Real-Time Rolling Door Detection and Classification for Security Application

Rolling Doors or shutters play an important role as a means of security and privacy of street shops in many countries. These shutters have flimsy locking mechanisms which makes them vulnerable and can be broken in easily. In this paper, we have implemented the detection and classification of rolling doors, intending to improve the surveillance system by detecting whether a particular rolling door is open during off times, leading to a possible break-in. The tool used for detection in this system is the Google Tensorflow Object Detection (GTOD) API [3]. This API enables us to train custom datasets on various latest cutting-edge models. The OpenCV library [2] provides real-time optimized computer vision applications which are used in various aspects of image processing for this system such as data augmentation, image grading, and application of the flood fill algorithm [1], which is being utilized as the main technique for classification in this system. In this system, OpenCV is used as the primary interface for transporting input frames to the detection model and displaying the final output frame.