Keyframe Extraction for Salient Human Activity Recognition Using a Canny Edge Detector

• Published in: SN computer science Vol. 6; no. 4; p. 380
• Main Authors: Pandey, Madhuri, Mishra, Richa, Khare, Ashish, Rathore, Vandana
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 15.04.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Complexity; Computer Imaging; Computer Science; Computer Systems Organization and Communication Networks; Computing time; Data Structures and Information Theory; Datasets; Deep learning; Frames (data processing); Human activity recognition; Image processing; Information Systems and Communication Service; Literature reviews; Machine learning; Methods; Original Research; Pattern Recognition and Graphics; Research Advancements in Intelligent Computing; Sensors; Software Engineering/Programming and Operating Systems; Subject specialists; Video; Vision; Wavelet transforms; Deep learning; Canny edge detector-based; 2DCNN; Vision-based HAR; Keyframe
• ISSN: 2661-8907; 2662-995X; 2661-8907
• DOI: 10.1007/s42979-025-03911-8

Technological advancements in surveillance cameras require substantial storage spaces to store human actions. Processing large video frames is complex and requires a lot of computational time to recognize human activity. This paper proposes a canny edge detector-based keyframe extraction technique for human activity recognition to reduce the complexity and computational time of video processing. The proposed technique captures edge information to identify prominent keyframes by discarding redundant and similar frames. The extracted keyframes are then fed to a 2D CNN deep learning (DL) model to efficiently recognize human activity in multi-view and complex environments. To evaluate the efficacy of the proposed approach, we have experimented with two publicly available datasets, namely IXMAS and HMDB51, and achieved an accuracy of 97.69% and 99.53%, respectively. Moreover, the proposed work outperforms the existing state-of-the-art approaches.