Stabilizing 2-Axis

Tracking Across Multiple Automatically Calibrated

Cameras With Disjoint and overlapping Views
Text Box: 40Mb

The system uses  Servo Engines controlled by RT Machine-Vision  software programmed with OpenCV package.  Harris Detectors were used to lock on shoreline features.
Text Box: The system is built upon matching Algorithms that compare between panoramic spheres in order to determine the overlapping views of two or more cameras. Text Box: Overview Surveillance of wide area requires a network of cameras. In such a setup, it is not always possible to have overlapping camera views. Hence, multiple observations of the same object are obtained that can be widely separated in time and space. Moreover, it is preferable that the tracking system does not require camera calibration or complete site modeling, since in most situations it is not cost effective or practical to calibrate real life surveillance systems comprised of dozens of cameras. Text Box: Our current focus Is set on the problem of multi-camera tracking in a system of non-overlapping non-calibrated cameras. The task of a multi-camera tracker is to establish correspondence between observations of objects across cameras. We assume that tracking information is available for individual cameras, and the objective is to find correspondences between these tracks, obtained by different cameras, in such a way that enables us to reconstruct the entire path of the object moving across the area covered by the surveillance system .After determining the overlapping field of view and figuring the neighboring cameras, our next objective is to have on the fly optimization of the system resources. This can achieved by rotating and focusing the cameras where they are mostly needed in order to provide seamless tracking over multiple targets. Text Box: The image demonstrates the need to match between image “spheres” in order to deduce overlap sections between two or more cameras

Mixing technologies

Having many cameras and image processors deployed in field conditions usually involves partial failure and demands adding and subtracting cameras from the distributed surveillance system. By using Sonarion’s CLAN distributed infrastructure we can support  hot swap of resources in and out of the system and provide self –healing and failover capacity. CLAN is founded on Sun’s Jini and Javaspaces technology.

Shades and color variations are used to extract more knowledge about  the scene. Some of the center’sHuman detection projects use this concept as well.

Overview

Machine vision is used at the Sonarion-Hadassah Academic VR Center for detecting humans, performing vehicle classification and other robotic missions such as camera stabilization, automatic multiple view matching and eye-in-hand types of projects.

Working hand in hand with leading research institutes such as Hadassah Hospital, The Volcani Agricultural Institute and Industrial entrepreneurships like GolanTech has further broadened the scope and depth of our vision projects. 

Roni & Omri, moments before testing the Stabilized Vision System while sailing a yacht off the Tel Aviv shore

Machine Vision