Detecting the violence from video using computer vision under 2dspatio temporal representations
M. Srividhya, R. Ramyadevi, V. Vijaya Chamundeeswari · 2023
A current hot topic in computer vision is action recognition in videos, particularly for the detection of aggression. The complexity of the 2D + t data produced by the proliferation of videos by a security camera or television material is what makes this work interesting. Modern techniques learn using 3D neural network approaches, which require data to train to produce discriminating features. This research intends to address topics such as cutting-edge techniques for video violence detection, datasets for real-time video violence detection framework development, and discussion and identification of unresolved concerns in the given subject. To overcome these restrictions, we offer throughout this piece a technique for putting videos into categories for violence recognition with a traditional 2D CNN. The methods comprise two parts.: (I) From an input video, we first construct several 2D spatial-temporal representations, (II) According to speculation, the new representations will feed CNN&s;s test/training data. The methodology by combining the separate conclusions from the video&s;s various 2D spatial-temporal representations and the categorization decision is made. An experiment on publicly available datasets with violent video demonstrates the usefulness of the suggested approach.