Photographing big things
View Patent ↗The subject application provides an architecture that facilitates producing a single image that can visualize a scene too large to depict from any single perspective view. These images can be stitched together on a 2-dimensional picture surface to form a multi-perspective image of the entire extent of the scene depicted by the input images. Where the respective images overlap, an objective function can be employed to determine where to make a transition between one image and the next. The objective function can employ a data cost and a seam cost to make this determination.
1. A computer implemented system that facilitates optimized, multi-perspective digital image generation, comprising:
an interface that receives a plurality of images that correspond to a plurality of camera locations along a camera path, each respective image depicts a portion of a 3-dimensional scene from the point of view of one of the plurality of camera locations; and
an image generation component that generates an optimized image, the image generation component comprising:
a surface definition component that determines a dominant depth for the scene and generates a 2-dimensional picture surface at the dominant depth;
a projection component that projects each pixel of an image from the plurality of images onto the picture surface and identifies overlapping pixels; and
an optimization component that selects a best pixel to be projected from among the overlapping pixels based on an objective function.
2. The system of claim 1 , the objective function is based upon a seam cost, the seam cost is a function of consistency of a pixel with adjacent pixels, a strong edge in the picture surface, and the dominant depth.
3. The system of claim 2 , the optimization component minimizes the objective function, the objective function is the sum of the seam cost and a data cost, the data cost is proportional to the distance of a pixel from the center of an image that includes the pixel.
4. The system of claim 2 , the seam cost is based upon a depth discontinuity between objects in the scene.
5. The system of claim 2 , the seam cost is based upon item recognition of items in the scene.
6. The system of claim 1 , the dominant depth is based upon at least one dominant feature of the scene.
7. The system of claim 1 , the dominant depth is based upon the alignment between images projected onto the picture surface.
8. The system of claim 1 , the plurality of camera locations are each separated by a distance of at least 6 inches.
9. The system of claim 1 , at least one of the plurality of images is a multi-perspective image.
10. A computer implemented methodology that facilitates generating a 2-dimensional picture surface of a scene, comprising:
receiving a plurality of images relating to a scene, each image corresponding to a different camera position on a camera path;
computing a plurality of projection matrices, one projection matrix for each camera position corresponding to the plurality of images;
constructing a 3-dimensional model of the scene;
determining a plurality of feature points for the model based upon dominant objects in the scene;
determining a dominant depth for the model based at least in part upon the plurality of feature points; and
defining a 2-dimensional picture surface at the dominant depth of the model.
11. The method of claim 10 , the dominant depth is determined based upon a most dominant object in the scene.
12. The method of claim 10 , the dominant depth is determined based upon an efficient alignment between a plurality of feature points within the model.
13. The method of claim 10 , further comprising spacing each camera position on the camera path by a distance of at least 6 inches.
14. The method of claim 10 , the plurality of images received comprises a multi-perspective image.
15. The method of claim 10 , the scene is a 3-dimensional scene.
16. A computer implemented methodology that facilitates optimizing a multi-perspective picture surface, comprising:
receiving a plurality of images relating to a scene, each image corresponding to a different camera position on a camera path;
projecting each pixel of an image from each of the plurality of images onto a multi-perspective picture surface;
identifying overlapping pixels;
selecting a best pixel between the overlapping pixels based upon an objective function, the objective function employing at least a seam cost; and
determining the seam cost based upon uniformity between neighboring pixels, an edge feature in the picture surface, and a dominant depth of the scene.
17. The method of claim 16 , the best pixel is selected by minimizing the objective function, which is determined by summing the seam cost and a data cost, the data cost is proportional to the distance of a pixel from the center of an image comprising the pixel.
18. The method of claim 16 , the determining a seam cost is further based upon a depth discontinuity between objects in the scene.
19. The method of claim 16 , the determining a seam cost is further based upon item recognition of items in the scene.
20. The method of claim 16 , the picture surface is a 2-dimensional picture surface residing at the dominant depth of the scene.