REDA: A Real-Time Event-Detection Approach to Minimize IoT Visual Data Generation With Computation Efficiency

The Internet of Things (IoT) offers vast potential to enhance the quality of life, but the excessive visual data generated during environmental monitoring presents significant challenges. Existing visual data minimization methods struggle with real-time data reduction, often applying uniform minimization ratios to compress already generated data, which leads to high computational overhead and distortion. To address these limitations, this paper introduces REDA, a real-time event-driven approach for minimizing visual data generation. REDA employs an event estimation method that integrates motion and multi-scale object detection to reduce false alarms, missed detections, and computational costs. Additionally, it introduces an Optimal-IoU loss function to handle gradient challenges and applies contextual optical flow and filtering techniques to minimize data loss and distortion. Theoretical analysis and experimental results demonstrate that REDA achieves superior real-time data minimization and efficiency compared to existing state-of-the-art solutions.