This document proposes an approach for abnormal event detection in unseen scenarios using blob-level statistical analysis of video frames. The approach extracts frame-level features from foreground blobs using measures like blob area, count, and distance. These are transformed into temporal features considering speed and order. Event models are built offline using these features and an SVM classifier. During operation, only selected frame and temporal features are extracted in real-time, without additional training or tuning. Experiments on public safety and crowd datasets show the approach can detect events in unseen environments, outperforming optical flow-based methods.