Domitille Commun 

(Advisor: Prof. Mavris)

will propose a doctoral thesis entitled, 

An approach for UAV-Enabled surveillance camera calibration in various environments

On

Monday, March 23 at 9:00 a.m.
https://bluejeans.com/406001596

ABSTRACT:

Several video applications for public safety, infrastructure security, manufacturing, building construction and 3D reconstruction of an object or a scene rely on the precise setup of cameras. Camera calibration is a key enabler for these video applications, and is possible through determining the mapping between 3D real world coordinates of an object and its 2D pixel coordinates on an image taken by a camera. It involves determination of internal parameters, such as the focal length, and external parameters, such as camera location and orientation. Calibrating a camera is necessary for the implementation of accurate computer vision techniques such as object detection and for determination of object locations. Camera calibration methods enable well performing machine vision techniques and can support the automated analysis of video footages, which is key for achieving cost and time savings in a variety of fields such as manufacturing, civil engineering, architecture and safety.

Existing calibration methods have not been adapted for use with surveillance cameras that are already installed. Such methods either require easy access to the camera, or are only effective for certain kinds of scenes. Self-calibration approaches do not require camera access, but require the presence of an object or infrastructure with three orthogonal directions. They apply, for instance, when a building is seen by the camera. Yet they are not effective for a number of cases, e.g. when the camera is facing the sea (harbor, offshore platform), is located in a non-urban environment (forest, fields, parks etc.) or it is installed at the top of buildings.

The objective for this thesis is the development of a camera calibration technique which is applicable across a broader range of operating environments, which would accommodate cameras that are already installed. The calibration is enabled with the use of Unmanned Aerial Vehicles (UAVs) that are equipped with location sensors such as an altimeter and a GPS. According to the proposed approach, the UAV will be used to sample points in the 3D space and the corresponding 2D pixel coordinates on the image provided by the surveillance camera will be automatically extracted. Once the 3D and corresponding 2D coordinates of the sampled points are obtained, a parametric regression will be implemented using these sampled points to model the mapping from 3D to 2D. As the mapping obtained through the parametric regression does not account for manufacturing errors or weather impact on the camera lens, a non-parametric regression will consequently be implemented in order to take these uncertainties into account. Once the 3D to 2D mappings are obtained, i.e. the camera is calibrated, their precision will be evaluated. The approach will be first developed and implemented using a simulation environment and will then be tested in a real environment. 

It is expected that this new method will enable the calibration of cameras that are difficult to access and located in urban areas or more complex environments such as forests, parks, offshore platforms and other non-urban areas. Thus, enabling the automated analysis of videos from those cameras. One application of the method developed in this thesis is to enable the collection of additional traffic data which would be useful to municipalities, infrastructure planners and local police departments. Based on the availability of the resources required a test of the methodology may be possible with real time traffic video on the intersection at a university campus.

Committee Members:

• Dr. Dimitri Mavris – School of Aerospace Engineering (Academic Advisor)
• Dr. Olivia Pinon Fischer – School of Aerospace Engineering
• Dr. Cédric Pradalier – School of Interactive Computing
• Dr. Graeme Kennedy – School of Aerospace Engineering
• Dr. Michael Balchanos – School of Aerospace Engineering