Adaptive Motion Detection for Enhanced Video Surveillance

The field of video surveillance demands precision and operates within strict constraints. In particular, object detection models must strike a delicate balance between accuracy and efficiency to run effectively on edge devices with limited resources. An effective approach to enhancing model performance in dynamic scenarios involves the integration of motion detection techniques. This paper introduces an innovative motion detection method featuring several novel mechanisms. Notably, the proposed technique incorporates an adaptive learning rate, determined by the intensity of motion within the scene, and an adaptive threshold, based on the mean intensity of each frame. These adaptations render the algorithm highly responsive to variations in motion and lighting conditions. Furthermore, the algorithm employs HSV color space to detect shadows, effectively distinguishing them from actual objects, and a simple Gaussian Blur method for background model initialization. This approach yields an algorithm that is both robust and efficient, a critical requirement for real-time performance on resource-constrained edge devices deployed in dynamic environments.