Eco-Efficient Surveillance: Transforming Video Data into Actionable Text Summaries

ICCV 2025 Workshop CV4A11y Submission10 Authors

Keywords: Object Detection, Video-to-text algorithms

Abstract:

The growing reliance on surveillance cameras has resulted
in massive storage requirements, leading to frequent dele-
tion of video footage due to limited storage capacity. This
practice not only raises concerns about the loss of cru-
cial evidence for investigations but also exacerbates en-
vironmental issues due to the energy-intensive nature of
video storage systems. To address these challenges, this pa-
per introduces Eco-Surve, a novel approach for transform-
ing surveillance video data into a compact, queryable sys-
tem without the necessity of storing the raw video footage.
This method enhances data efficiency, retrieval speed, and
privacy while maintaining the integrity of critical surveil-
lance information. By employing advanced object de-
tection algorithms(YOLO),video-to-text algorithms (Gem-
ini 1.5 pro and GPT 4) and reasoning large language mod-
els(DeepSeek), this method captures key details such as
timestamps, motion events, and object activities, ensur-
ing critical information is retained. Eco-Surve eliminates
the need for time-consuming manual searches from video
footage, significantly reducing the time required to identify
specific events or objects from hours to minutes, accounting
for reduction in time consumption by nearly 80%. Addi-
tionally, by reducing high-volume video storage demands
by 90%,it minimizes the energy and hardware resources
needed, thus mitigating environmental impacts like carbon
emissions and digital wastage. This dual benefit of saving
time and resources makes the proposed solution an impact-
ful tool for industries reliant on video monitoring systems,
ensuring efficient data management while retaining vital in-
formation for legal and investigative purposes.

Submission Number: 10