The details on activity recognition is yet to be added.

Opposed to token based (e.g. possession of smart card) or knowledge based (e.g. knowing a PIN/password), biometric systems authenticate a returning user based on his traits that are inherent to him/her. These traits can be physiological (cannot be altered easily as per user's choice, e.g. eye image), or behavioral (can be altered by an user's choice, e.g. signature). Biometric systems are excepted to deal with reliable traits. However, usually the traits are subject to change with time, varying illumination, pose etc, making it challenging for the system to work properly. Therefore simple mathematical comparison of saved data and new data by the user is not functional, leading to the usage of pattern recognition techniques.
Broadly, the current researches in the field of biometric can be classified into following groups in which our laboratory has done extensive research in most of these:

1. Testing performance of different traits with different feature extraction and matching algorithms
2. Developing tailor-made features for a particular trait
3. Classification of subjects based on gender, ethnicity; kinship verification
4. Indexing and organization of biometric data for fast matching
5. Soft and cancellable biometrics

The details on visual forensic is yet to be added.

The details on UAV navigation and automatic signaling is yet to be uploaded.

A typical outdoor video surveillance system deploys a network of cameras to monitor a wide-area scene, e.g. airport, car parking zone, and shopping complex. For global monitoring, it is necessary that the network bears a semantic communication among the member cameras. Also, with advent of PTZ cameras, the movement of these cameras are utilized intelligently for effective surveillance.
Broadly, the research in this field is conducted on the following domains in which our laboratory has done extensive research in most of these:

1. Calibration and topology selection for multi-camera network
2. Covering maximal surveillance area with minimal cameras
3. Developing algorithms to make cameras 'smart' by embedding functionalities like 'follow the pedestrian'
4. Architecture, middleware, and applications of smart camera networks

Person re-identification can be modelled as a recognition/matching task, where a probe individual is matched against a gallery of templates (representing the individuals previously seen by the camera network). The person re-identification process can be seen as a combination of following sequential tasks: multi-person detection, feature extraction from appearances cues of the detected persons and storing them in a gallery, feature extraction and matching of the probe subject with the gallery. Adding to this, the challenges become multi-folded when external factors like camera resolution, intensity variation, occlusion, pose variation come into play.
Broadly, the research in this field is conducted on the following domains in which our laboratory has done extensive research in most of these:

1. Engineering suitable visual features for re-identification
2. Finding low-cost computational model for feature extraction and matching
3. Developing algorithms to make cameras 'smart' by embedding functionalities like 'follow the pedestrian'
4. Coping with external factors like intensity variation, clothing condition, pose variation etc.

The details about video summarization is about to be uploaded.