Three-dimensional scene reconstruction

1. On a computing system, a method, comprising:

receiving a two-dimensional dataset comprising two-dimensional image data of a physical scene;

performing a three-dimensional reconstruction of the physical scene by:

processing the two-dimensional dataset;

reconstructing one or more depth values associated with the two-dimensional dataset;

constructing a three-dimensional representation of the physical scene based at least on the one or more depth values;

outputting the three-dimensional representation;

wherein the three-dimensional representation of the physical scene is constructed based further on a result of a prior three-dimensional representation constructed by the computing system; and

wherein the result is stored in a data store communicatively coupled to the computing system, the data store further storing one or more unknown three-dimensional objects that failed to be classified.

2. The method of claim 1 , wherein performing the three-dimensional reconstruction of the physical scene includes:

acquiring one or more parameters associated with an imaging device used to capture the two-dimensional dataset; and

normalizing the two-dimensional dataset based on the one or more parameters associated with the imaging device.

3. The method of claim 1 , wherein performing the three-dimensional reconstruction of the physical scene includes selecting a number of computing devices of the computing system to perform the three-dimensional reconstruction based on the two-dimensional dataset.

4. The method of claim 1 , wherein performing the three-dimensional reconstruction of the physical scene includes selecting a set of computing resources of the computing system to perform the three-dimensional reconstruction based on a processing load of the computing system.

5. The method of claim 1 , wherein performing the three-dimensional reconstruction of the physical scene includes classifying at least one object in the two-dimensional dataset, and wherein the three-dimensional representation of the physical scene is constructed based on the at least one classified object.

6. The method of claim 5 , wherein the at least one object is a dynamic object having a variable state, and wherein the three-dimensional representation of the physical scene includes the dynamic object and encodes the variable state.

7. The method of claim 1 , wherein performing the three-dimensional reconstruction of the physical scene includes fusing each portion of the two-dimensional dataset that is of a threshold similarity into a respective group, and wherein the one or more depth values are reconstructed for each group as a whole.

8. The method of claim 1 , wherein performing the three-dimensional reconstruction of the physical scene includes filtering the three-dimensional representation of the physical scene by identifying and discarding one or more disconnected components of the three-dimensional representation.

9. The method of claim 1 , further comprising generating one or more texture files that are spatially associated with the three-dimensional representation of the physical scene, wherein the one or more texture files are generated based on the two-dimensional dataset.

10. A computing system, comprising:

one or more logic machines configured to:

receive a two-dimensional dataset comprising two-dimensional image data of a physical scene;

schedule a three-dimensional reconstruction of the physical scene to be executed by a number computing devices of the computing system, the number of computing devices selected based on the two-dimensional data set and on a processing load of the computing system;

via the selected number of computing devices,

process the two-dimensional dataset;

reconstruct one or more depth values associated with the two-dimensional dataset;

construct a three-dimensional mesh of the physical scene based at least on the one or more depth values;

output the three-dimensional mesh; and

wherein the three-dimensional representation of the physical scene is constructed based further on a result of a prior three-dimensional representation constructed by the computing system; and

wherein the result is stored in a data store communicatively coupled to the computing system, the data store further storing one or more unknown three-dimensional objects that failed to be classified.

11. The computing system of claim 10 , wherein the one or more logic devices are further configured to vary the number of computing devices selected to execute the three-dimensional reconstruction of the physical scene as the processing load of the computing system changes.

12. The computing system of claim 10 , wherein the one or more logic devices are further configured to select a set of computing resources with which to execute the three-dimensional reconstruction of the physical scene, the selected set of computing resources including one or more of processor cores, functional units, threads, time slices, and memory ranges.

13. The computing system of claim 10 , further comprising a data store storing a result of a prior three-dimensional reconstruction, and wherein the three-dimensional mesh of the physical scene is constructed based further on the result.

14. The computing system of claim 10 , wherein executing the three-dimensional reconstruction of the physical scene includes classifying at least one object in the two-dimensional dataset, and wherein the three-dimensional mesh of the physical scene is constructed based on the at least one classified object.

15. The computing system of claim 10 , wherein executing the three-dimensional reconstruction of the physical scene includes fusing each portion of the two-dimensional dataset that is of a threshold similarity into a respective group, and wherein the one or more depth values are reconstructed for each group as a whole.

16. The computing system of claim 10 , wherein executing the three-dimensional reconstruction of the physical scene includes filtering the three-dimensional representation of the physical scene by identifying and discarding one or more disconnected components of the three-dimensional representation.

17. A computing system, comprising:

one or more logic machines configured to:

receive a two-dimensional dataset comprising two-dimensional image data of a physical scene;

acquire one or more parameters associated with an imaging device used to capture the two-dimensional dataset;

schedule a three-dimensional reconstruction of the physical scene to be executed by a set of computing resources of the computing system, the set of computing resources selected based on the two-dimensional data set and on a processing load of the computing system;

via the selected set of computing resources,

normalize the two-dimensional dataset based on the one or more parameters associated with the imaging device;

classify at least one object in the two-dimensional dataset;

reconstruct one or more depth values associated with the two-dimensional dataset;

construct a three-dimensional mesh of the physical scene based on the one or more depth values and the at least one classified object;

output the three-dimensional mesh for display by a display device;

wherein

the at least one object is a dynamic object having a variable state, and wherein the three-dimensional mesh of the physical scene includes the dynamic object and encodes the variable state,

executing the three-dimensional reconstruction of the physical scene includes fusing each portion of the two-dimensional dataset that is of a threshold similarity into a respective group,

the one or more depth values are reconstructed for each group as a whole,

executing the three-dimensional reconstruction of the physical scene includes filtering the three-dimensional mesh of the physical scene by identifying and discarding one or more disconnected components of the three-dimensional mesh,

the one or more logic devices are further configured to, via the selected set of computing resources, generate one or more texture files that are spatially associated with the three-dimensional mesh of the physical scene, wherein the one or more texture files are generated based on the two-dimensional dataset, and

the three-dimensional mesh of the physical scene is constructed based further on a result of a prior three-dimensional mesh constructed by the computing system.