improved by 6.6%.
• CPU usage decreased by 45.49% and latency by over 60%.

Conclusions:

• The proposed environment-context-aware, multi-model object detection framework enhances robot mission success.
• This approach effectively mitigates performance degradation caused by class imbalance.
• Significant improvements in efficiency (reduced CPU usage and latency) alongside accuracy gains were demonstrated.