Segmentation, Detection and Tracking of Video Objects for Interactive Video Editing Application

Abstract

Artificial intelligence applications have become important in various aspects of our lives, including the field of image processing. One prominent area in this regard is the detection and tracking of objects in videos, which has numerous application examples. Objects present in visual media elements are considered valuable data, particularly by the advertising sector. This study aims to detect objects from video images, segment these objects in the images, and track them seamlessly across consecutive frames. This enables the creation of interactive areas, positioning of multimedia content, overlaying visual layers such as advertisements onto objects, and covering the texture of the object area within the commercial application, Cinema8, a video editing application. To achieve this, the YOLO architecture is utilized to identify the classes of objects in the images and detect bounding boxes for these objects. Based on these bounding boxes, a plugin infrastructure is developed to edit multimedia images in a commercial product, offering users the ability to generate interactive videos.