Back to results
US 11,106,903 B1Granted
Object detection in image data
AssigneeAmazon Technologies, Inc.
InventorsCong Phuoc Huynh, Mashhour Solh, Tiantong Guo
Life of this patent
Estimating the term…
Abstract
Techniques are generally described for object detection in image data. A first frame of image data associated with a first domain is received by a detector executing on at least one computing device. The detector generates a first feature data in the first domain. The first feature data is transformed from the first domain into a second feature data in a second domain. The detector may be effective to detect objects in the second domain. A location of an object in the first frame of image data is determined based at least in part on the second feature data.
Claims20
A computer-implemented method of object detection, comprising:
- receiving, by a single shot detector (SSD) executing on at least one computing device, a first frame of RGB image data, the SSD configured to detect humans represented in image data;
- generating, by a first layer of the SSD, first feature data in an RGB domain, the first feature data representing the first frame of RGB image data;
- determining, by the SSD, a first plurality of contiguous pixels representing a first human from the first feature data;
- generating, by an output layer of the SSD, a first bounding box defining the first plurality of contiguous pixels in the first frame of RGB image data;
- receiving, by the SSD, a first frame of near infrared (NIR) image data;
- generating, by the first layer of the SSD, second feature data in an NIR domain, the second feature data representing the first frame of NIR image data;
- sending, by the SSD, the second feature data to a generator neural network configured to transform NIR feature data into RGB feature data;
- transforming, by the generator neural network, the second feature data to a third feature data, wherein the third feature data represents first synthetic RGB feature data in the RGB domain;
- determining, by the SSD, a second plurality of contiguous pixels representing a second human from the third feature data;
- and
- generating, by the output layer of the SSD, a second bounding box defining the second plurality of contiguous pixels representing the second human in the first frame of NIR image data.
8 more elements
A computer-implemented method of object detection, comprising:
- receiving, by a detector, a first frame of image data, the first frame of image data being in a first domain;
- generating, by the detector, first feature data comprising a representation of the first frame of image data in the first domain;
- determining, by the detector, a first plurality pixels representing a first object based at least in part on the first feature data;
- generating, by the detector, a first bounding box defining a location of the first plurality of pixels in the first frame of image data;
- receiving, by the detector, a second frame of image data, the second frame of image data being in a second domain different from the first domain;
- generating, by the detector, second feature data comprising a representation of the second frame of image data in the second domain;
- sending the second feature data to a generator configured to transform feature representations of image data from the second domain to the first domain;
- transforming the second feature data from the second domain into third feature data in the first domain;
- determining, by the detector, a second plurality of pixels representing a second object from the third feature data;
- and
- generating, by the detector, a second bounding box defining a location of the second plurality of pixels in the second frame of image data.
8 more elements
A system comprising:
- at least one processor;
- and
- a non-transitory, computer-readable memory storing instructions that when executed by the at least one processor program the at least one processor to perform a method comprising:
- receiving, by a detector, a first frame of image data, the first frame of image data being in a first domain;
- generating, by the detector, first feature data comprising a representation of the first frame of image data in the first domain;
- determining, by the detector, a first plurality pixels representing a first object based at least in part on the first feature data;
- generating, by the detector, a first bounding box defining a location of the first plurality of pixels in the first frame of image data;
- receiving, by the detector, a second frame of image data, the second frame of image data being in a second domain different from the first domain;
- generating, by the detector, second feature data comprising a representation of the second frame of image data in the second domain;
- sending the second feature data to a generator configured to transform feature representations of image data from the second domain to the first domain;
- transforming the second feature data from the second domain into third feature data in the first domain;
- determining, by the detector, a second plurality of pixels representing a second object from the third feature data;
- and
- generating, by the detector, a second bounding box defining a location of the second plurality of pixels in the second frame of image data.
11 more elements
Dependencies are inferred from the claim text.