You are most welcome to attend the Work Completion Seminar by Mr Mr. Low Yi Qian (Master of Engineering Science), on his research title of New SIFT-based Calibration Methods for Hybrid Camera System. Details are as follow:
Date: 25/11/2011 (Fri)
Venue: 5A meeting room, 5th floor, SA Block, FES, UTAR KL Campus.
Research title: New SIFT-based Calibration Methods for Hybrid Camera System
Candidate: Low Yi Qian
Multi-cameras networks have an important role in surveillance systems. In traditional cameras networks mainly based on static cameras. Static camera provides low resolution and poor image quality in modeling. At the meanwhile, Pan-Tilt-Zoom (PTZ) camera is able to obtain multi-angles of views and multi-resolution information; it provides more functionality compared with static camera. In multi-camera networks, they obtain two or more different perspectives views of scene to perform calibration within one another. Traditional approach to solve this problem is by using human inputs or referenced objects to find matched correspondences among all images. The recent works by some researchers have brought remarkable increase of automation to these problems. One of the main challenges in multi-cameras calibration is to obtain an accurate and fast estimation of disparities of two different views of scene. Besides disparities, differences might be caused by occlusion of the object, specular reflection, sensor noise and various other causes.
In this project, we designed and constructed a new hybrid camera system instead of using two static cameras. More precisely, our hybrid camera system consists of a static wide angle camera and a PTZ camera. Both cameras obtain different optical elements and resolutions. We proposed a master-slave concept to represent both cameras. A static camera is master camera with wide angle view is used to monitor the environment from a distance. The images taken from the master camera will be used as reference images. When the master camera detects an object of interest, the slave camera - PTZ camera, will zoom into the region of interest (ROI). PTZ camera plays an important role at pointing towards the ROI in high resolution as well as providing different parameters of PTZ values. Then, we studied and proposed calibration methods for the hybrid camera system. In particular, Scale Invariant Features Transform (SIFT) was used to identify keypoints to perform multiple views matching. One of the major contributions of this project is to design and develop a new smart filtering technique based on the principle of trigonometry, called Trigonometry Filter Technique (TFT). TFT is used to filter the false keypoints by calculating and classifying the orientation of the images. Furthermore, TFT also increases the determination and detection rate of ROI image localization. Secondly, we proposed a novel calibration method based on multiple-image stitching approach. By stitching images (by PTZ camera), a high resolution panorama image will be created. Finally, we are able to stitch and calibrate both static master camera and PTZ camera images without the use of any object as reference. By eliminating the use of reference objects, the obtained empirical results show good performance to handle images from various kinds of angles, scaled and target uncertainties.