This document presents a method for panic-driven event detection from surveillance video without using track and motion features. The method extracts frame-level features from each video frame independently, then transforms those into temporal features considering speed and order to characterize events. It was tested on four events from the CAVIAR dataset with over 80% accuracy and shown to detect abnormal events in three sites from the UMN dataset. The approach aims to develop a more generic event detection framework compared to tracking or motion-based methods.