MSc dissertation proposal 2016/2017
Forensic use of Mobile Phone Cameras
-- Information at fenix:
Objectives:
Surveillance cameras, and mobile phone cameras, capture many times suspects that need to be arrested. A key step for re-identifying those suspects is the image based measurement of the body height. The main objective of this thesis is to test methodologies for measuring heights of people given the video streams. For that purpose, one of the project goals is the estimation of the pose of the camera that captured the image of the suspect. Given the forensic nature of the application, it is assumed that the geometric scene data can be obtained after the video acquisition.
Goals:
1) Review of methodologies for acquiring 3D scene data to combine with the image data;
2) Estimating the height of a person imaged by a fixed camera or from a mobile phone;
3) Uncertainty analysis of the height estimation process.
Requirements (grades, required courses, etc):
-
Place for conducting the work-proposal:
ISR / IST
Observations:
The increasing need of surveillance in public spaces and the recent technological advances on embedded video compression and the communications made camera networks ubiquitous. Typical environments include single rooms, complete buildings, streets, highways, tunnels, etc. While the technological advances already allowed such a wide installation of camera networks, the calibration of these cameras (namely the external parameters calibration) is still an active research area.
For these type of scenarios, one of the crucial problems is to estimate the camera pose in the scenario. After the computation of the camera location on the scenario, the height of the suspect can be estimated (assuming some errors due to the person’s posture).
Given the estimated height of the person, it is important for the forensic application to obtain also an estimate of uncertainty. The estimation of the uncertainty can be simply based in adding noise to the data and reestimating the height, i.e. applying a Monte Carlo analysis methodology.
-- More information:
The increasing need of surveillance in public spaces and the recent technological advances on embedded video compression and communications made surveillance cameras ubiquitous. Even in places not equipped with surveillance cameras, cameras ubiquiti still happens - people use the cameras of their mobile phones and effectively acquire relevant forensic information.
While the technological advances already allowed such a ubiquitous presence of cameras, the still infancy stage of video analytics and the high variability of forensic research challenges make per si the extraction of information from the cameras an active research area.
The main objective of the work consists in the development and implementation of algorithms for the calibration of perspective cameras, using features obtained in the (forensic) scenarios, in order to acquire forensic information as e.g. people heights. More in detail, a major difference of the this proposal is that coordinates of 3D straight lines and points can be extracted from the scenarios and used for calibration [Leite08, Silva12, Galego14].
Given the calibration with respect to the ground plane and some priors, such as people standing up (legs straightened), then the height of persons can be measured. In addition, constraints regarding the position of the camera can be used, to ensure better calibration results. For example, if the camera is located in some wall, assuming that the scenario is known, some coordinates associated with the camera position can be predefined [Silva12, Galego14]. More, data of interest can be measured in site or based in auxiliary databases, e.g. architectural floor plans [Galego14]. Alternatively, or complimentarily, in most forensic investigations it is possible to visit again the scenario and use measurement equipment such as laser range finders to obtain 3D data of the scenario.
A critical aspect in forensic applications is the uncertainty of the data. After having estimated e.g. the height of a person, it is important to have an indication of the uncertainty of the estimate. Given typical noises, this information may be obtained using Monte Carlo simulation.
Work-proposal detailed steps:
- Review of methodologies for acquiring 3D scene which allows extracting 3D information from an image
- Estimating the height of a person imaged by a fixed camera or from a mobile phone
- Uncertainty analysis of the height estimation process
References:
[Galego14] Uncertainty analysis of the DLT-Lines calibration algorithm for cameras with radial distortion, R. Galego, A. Ortega, R. Ferreira, A. Bernardino, J. Andrade-Cetto, J. Gaspar. Computer Vision and Image Understanding (CVIU) 140, pp115-126, Elsevier 2015
[Silva12] M. Silva, R. Ferreira and J. Gaspar. Camera Calibration using a Color-Depth Camera: Points and Lines Based DLT including Radial Distortion. In WS in Color-Depth Camera Fusion in Robotics, held with IROS 2012.
[Leite08] Calibrating a Network of Cameras Based on Visual Odometry, Nuno Leite, Alessio Del Bue, José Gaspar, in Proc. of IV Jornadas de Engenharia Electrónica e Telecomunicações e de Computadores, pp174-179, November 2008, Lisbon, Portugal.
Expected results:
At the end of the work, the students will have enriched their experience in computer vision applied to forensic applications.
More MSc dissertation proposals on Computer and Robot Vision in:
http://omni.isr.ist.utl.pt/~jaghttp://omni.isr.ist.utl.pt/~jag