Back to results
US 10,949,982 B1Granted
Moving object recognition, speed estimation, and tagging
AssigneeAmazon Technologies, Inc.
InventorsQia Wang, Siva Somasundaram, Hefeng Chen, Vijay Gopal Joshi
Life of this patent
Estimating the term…
Abstract
Techniques for estimating speed for an object in a video (either stream or file) are described. In particular, once an object has been recognized, additional data (such as size) for that recognized object is used in a determination of a 3D position of that object in a plurality of frames. An estimated speed may be found using a difference between 3D positions of the object in different frames over the time between the frames.
Claims20
A computer-implemented method comprising:
- receiving a video at a video processing service of a provider network;
- processing the received video using the video processing service of the provider network to associate individual frames of the video with individual timestamps;
- receiving request to determine an estimated speed of one or more objects of the received video;
- detecting and recognizing one or more objects of interest in at least two frames of the video to be used in determining an estimated speed using an object recognition service;
- querying for a size of the one or more recognized objects of interest;
- calculating, using a hosted machine learning model, for at least the two frames of the video, a 3-dimensional (3D) position of the object of interest, using intrinsic values for a source of the video, based on the size of the recognized object of interest and bounding box for the recognized object of interest;
- calculating an estimated speed of the recognized object of interest based upon a difference between 3D positions of the recognized object of interest in two frames and a time difference between said frames;
- adding a tag to metadata of the video for the object including the estimated speed of the recognized object of interest;
- and
- providing the estimated speed to a requester from the metadata.
7 more elements
A computer-implemented method comprising:
- receiving a video;
- receiving request to determine an estimated speed of one or more objects of the received video;
- recognizing one or more objects of interest in at least two frames of the video to be used in determining an estimated speed;
- querying for a size of the one or more recognized objects of interest;
- calculating, for at least the two frames of the video, a 3-dimensional (3D) position of the object of interest using intrinsic values for a source of the video based on the size of the recognized object of interest and points of interest on the recognized object of interest;
- calculating an estimated speed of the recognized object of interest based upon a difference between 3D positions of the recognized object of interest in two frames and a time difference between said frames;
- adding a tag to metadata of the video for the recognized object including the estimated speed of the object of interest;
- and
- providing the estimated speed to a requester from the metadata.
6 more elements
A system comprising:
- an object recognition service implemented by a first one or more electronic devices to recognize one or more objects of interest in at least two frames of a video to be used in determining an estimated speed and position of the one or more objects of interest;
- an estimated speed and estimated 3D position service implemented by a second one or more electronic devices, the estimated speed and estimated 3D position service including instructions that upon execution cause the estimated speed and estimated 3D position service to:
- calculate for at least the two frames of the video, a 3-dimensional (3D) position of the object of interest, using intrinsic values for a source of the video, based on the size of the recognized object of interest and points of interest of the recognized object of interest,
- calculate an estimated speed of the recognized object of interest based upon a difference between 3D positions of the recognized object of interest in two frames and a time difference between said frames, and
- add a tag to metadata of the video for the object including the estimated speed of the recognized object of interest.
2 more elements
Dependencies are inferred from the claim text.