A hybrid filter based image classification framework for real-time anomaly detection video databases

Human anomaly detection has been one of the most promising fields of study in the last few years. Auto-detection of multi-class human anomalies will make it easier to understand more complicated actions and their variations. It s because there are so many features and training images that most multi-class anomaly detection models don t need to deal with noise elimination or figure out how to separate features. These models, on the other hand, use only a few features to look for multi-class anomalies. There are more and more different types of human anomalies. It takes a lot of memory and time to find the multi-class anomaly because it takes a lot of memory and time. In order to improve the process of detecting multiple-class human anomalies, a hybrid multiple feature extraction method is proposed to find the important multiple features in the motion vectors for the classification problem. Non-linear SVM classification is used to make a hybrid convolution neural network framework even better. Using experiments, it was found that the proposed model has a better human anomaly detection rate than the traditional multi-class segmentation models that have been around for a long time.