C10707: Novel Video Registration ToolNovelty:
An efficient method to align and register the views from multiple video cameras into one composite video sequence with a common reference frame.
Value Proposition:
Video registration refers to a method by which multiple video sequences from multiple cameras are aligned to produce a panoramic video, currently considered an expensive and inefficient process. The technique presently involves extracting and tracking features through every frame in a video sequence in order to produce one combined video. This invention, developed at JHU is a novel algorithm that can register multiple video sequences both spatially and temporally without going through every frame of multiple videos. Additional advantages include:
• Efficient procedure since feature is extracted from only a few image frames
• Does not require expensive optimization procedure
• Easier to align deforming objects like water
• Temporal lag between video sequences does not affect spatial registration
• Software can be implemented to produce camcorder panoramic videos and surveillance videos
Technical Details:
Johns Hopkins researchers have developed an algorithm that can register two video sequences of rigid or non-rigid scenes taken from two synchronous or asynchronous cameras. The design method involves: 1) determining which images of the videos should be associated with each other using linear dynamic system (LDS) modeling and 2) comparing the associated images with each other in order to align the videos. Since LDSs do not depend on the temporal alignment, the spatial registration of the two videos can be obtained using classical image registration techniques to the parameters of the LDSs. Testing the algorithm on a variety of challenging video sequences has proven to be an efficient and effective alternative to the current expensive methods in practice.
Looking for Partners:
To develop and commercialize the technology as a video registration software for multi camera systems.
Stage of Development:
Proof of Concept
Data Availability:
Data Validated
Publications/Associated Cases:
IEEE Trans Pattern Anal Mach Intell. 2011 Jan;33(1):158-71
