Improved understanding of crowd dynamics would mitigate the increasing threat of crowd-related disasters. Under a program for the Pacific Northwest National Laboratory, SRI is developing novel solutions for the detection and tracking of humans in crowds. Leveraging its software for the real-time video exploitation and visualization of data, SRI is enhancing its deployed system for real-time, multi-camera tracking to achieve robust performance in crowds.
The new prototype system features several innovative technologies.
- It provides robust human detection with a low false alarm rate in the presence of environmental occlusions, illumination variations, camera shaking, and extraneous motion from foliage, flags, moving vehicles, and other objects.
- The system also offers comprehensive appearance, shape, and motion modeling for robust tracking, even when objects occlude each other, tracks are lost temporarily, and humans change their poses.
- Its architecture enables the maintenance of tracks over wide areas by using geolocations of tracks, object fingerprinting, and dynamic and static occlusion modeling.
The system's core technology is computationally amenable to real-time implementation on commercial-off-the-shelf processors with a frame rate greater than 10 Hertz and a latency of less than 0.25 seconds. In addition, the scalable system architecture can interface with hundreds of cameras.