Patent Issued For Abandoned Object Recognition Using Pedestrian Detection

patent application

A patent by the inventors Brown, Feris, Kjeldsen, and Scherbaum was assigned patent number 8675917 to International Business Machines Corporation.

"There is a growing demand for automated video surveillance (AVS) systems for public safety and security enhancement. While traditional video surveillance methods require constant human attention, automated visual analysis performs real-time monitoring of people, vehicles and other objects, and generates alerts when suspicious persons or abnormal activity are detected. Such automatic analysis significantly increases the effectiveness of the monitoring by reducing the number of human operators needed, thus is crucial for urban surveillance where over thousands of cameras are set up to monitor a large area on the scale of a city.

"AOD systems typically detect static objects in a scene, using background modeling and subtraction (BGS). However, a number of non-threatening objects are often observed staying static (such as cars stopping at a red light) or near static (pedestrians standing still on the street) for a short period of time. Moreover, temporarily static objects, if not properly handled, would pose serious adverse effects on background subtraction. Generally, traditional BGS approaches such as the Gaussian mixtures model will gradually adapt people that are standing or sitting still into the background, and as a result, AOD techniques based on BGS may confuse the still people with a suspicious object. […]