Mining Images in Publicly-Available Cameras for Homeland Security

Abstract

A dramatic increase or decrease in the number of people appearing at a location can be an indicator that something has happened that may be of interest to law-enforcement, public health, or security. This work demonstrates how low quality camera images can be used to automatically alert when an unusual number of people are absent or present at a location. We report on experiments using publicly available, inexpensive cameras already operational over the Web. A “historical database” (H) for each camera is constructed by capturing images at regular time intervals and applying a face detection algorithm to store the number of faces appearing in each image (“face count”). Later, given an image X having timestamp t, if the face count of X is significantly higher or lower than the expectation inferred from H for time t, an unusual number of people are considered to be present in the image.

Keywords: Publicly-available cameras, CameraWatch, homeland security, privacy-preserving surveillance, video surveillance

Citation:
L. Sweeney and R. Gross. Mining Images in Publicly-Available Cameras for Homeland Security. AAAI Spring Symposium, AI Technologies for Homeland Security, 2005. (PDF).

Poster

Related Links

• Publicly-available Cameras, CameraWatch
• Preserving Privacy by De-identifying Facial Images
• Homeland security
• Privacy-preserving surveillance
• Privacy-Enhanced Linking
• Other Projects at the Data Privacy Lab
• Data Privacy Laboratory