A Novel Method for Automatic Detection of Basic Shapes on Whiteboard Images using Faster RCNN

In meetings, whiteboards are a useful tool for speakers to illustrate information, show diagrams and charts. After meetings, users always need to store the information that has been written, drawn on the board in the form of appropriate digitization as PowerPoint format. In this study, we first used pre-processing techniques to detect the frame of whiteboard from the whiteboard image and perform defatting. Next, we built the Faster RCNN model to detect three basic types of shapes, including circles, triangles and rectangles. Finally we use Line Segment Detector to detect the line segments. The research results show that our method with Faster RCNN model accurately detected the basic shapes on the whiteboard image with the Faster RCNN model, resulting in mAP@0.5 reaching 92.35% and the rate of detection of straight line with recall = 93.8%.