Damage detection from multi-view visual data
Reference images of an object may be mapped to an object model to create a reference object model representation. Evaluation images of the object may also be mapped to the object model via the processor to create an evaluation object model representation. Object condition information may be determined by comparing the reference object model representation with the evaluation object model representation. The object condition information may indicate one or more differences between the reference object model representation and the evaluation object model representation. A graphical representation of the object model that includes the object condition information may be displayed on a display screen.
1. A method comprising:
capturing a video of a designated object via a camera at a mobile computing device as the mobile computing device moves through space around the designated object;
mapping a sequence of evaluation images to an object model via a processor to create an evaluation object model representation, the sequence of evaluation images being selected from the video of the designated object;
determining that the sequence of evaluation images exhibit a coverage level that is beneath a designated threshold;
providing a recording guidance instruction via a display screen at the mobile computing device to capture additional video data;
determining object condition information for the designated object via the processor by comparing the evaluation object model representation with a reference object representation, the reference object representation being created by mapping a plurality of reference images of the designated object to the object model, the object condition information indicating one or more differences between the reference object model representation and the evaluation object model representation; and
storing a graphical representation of the object model that includes the object condition information.
2. The method recited in claim 1 , wherein the object model includes a three-dimensional skeleton of the designated object.
3. The method recited in claim 1 , wherein mapping the sequence of evaluation images to the object model includes applying a neural network to determine a pixel-by-pixel correspondence between the sequence of evaluation images and the object model.
4. The method recited in claim 3 , wherein the neural network receives as an input depth information captured from a depth sensor.
5. The method recited in claim 1 , the method further comprising:
estimating pose information for the sequence of evaluation images by identifying tracking points in the sequence of evaluation images and linking the tracking points across the sequence of evaluation images.
6. The method recited in claim 5 , wherein the sequence of evaluation images are associated with corresponding inertial measurement unit data captured via an inertial measurement unit at the mobile computing device, and wherein the tracking points are linked across the sequence of evaluation images based at least in part on the corresponding inertial measurement unit data.
7. The method recited in claim 1 , wherein the object condition information includes damage information selected from the group consisting of: an estimated probability of damage to the designated object, an estimated severity of damage to the designated object, and an estimated type of damage.
8. The method recited in claim 1 , wherein the graphical representation of the object model includes a standard view of the designated object that illustrates damage to the designated object.
9. The method recited in claim 8 , wherein the standard view of the designated object is selected from the group consisting of: a top-down view of the designated object, a multi-view representation of the designated object navigable in one or more directions, and a three-dimensional model of the designated object.
10. The method recited in claim 8 , wherein the damage to the designated object is illustrated via a heatmap.
11. The method recited in claim 1 , wherein the object model includes a two-dimensional skeleton of the designated object.
12. The method recited in claim 1 , the method further comprising:
identifying a plurality of components of the object model; and
determining component condition information for one or more of the components based on the object condition information.
13. The method recited in claim 12 , wherein a designated one of the components corresponds with a designated subset of the sequence of evaluation images and a designated portion of the designated object, the method further comprising:
constructing a multi-view representation of the designated portion of the designated object based on the designated subset of the sequence of evaluation images, the multi-view representation being navigable in one or more directions.
14. The method recited in claim 1 , the method further comprising:
providing live recording guidance for capturing one or more additional images via a camera, wherein the object condition information includes a statistical estimate, and wherein the live recording guidance is provided so as to decrease statistical uncertainty for the statistical estimate.
15. The method recited in claim 1 , the method further comprising:
constructing a multi-view representation of the designated object based on the sequence of evaluation images, the multi-view representation being navigable in one or more directions.
16. The method recited in claim 1 , wherein the designated object is a vehicle, and wherein the object model includes a three-dimensional skeleton of the vehicle, and wherein the object model includes a plurality of components, the plurality of components including each of a left vehicle door, a right vehicle door, and a windshield.
17. The method recited in claim 1 , wherein the video is captured at a first level of clarity or detail, and wherein the recording guidance instruction indicates a second level of clarity or detail.
18. One or more non-transitory computer readable media having instructions stored thereon for performing a method, the method comprising:
capturing a video of a designated object via a camera at a mobile computing device as the mobile computing device moves through space around the designated object;
mapping a sequence of evaluation images to an object model via a processor to create an evaluation object model representation, the sequence of evaluation images being selected from the video of the designated object;
determining that the sequence of evaluation images exhibit a coverage level that is beneath a designated threshold;
providing a recording guidance instruction via a display screen at the mobile computing device to capture additional video data;
determining object condition information for the designated object via the processor by comparing the evaluation object model representation with a reference object representation, the reference object representation being created by mapping a plurality of reference images of the designated object to the object model, the object condition information indicating one or more differences between the reference object model representation and the evaluation object model representation; and
storing a graphical representation of the object model that includes the object condition information.
19. A mobile computing device comprising:
one or more cameras configured to capture a video of a designated object as the mobile computing device moves through space around the designated object;
a processor configured to:
map a sequence of evaluation images to an object model to create an evaluation object model representation, the sequence of evaluation images being selected from the video of the designated object,
determining that the sequence of evaluation images exhibit a coverage level that is beneath a designated threshold,
providing a recording guidance instruction via a display screen at the mobile computing device to capture additional video data, and
determining object condition information for the designated object by comparing the evaluation object model representation with a reference object representation, the reference object representation being created by mapping a plurality of reference images of the designated object to the object model, the object condition information indicating one or more differences between the reference object model representation and the evaluation object model representation; and
a storage device configured to store a graphical representation of the object model that includes the object condition information.