Bandwidth Reduction and Object Detection Model in Close-circuit Camera Systems

The use of video streaming is a prevalent platform nowadays, such as providing online movie services, online learning services, and CCTV systems. Video streaming requires substantial bandwidth to maintain data clarity, leading to exponential bandwidth usage within internal networks. CCTV systems also use video streaming to transmit data within the network between the cameras and the server. The data transmission between these two devices leads to exponentially increased bandwidth usage compared to other devices. We proposed a method to reduce the bandwidth of video streaming between CCTV cameras and the server. Starting with a video compression model using YOLO for image detection to classify the video into subframes, combined with reducing the frame rate and constant rate factor when there is no target object, and increasing the frame rate and constant rate factor when there are target object, result in a bit rate reduction of up to 95.797%. During decompression, YOLO is used to divide the video into subframes, with the frame rate and constant rate factor adjusted uniformly across all frames. The Real-ESRGAN method is then applied to restore the quality of each subframe to match the original video. The Model performance is evaluated using YOLO, a state-of-art object detection algorithm. By comparing the video quality and traffic in different settings including, no compression, during compression, and after decompression.