Sensors or electronic devices are used for remote monitoring. Real-time traffic, forest,
military, commerce, and medical remote monitoring detects abnormalities. The computational
complexity of computer vision-based video processing systems is substantial. This research
develops a Slow-Fast Convolution Neural Network (SF-CNN) to detect and classify anomalous
behaviour in surveillance videos. The suggested CNN architecture automatically learns video
frames and selects the best object behavior attributes from a wide sample of films. SF-CNN
learns slowly or quickly. When the frame rate is low, slow learning is enabled, and when high,
rapid learning is enabled. The input video teaches both both spatial and temporal information.
Actions identify humans, vehicles, and animals. All movies contain normal and aberrant
activities in different circumstances. The SF-CNN architecture solves numerous limitations
anomalous motions end-to-end. SF-CNN performance is tested on many benchmark datasets.
The proposed method had 99.6% accuracy, higher than prior methods.