My primary research focus is big data analysis of large scale camera networks. A testbed consisting of large scale distributed camera network deployed in and around the UCSB campus is used to test the efficacy of computer vision algorithms.
Tracking in Single and Multiple Cameras
Existing state of the art tracking algorithms tuned for one dataset fail miserably when applied on a completely different dataset due to reasons such as illumination changes, insufficient appearance and motion modeling. The primary focus of this project is to develop robust tracking algorithms that exploit contextual information, multi-camera fusion and global appearance to improve the tracking.
View Relevant Publications
Search and Retrieval in Single and Multiple Cameras
The advent of smart cameras enabling wide area network coverage resulted in a big-data. Given the bandwidth and network constraints, de-centralized approaches taking full use of limited computation power in individual cameras have been gaining prominence. Also, it is a huge challenge to handle the collected big-data in an efficient manner. Towards this extent, this project analyses distributed video surveillance systems that enable users to query and find a set of both “informative” and “diverse” snapshots to infer network events.
View Relevant PublicationsActivity Recognition in Surveillance Videos
With the advances in technology, cheap and miniature network-enabled cameras are available for large scale deployment and monitoring wide areas. However, due to limited processing power, most of the computational bandwidth is spent either on computing visual features or lowlevel processing of video frames in realtime. This project focuses on building a computationally efficient activity recognition algorithms for uncontrolled scenes.
View Relevant Publications
Calibration of Camera Networks
In a wide-area camera network, cameras are often placed such that their views do not overlap. Collaborative tasks such as tracking and activity analysis still require discovering the network topology including the extrinsic calibration of the cameras. This project focuses on calibration of fixed cameras in a wide-area network by combining measurements from image sensors, GPS receivers, and orientation sensors.
View Relevant Publications