Video classification is an important domain within computer vision. It categorizes video content into meaningful classes such as actions or emotional states. In relation to image classification, it has to deal with the problem of spatiotemporal dimensions as well as a large data volume that is present in a video. In this work we introduce a novel distance metric based video summarization technique which minimizes the size of the dataset while maintaining key temporal information. We performed our experiments using distance metrics such as norm of rows distance other than euclidean distance, norm of columns distance and eigenvalue based distance metrics. Our results show that the norm of rows distance performed well and provides a suitable balance between efficiency and accuracy. Our proposed method achieved significant accuracy of \(81.23\%\) , \(92.42\%\) , \(98.89\%\) and \(90.27\%\) on MMAC, UCF101, UCF11 and HMDB51 benchmark datasets. Our proposed technique continuously tracked temporal information while recalculating the distance from each key frame. Due to less computational demands, our approach performs effectively in real-world application scenarios.
Khan et al. (2026) studied this question.