Anomaly detection is a crucial task in video for numerous applications such as surveillance, security, and detecting anomalies in industrial processes. This research proposes a new approach for anomaly detection in surveillance data using an ensemble method of isolation forest and robust covariance. The proposed method involves three stages: feature extraction, key frame extraction, and anomaly detection. Our approach for detecting anomalies involves using robust covariance-based feature extraction and thresholding to identify key frames. These key frames can then be inputted into an isolation forest algorithm. The proposed approach is tested on a surveillance dataset that is publicly available to study the effectiveness, and its performance is compared to other methods that are currently considered the most advanced in the field. The outcomes show that the proposed approach surpasses alternative approaches in accurately identifying anomalies, providing precise detections, and recalling a higher number of true anomalies.