Mixed depth object detection

1. A method, comprising:

obtaining (i) a point cloud captured by a depth sensor, and (ii) image data captured by an image sensor, the point cloud and the image data representing a support structure bearing a set of objects;

detecting, from the image data, an image boundary corresponding to an object from the set of objects;

determining a portion of the point cloud corresponding to the image boundary;

selecting, from the determined portion, a subset of points corresponding to a forward surface of the object; and

generating a three-dimensional position of the object based on the image boundary and the subset of points corresponding to the forward surface.

2. The method of claim 1 , wherein the portion of the point cloud corresponds to a portion of a camera field of view.

3. The method of claim 1 , wherein selecting the subset of points includes allocating the portion of the point cloud to bins corresponding to successive depths; and selecting the bin having the greatest number of points allocated thereto.

4. The method of claim 1 , further comprising:

obtaining a second image boundary;

generating a second three-dimensional position of the object based on the second image boundary;

determining that the second image boundary corresponds to the object; and

generating a combined three-dimensional position based on the three-dimensional position and the second three-dimensional position.

5. The method of claim 4 , wherein generating the combined three-dimensional position includes:

projecting each of the three-dimensional position and the second three-dimensional position to a sequence of candidate depths;

determining, at each candidate depth, a cost function corresponding to the three-dimensional position and the second three-dimensional position;

selecting the candidate depth yielding the lowest cost function; and

generate the combined three-dimensional position at the selected candidate depth.

6. The method of claim 4 , wherein determining that the second image boundary corresponds to the object includes comparing object metadata associated with the image boundary and the second image boundary.

7. The method of claim 6 , wherein the object metadata includes barcode data.

8. The method of claim 5 , wherein the cost function is a distance between the projections of the three-dimensional position and the second three-dimensional position.

9. The method of claim 5 , wherein the combined three-dimensional position is an average of the projections of the three-dimensional position and the second three-dimensional position to the selected candidate depth.