|
About AMP Lab Projects Downloads Publications People Links
 |
Qi Wu Personal Homepage:
http://amp.ece.cmu.edu/people/qiwu/ |
|
|
Office: Porter Hall B41 Lab: Porter Hall B6 Phone: 412-268-7113 Fax: 412-268-3890 |
Mailing Address: Department of ECE, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA 15213-3890 |
|
[Research Interests] [Project] [Publications]
Research Focus:
Computer Vision and Geometry
Focus on the camera autocalibration in single and multiple view. We would like to propose algorithms to estimate intrinsic parameters and extrinsic parameters, which means its focal length, skew, position and orientation with respect to some reference, from one or multiple image observations without requiring measurements of scene objects.
Object Detection/Recognition
Because of the feature extraction and segmentation in image or video are not always stable, we would like to Introduce the segmentation label as a hidden information. After extraction, some unknown or ambiguous regions will be given different weights depends on the neighboring segmentation inference. Combined with Markov Random Field (MRF) algorithm, our goal is to propose a novel recognition framework used in general detection and recognition task, such as parking lot space detection or face recognition.
|
AutoCalibration of One or Multiple Cameras
In the context of visual surveillance, knowledge about a camera's internal and external parameters is useful, as it allows for the establishment of a connection between image and world measurements. In our project, we describe an approach for estimating both intrinsic and extrinsic parameters from vanishing points, vanishing line( horizon line) and an objects of known constant height. Assuming all lines connecting the upper points and lower points of objects, which are parallel, intersect at the image of the scene horizon, we use geometrical properties of vanishing points to obtain the them and vanishing line. In addition, we also use a defined foot-to-head homology to obtain them. After that, under the common assumptions of zero skew, unit aspect ratio and known principal point, the parameters can be estimated.
|
 |
Parking Lot
Space Detection
For the convenience of finding empty parking space in the parking lot, we design the algorithms for parking lot space detection using single camera. By dividing the input video frames into space detection patches. we first do the car segmentation to obtain color profile. After reducing their dimension with PCA, we set up Multi-SVM to classify and use Markov Random Field (MRF) for global optimization in order to solve the potential classify conflictions between neighboring patches. Our goal is to build a industrial and stable system for space detection with high accuracy. |
Conference Papers:
|
This
website is maintained by Devi Parikh
|
Top of this page