Research

This page is incomplete and is being updated. :-)

Active Authentication on Mobile Devices

Rama Chellappa, Larry Davis, Vishal Patel, Vlad I. Morariu

As mobile devices are becoming more ubiquitous, it becomes important to verify the identity of the user during all interactions rather than just at login time. Active authentication (AA) systems deal with this issue by continuously monitoring smartphone sensors after the initial access has been granted. However, AA remains an unsolved problem specially for smartphones.

This project aims to develop more robust algorithms for AA on smartphones using various modalities such as face, touch gestures and other physiological and behavioral traits.

Downloads

Mobile Face Video Dataset:

(Download link available only by email)

The dataset includes 750 face videos of 50 users captured while using a smartphone by its front-facing camera. For each user, five of the videos are taken in a well-lit room, another five are taken in the same room with dim lighting, and the remaining five are taken in a different room with natural daytime illumination. Each group of five videos consists of one video for enrollment and four videos taken while the user is using a custom-written app to perform one of four tasks.

If you want to download the dataset, please send an email to rama (at) umiacs*umd*edu (replace the *'s with dots) with "Mobile Face Dataset Download Request" in the subject line. Please also include basic information about yourself, your institute and the reason you need the dataset. You will receive the download link in a reply to your email.

UMDAA-02:

The University of Maryland Active Authentication Dataset-02 (UMDAA-02) data collection drive (15 Oct. to 20 Dec., 2015) yielded 141.14 GB of smartphone sensor signals collected from 48 volunteers using Nexus 5 phones as their primary device for around one week. The data collection sensors include the front-facing camera, touchscreen, gyroscope, accelerometer, magnetometer, light sensor, GPS, Bluetooth, WiFi, proximity sensor, temperature sensor and pressure sensor. The data collection application also stored the timing of screen lock and unlock events, start and end time stamps of calls, currently running foreground application etc. The usage information is arranged in "Sessions" which starts when the user unlocks the phone and ends when the phone goes to the locked state. Here we upload 2 modalities a) front camera captures (Images are captured at 3 fps for the first 60 seconds for each session) and b) touch.

Downloads: UMDAA-02 face dataset UMDAA-02 Touch dynamics datsaset

Independent Moving Object Detection

Rama Chellappa, Joshua Broadwater

The Science of Land Target Spectral Signatures

Rama Chellappa, Joshua Broadwater

This project is a collaboration with the Georgia Institute of Technology, University of Hawaii, Rochester Institute of Technology, Clark Atlanta University, and University of Florida. The Multiple University Research Initiative (MURI) team is focused on understanding the physical phenomena that affect target spectral signatures such as weathering, atmospheric effects, and camera position. Our part of the project focuses on incorporating physical phenomena into automatic target detection/recognition (ATD/R) algorithms for hyperspectral imagery. Our past research has provided improved parametric detectors that incorporate sum-to-one and non-negativity abundance constraints. Our current research revolves around the development of physics-based kernels for non-parametric detectors and the development of advanced adaptive threshold techniques based on importance sampling.

Algorithms and Architectures For Vision Based Inference From Distributed Cameras

Rama Chellappa, Ashok Veeraraghavan, Gaurav Aggarwal
Project page

This project is a collaboration with Dr.Shuvra Bhattacharya and Dr.Wayne Wolf .This project develops new techniques for distributed smart camera networks through an integrated exploration of distributed algorithms, embedded architectures, and software synthesis techniques. The objective is to build real time systems for coordinated inference from multiple cameras. Specifically, we are investigating a series of complex smart camera algorithms and applications, specifically, human gesture recognition; self-calibration of the distributed camera network; detection, tracking and fusion of trajectories using distributed cameras; view synthesis using image based visual hulls; gait-based human recognition; and human activity analysis.

New Technology for Capture, Analysis and Visualisation of Human Movement Using Distributed Cameras

Rama Chellappa, Aravind Sundaresan, James Sherman
Project page

This project is a collaboration with the Biomotion Laboratory at Stanford University and the Media Research Laboratory at New York University. The objective is to perform markerless motion capture using multiple calibrated cameras. We use shape models such as super-quadrics to represent the humans. Such models are essential for tracking the articulated motion accurately.

Insect Inspired Navigation of Micro Air Vehicles

Rama Chellappa, Mahesh Ramachandran, Kaushik Mitra, Ashok Veeraraghavan
Project page

This project aims at developing autonomous Micro Air Vehicles capable of navigation and terrain understanding. The project is a collaboration with Aerospace Engineering , Mechanical Engineering Departments at the University of Maryland and Visual Sciences Group at Australian National University.

Current focus is on the following issues.

Estimating the ego-motion of Micro Air Vehicles(MAV) from a combination of visual and inertial sensors.
Terrain Understanding from videos acquired from MAVs.
Simultaneous tracking and Behavior analysis of social insects.

Multi-sensor Multi-nodal Fusion for Audio-Video Survillence

Rama Chellappa, Qinfen Zheng, Aswin Sankaranaryanan, Wu Hao, Xue Mei, Seong-Wook Joo
Project page

This project is a collaboration with BAE Systems and Prof. Jim McClellan of Georgia Institute of Technology. This project addresses issues related to surveillance and outdoor scene monitoring and develops algorithms for the following.

Develop PDA based remote sentry devices for monitoring scene acoustic and video through secure wireless networks.
Multi-nodal Multi-sensor fusion for robust tracking with efficient use of the individual modality for low power consumption and prolonged battery life.
Multi-camera tracking (especially for non-overlapping views).

Verification and Identification for Surveillance Video

Rama Chellappa, Zhanfeng Yue, and Arunkumar Mohananchettiar
Project page

This project is a collaboration with SAIC Corporation. The objective is to develop automatic target verification and identification for airborne surveillance video. The following issues are addressed in the project: Target detection from airborne surveillance video (moving platform). Reliable target tracking. Shadow detection. A verification by synthesis algorithm using homography and template matching. An alternative verification system accomplishing tracking and recognition simultaneously.