Object recognition enhancement using depth data
Methods for improving object recognition using depth data are disclosed. An image is captured of a 3-D scene along with depth data, such as in the form of a point cloud. The depth data is correlated with the image of the captured scene, such as by determining the frame of reference of each of the image and the depth data, thereby allowing the depth data to be mapped to the correct corresponding pixels of the image. Object recognition on the image is then improved by employing the correlated depth data. The depth data may be captured contemporaneously with the image of the 3-D scene, such as by using photogrammetry, or at a different time.
1. A method, comprising:
capturing, using a camera, a video of a scene;
detecting, within a reference frame from the video of the scene, a location of one or more objects;
obtaining depth data of the scene by capturing the depth data contemporaneously with capturing the video of the scene; obtaining motion and location data of the camera; correlating, using the motion and location data, positions of points of the depth data to the reference frame; correlating, using at least the correlated positions of the depth data points, the depth data with the one or more detected objects; and performing object recognition on the one or more detected objects using the depth data.
2. The method of claim 1 , wherein obtaining the depth data of the scene comprises capturing the depth data at a different time from capturing the scene video.
3. The method of claim 1 , wherein performing object recognition further comprises performing object recognition on the scene using a model library of 3-D objects.
4. The method of claim 1 , further comprising performing edge detection upon the scene, and correlating one or more depth points in the depth data with one or more detected edges.
5. A non-transitory computer-readable medium (CRM) comprising instructions that, when executed, cause the CRM to:
capture, using a camera, a video of a scene;
detect, within a reference frame from the video of the scene, a location of one or more objects;
obtaining depth data of the scene by capturing the depth data contemporaneously with capturing the video of the scene; obtain motion and location data of the camera; correlate, using the motion and location data, positions of points of the depth data to the reference frame from the scene video; correlate, using at least the correlated positions of the depth data points, the depth data with the one or more detected objects; and perform object recognition on the one or more detected objects with the depth data.
6. The CRM of claim 5 , wherein the instructions are to further cause the apparatus to perform edge detection upon the scene, and correlate one or more depth points in the depth data with one or more detected edges.
7. The CRM of claim 5 , wherein the instructions are to further cause the apparatus to capture the depth data of the scene.
8. The CRM of claim 5 , wherein the instructions are to further cause the apparatus to perform object recognition on the scene using a model library of 3-D objects.
9. An apparatus, comprising: an image store;
and an object detector, wherein: the image store is to receive a video of a scene captured by a camera, detecting, within a reference frame from the video of the scene, a location of one or more objects;
obtaining depth data of the scene by capturing the depth data contemporaneously with capturing the video of the scene; obtaining motion and location data of the camera, and a 3-D point cloud, the object detector is to detect one or more objects within the reference frame from the video, the object detector is to correlate, using the motion and location data, the 3-D point cloud to the one or more objects detected within the reference frame, and the object detector is to recognize the one or more objects based at least the 3-D point cloud and the depth data.
10. The apparatus of claim 9 , wherein the image store is to receive the video and 3-D point cloud from a mobile device.
11. The apparatus of claim 10 , wherein the 3-D point cloud is captured contemporaneously with the video.
12. The apparatus of claim 9 , wherein the object detector is to further detect objects within the video based at least in part upon a model library of 3-D objects.
13. The apparatus of claim 9 , wherein the apparatus is a mobile device.