Human anomaly detection is crucial for enhancing security in high-risk areas such as banks, parking lots, hospitals, shopping centers, educational institutions, borders, airports, and transit stations. Traditional manual CCTV monitoring is labor-intensive and inefficient, highlighting the need for automated video surveillance systems capable of real-time anomaly detection. Most previous approaches were based on a single model and thus suffered from high bias, high variance, extensive data requirements, varying conditions, and difficulty handling subtle anomalies. In contrast, the proposed model, bootstrap ensemble long-term recurrent convolution network, addresses some of the limitations mentioned above and enhances the efficiency of anomaly detection in video surveillance. The proposed approach is validated on four benchmark datasets available ib public, along with a small real-time dataset created by us in our lab. Empirical outputs illustrate that the proposed method’s performance is quite satisfactory.