Methods and systems for performing object dimensioning
Aspects of the present disclosure include isolating a target scene object from multiple images of a time-of-flight (TOF) sensor. Multiple images from a TOF sensor can be transformed into multiple three-dimensional (3D) point clouds. In each of the multiple 3D point clouds, one or more large planes having a minimal threshold size can be identified. A common 3D point cloud can be generated by aligning rotated and translated point clouds corresponding to at least a portion of the multiple 3D point clouds. The common 3D point cloud can be rotated and translated to have one of one or more large planes identified as a ground plane within a common coordinate system of the common 3D point cloud, and a target object can be isolated.
1 . A method for isolating a target scene object from multiple images of a time-of-flight (TOF) sensor, comprising:
transforming the multiple images from a TOF sensor into multiple three-dimensional (3D) point clouds;
identifying, in each of the multiple 3D point clouds, multiple large planes having a minimal threshold size;
identifying one of the multiple large planes as a ground plane;
generating a common 3D point cloud by aligning rotated and translated point clouds corresponding to at least a portion of the multiple 3D point clouds;
rotating and translating the common 3D point cloud to have the ground plane within a common coordinate system of the common 3D point cloud; and
isolating a target object from the common 3D point cloud as rotated and translated, including:
projecting an average location of reconstructed TOF frusta in the common 3D point cloud into the ground plane as an origin;
averaging a pointing direction for the reconstructed TOF frusta within the ground plane; and
modifying the common 3D point cloud by removing points lying outside a subcloud annulus of the common 3D point cloud between two ground plane circles centered on the origin, and removing points outside a subcloud azimuthal wedge defined by a specified number of degrees from the pointing direction.
2 . The method of claim 1 , wherein isolating the target object includes selecting a subcloud having points closest in position to a preferred axis defined by the pointing direction from the origin.
3 . The method of claim 1 , wherein isolating the target object includes further removing points from the common 3D point cloud that are associated with the multiple large planes.
4 . The method of claim 1 , wherein isolating the target object includes:
converting the common 3D point cloud to a height map; and
removing points from the common 3D point cloud that are associated with objects less than a threshold size.
5 . The method of claim 1 , further comprising displaying, on a screen interface, the target object as isolated from the common 3D point cloud.
6 . The method of claim 1 , further comprising measuring a width, length, and height dimension of the target object as isolated from the common 3D point cloud.
7 . The method of claim 6 , further comprising searching a seven dimensional parameter space for a bounding box that surrounds the target object, wherein measuring the width, length, and height of the target object is based on the width, length, and height of the bounding box.
8 . The method of claim 7 , further comprising displaying, on a screen interface, the common 3D point cloud and the bounding box surrounding the target object.
9 . The method of claim 8 , wherein displaying the target object includes jointly displaying the common 3D point cloud and the bounding box surrounding the target object within three orthographic projections.
10 . A system comprising:
a memory; and
a processor coupled to the memory, wherein the processor is configured to:
transform multiple images from a time-of-flight (TOF) sensor into multiple three-dimensional (3D) point clouds;
identify, in each of the multiple 3D point clouds, multiple large planes having a minimal threshold size;
identify one of the multiple large planes as a ground plane;
generate a common 3D point cloud by aligning rotated and translated point clouds corresponding to at least a portion of the multiple 3D point clouds;
rotate and translate the common 3D point cloud to have the ground plane within a common coordinate system of the common 3D point cloud; and
isolate a target object from the common 3D point cloud as rotated and translated including:
projecting an average location of reconstructed TOF frusta in the common 3D point cloud into the ground plane as an origin;
averaging a pointing direction for the reconstructed TOF frusta within the ground plane; and
modifying the common 3D point cloud by removing points lying outside a subcloud annulus of the common 3D point cloud between two ground plane circles centered on the origin, and removing points outside a subcloud azimuthal wedge defined by a specified number of degrees from the pointing direction.
11 . The system of claim 10 , wherein the processor is configured to isolate the target object at least in part by selecting a subcloud having points closest in position to a preferred axis defined by the pointing direction from the origin.
12 . The system of claim 10 , wherein the processor is configured to isolate the target object at least in part by further removing points from the common 3D point cloud that are associated with the multiple large planes.
13 . The system of claim 10 , wherein the processor is configured to isolate the target object at least in part by:
converting the common 3D point cloud to a height map; and
removing points from the common 3D point cloud that are associated with objects less than a threshold size.
14 . The system of claim 10 , wherein the processor is further configured to display, on a screen interface, the target object as isolated from the common 3D point cloud.
15 . The system of claim 10 , wherein the processor is further configured to measure a width, length, and height dimension of the target object as isolated from the common 3D point cloud.
16 . The system of claim 15 , wherein the processor is further configured to search a seven dimensional parameter space for a bounding box that surrounds the target object, wherein measuring the width, length, and height of the target object is based on the width, length, and height of the bounding box.
17 . A non-transitory computer-readable medium, comprising code executable by one or more processors for isolating a target scene object from multiple images of a time-of-flight (TOF) sensor, the code comprising code for:
transforming the multiple images from a TOF sensor into multiple three-dimensional (3D) point clouds;
identifying, in each of the multiple 3D point clouds, multiple large planes having a minimal threshold size;
identifying one of the multiple large planes as a ground plane;
generating a common 3D point cloud by aligning rotated and translated point clouds corresponding to at least a portion of the multiple 3D point clouds;
rotating and translating the common 3D point cloud to have the ground plane within a common coordinate system of the common 3D point cloud; and
isolating a target object from the common 3D point cloud as rotated and translated, including:
projecting an average location of reconstructed TOF frusta in the common 3D point cloud into the ground plane as an origin;
averaging a pointing direction for the reconstructed TOF frusta within the ground plane; and
modifying the common 3D point cloud by removing points lying outside a subcloud annulus of the common 3D point cloud between two ground plane circles centered on the origin, and removing points outside a subcloud azimuthal wedge defined by a specified number of degrees from the pointing direction.