Recent advances in imaging and embedded computing technology enable widescale computer vision systems that incorporate arrays of low-cost, portable imagers connected through wireless networks. Video surveillance is a particularly important driving application area. This research focuses on embedded systems that can perceive, process, and interpret scene activity. It focuses on both fixed-camera surveillance applications and mobile camera scene analysis (e.g., onboard vehicles for intelligent transportation systems).

Sophisticated new video analysis algorithms are emerging for a broad range of important computer vision and surveillance applications, including activity recognition, anomalous behavior detection, wide-area distributed tracking, and accurate face and gait recognition. Integrating these algorithms into embedded environments creates new research challenges to balance their huge computational and communication demands with the stringent size, power, and memory resource constraints of embedded platforms.

Meeting these challenges requires a systems perspective, drawing on advances from a broad spectrum of fields, including signal processing, AI, computer vision, microelectronics, computer architecture, real-time systems, distributed computing/middleware, and rapid system prototyping. All of these areas are critical to solving real-time automated video surveillance problems with high efficiency and accuracy.

Sample Projects


Illumination Change Compensation

to improve the robustness of video analysis algorithms to sudden illumination changes
(also: GTILT Dataset)

Leap Segmentation

for efficient grouping of chromatic and illumination-
similar, non-contiguous regions
(also: GTTraffic Dataset)

Parallel Architectures for Multimedia
:
Portable Video Supercomputers

High-Efficiency Background Modeling

for low-cost, low-memory embedded platforms

Midground Object Detection

for identifying roadside threats, abandoned
luggage, and other suspicious activities

Video-Centric Applications

High-Performance Color Imaging
:
color-aware instruction sets

Automated Retargeting
of sequential imaging
software to parallel execution

Dynamic Optimization
of data
communication in multimedia architectures