Anomaly Detection in Video with Knowledge Distillation Radial Kernel Function Assisted Ensemble Model

One of the most well-known applications of computer vision is anomaly detection in videos, though it remains challenging due to its subjective and context-dependent nature.Traditional machine learning (ML) and deep learning (DL) approaches have been introduced but struggle with high error rates and inefficiency.To address this, the study proposes an ensemble learning model with an effective optimization mechanism.The process includes five phases: keyframe extraction, pre-processing, feature extraction, feature selection, and classification.Adaptive keyframe extraction is followed by noise removal using the Adaptive Gaussian normalized median filter (Ad-GauNMF).Features are extracted using Color Correlation (CC), Adaptive gray level co-occurrence matrix (Ad-coGLCM), and Modified Histogram of Oriented Gradients (Mod_HOG).The Hybrid tent chaos randomized single candidate optimizer (Hyb-Chr-Sop) selects features, while Knowledge distillation-based Radial kernel function assisted ensemble random forest with AdaBoost (KDRad-EnRFBoost) detects anomalies.The proposed model exhibited high performance in terms of accuracy, with values of 99.03%.99.1% and 99.21% for dataset for crime scene activity in the surveillance system (DCSASS), theft detection dataset and UCFCrime dataset.