Nighttime Respiration Detection Based on Improved Canny Edge Detection

This paper presents a non-contact nighttime breathing detection method based on an improved Canny edge detection algorithm, which enables real-time and accurate detection of breathing frequency in nighttime environments. During video-based breathing monitoring, the breathing motion is subtle and difficult to observe, making it easily influenced by environmental noise, which leads to low accuracy when directly analyzing video signals to extract breathing information. To adapt to variations in the monitored subjects, this paper proposes an adaptive region of interest (ROI) strategy. Based on the detection results of the RetinaFace algorithm, the facial area of the monitored subject is first identified. If only the face is visible, the ROI is limited to the face. However, if the chest or abdomen is also exposed in the video, the ROI is extended to include the face, chest, and abdomen, ensuring the comprehensive capture of breathing movements. Within the selected ROI, an improved Canny edge detection algorithm is used to extract image edges. Compared to the traditional Canny algorithm, the improved version adapts the edge detection parameters, allowing it to better capture subtle grayscale variations in low-light environments, thereby enhancing the detection of weak breathing movements. To further improve detection accuracy, this paper also incorporates a false peak elimination algorithm to remove interference signals caused by slight movements of the face or chest, or environmental noise. As a result, the algorithm can more stably identify the true breathing frequency.