Abstract: Edge video processing faces challenges due to limited bandwidth, storage, and computational resources. This work explores low-cost methods for video triage and summarization at the edge, specifically targeting long-format video streams from sensors like security cameras and drones. We investigate perceptual hashing and Shannon entropy as information-theoretic approaches to identify and remove semantically redundant frames, avoiding computationally expensive deep learning models. Through exploratory experiments using the MEVA dataset and airport ground operation videos, we demonstrate that these methods effectively correlate with video activity. A key finding is that transmitting frames only when perceptually different by more than one bit reduces frame transmission by 89-99%. These results suggest a promising avenue for efficient edge video processing, with future work focused on quantitative evaluation against annotated ground truth data.