Detection of Cut Transition in Videos Using Optical Flow and Clustering
Nilamadhab Dash, Sasmita Kumari Nayak, Jhama Majumdhar · 2021 Asian Conference on Innovation in Technology (ASIANCON) · 2021
In our everyday lives, videos have become an immersive and interactive media of communications. The huge number of video makes it very hard to peruse and get the necessary data. Manual indexing and analyzing the contents of videos is time consuming process. The automatic detection of some events within the video is a clear option of this problem. The collection of frames or images from a single camera is referred to as a shot in a digital video sequence. The abrupt changes of shots from one scene to another are the determination of shot boundary. Shot detection is just one of the basic techniques for digital video analysis. The process of video shot detection is also considered as basic temporal segmentation which is needed as a pre-processing for video summary. The proposed research presents the detection of abrupt transition CUT with varied constrictions based on the Optical Flow Model and clustering techniques. We have used two different corner detectors FAST and ORB to determine Optical Flow from the sequence of Video. The characteristic of the flow determines the CUT transition from the video sequence. We then used the corner points as features and use clustering methods CLARANS and WARD to get the number of clusters, which eventually ascertain the number of CUT transition in the Video sequence. To verify the correctness of the method, we used two different features from corner points, location of the corner points and corner strength. Results obtain ascertain the number of shot transition. Finally we demonstrate our approach showing the conventional ELBOW method as conclusion of the procedures adopted. These algorithms have the higher efficiency to detect cuts in speedy scene changes in the video.