Obviousness

Obviousness Analysis of U.S. Patent 10,979,693 under 35 U.S.C. § 103

This analysis evaluates whether the independent claims of U.S. Patent 10,979,693 would have been obvious to a Person Having Ordinary Skill in the Art (PHOSITA) at the time of the invention, based on the prior art references cited during the patent's examination. The standard for obviousness under 35 U.S.C. § 103 is whether the differences between the claimed invention and the prior art are such that the subject matter as a whole would have been obvious to a PHOSITA.

A PHOSITA in this context would be an engineer or computer scientist with a bachelor's degree in a relevant field and several years of experience in computer vision, image processing, and software development, particularly in the areas of stereoscopic video and virtual reality.

The central argument for obviousness rests on the combination of two key prior art references: US 2011/0141349 A1 (Albuz) and US 2013/0124471 A1 (Chen). A PHOSITA, faced with the well-known problem of stabilizing stereoscopic video for VR applications to reduce user discomfort and enhance immersion, would have been motivated to combine the teachings of these references.

---

Combination of Prior Art Rendering Claims Obvious

Primary Combination: Albuz (US '491) in view of Chen (US '471)

The independent claims of the '693 patent describe a specific method for video stabilization. This method involves:

1. Identifying a reference frame bracketed by preceding and succeeding frames.
2. Characterizing motion by comparing these bracketing frames.
3. Applying a combined matrix operation involving a projection matrix (derived from camera intrinsics like focal length), its inverse, and a motion matrix to stabilize the reference frame.

The combination of Albuz and Chen teaches or suggests these steps.

1. Motion Characterization from Surrounding Frames (Taught by Albuz):
   Albuz explicitly discloses analyzing motion between frames to create modified or interpolated frames. Critically, Albuz's method for motion estimation involves looking at frames both before and after a point in time to generate motion vectors. This directly teaches the claimed steps of "identifying a reference frame, a first set of frames before the reference frame, and a second set of frames after the reference frame" and "comparing the first and second set of frames to characterize a first motion." A PHOSITA seeking to create a robust stabilization algorithm would logically start with an accurate motion estimation technique, and the bracketing method taught by Albuz is a standard and effective approach for this.

2. Use of Camera Parameters for Processing (Taught by Chen):
   Chen discloses a system where image processing is driven by metadata, which includes intrinsic camera parameters such as focal length. This teaches the claimed step of using "at least one of a focal length of the stereoscopic camera at a time the reference frame was captured and a principal point of the stereoscopic camera" to calculate a matrix. While Chen describes this in the broader context of multi-image processing, a PHOSITA would immediately recognize its applicability to stabilization. Accurate stabilization requires correcting for the camera's specific optical properties, and Chen provides the rationale for using precisely this type of data.

3. Motivation to Combine and the Three-Matrix Operation:
   A PHOSITA tasked with stabilizing stereoscopic VR video would be highly motivated to combine Albuz's motion estimation technique with Chen's use of camera-specific metadata. The motivation is to achieve the highest possible quality of stabilization, a critical factor for a positive VR experience.

   • Albuz provides an effective method for finding the unwanted motion.
   • Chen provides the key to accurately correcting for that motion by taking the camera's unique optics into account.

   The final step—applying the specific three-matrix operation—would be an obvious implementation to a PHOSITA. The process of correcting for camera motion in 3D computer graphics is fundamentally a series of coordinate transformations, which are implemented via matrix multiplications. A standard, well-known method for changing a virtual camera's viewpoint involves:
   a) Transforming a 2D image point to 3D space using an inverse projection matrix (calculated from focal length and principal point, as suggested by Chen).
   b) Applying the corrective transformation in 3D space using a motion matrix (derived from the motion vectors found using Albuz's method).
   c) Projecting the newly corrected 3D point back onto the 2D image plane using the projection matrix.

   This sequence of operations is mathematically equivalent to the single, combined matrix operation claimed in the '693 patent. Therefore, combining the motion analysis of Albuz with the metadata-driven correction of Chen would lead a PHOSITA directly to the claimed method as a logical and standard implementation for solving the problem of video stabilization.

Conclusion

While the '693 patent claims a very specific sequence of calculations, the individual components of this method and the motivation to combine them were present in the prior art. Albuz teaches the method for characterizing motion from surrounding frames, and Chen teaches the use of intrinsic camera parameters like focal length to guide image processing. A Person Having Ordinary Skill in the Art would have been motivated to combine these teachings to create a more accurate video stabilization algorithm for the demanding application of virtual reality. The implementation of this combination using a projection matrix, its inverse, and a motion matrix represents a well-established technique in computer vision and would have been an obvious path to achieve the desired result. Consequently, the independent claims of U.S. Patent 10,979,693 appear to be obvious under 35 U.S.C. § 103.