Back to results
US 12,586,356 B2Granted
Artificial image generation with traffic signs
AssigneeFord Global Technologies, LLC
InventorsVanessa Ferreira Dalborgo, João Gabriel Lima Moraes, Vinicius Souza Madureira, Vitor Magno De Oliveira Santos Bezerra
Life of this patent
Estimating the term…
Abstract
A computer includes a processor and a memory, and the memory stores instructions executable by the processor to receive a plurality of initial images including traffic signs, identify the traffic signs in the initial images, and generate new images by applying manipulations to the traffic signs in the initial images. The manipulations simulate physical changes to the traffic signs.
Claims20
A computer comprising a processor and a memory, the memory storing instructions executable by the processor to:
- receive a plurality of initial images including traffic signs;
- identify the traffic signs in the plurality of initial images;
- generate a plurality of new images by applying manipulations to the traffic signs in the plurality of initial images, the manipulations simulating physical changes to the traffic signs, the physical changes including at least one of vegetation spaced between the respective traffic sign and a camera recording the respective initial image, graffiti on the traffic signs, physical bending of the traffic signs limited to subparts of the initial images including the traffic signs, erasure of symbols on the traffic signs limited to the subparts of the initial images including the traffic signs, or corrosion of the traffic signs limited to the subparts of the initial images including the traffic signs;
- and
- train a machine-learning model on training data, wherein the training data includes the plurality of new images;
- wherein the instructions to apply the manipulations include instructions to, for one of the plurality of initial images, at least one of overlay a template on the respective traffic sign, apply a geometric transformation to a first portion of the respective initial image containing the respective traffic sign, or replace the first portion of the respective initial image containing the respective traffic sign with a second portion of the respective initial image not containing the respective traffic sign.
3 more elements
A method comprising:
- receiving a plurality of initial images including traffic signs;
- identifying the traffic signs in the plurality of initial images;
- generating a plurality of new images by applying manipulations to the traffic signs in the plurality of initial images, the manipulations simulating physical changes to the traffic signs, the physical changes including at least one of vegetation spaced between the respective traffic sign and a camera recording the respective initial image, graffiti on the traffic signs, physical bending of the traffic signs limited to subparts of the initial images including the traffic signs, erasure of symbols on the traffic signs limited to the subparts of the initial images including the traffic signs, or corrosion of the traffic signs limited to the subparts of the initial images including the traffic signs;
- and
- training a machine-learning model on training data, wherein the training data includes the plurality of new images;
- wherein the instructions to apply the manipulations include instructions to, for one of the plurality of initial images, at least one of overlay a template on the respective traffic sign, apply a geometric transformation to a first portion of the respective initial image containing the respective traffic sign, or replace the first portion of the respective initial image containing the respective traffic sign with a second portion of the respective initial image not containing the respective traffic sign.
3 more elements
A computer comprising a processor and a memory, the memory storing instructions executable by the processor to:
- receive a plurality of initial images including traffic signs;
- identify the traffic signs in the plurality of initial images;
- generate a plurality of new images by applying manipulations to the traffic signs in the plurality of initial images, the manipulations simulating physical changes to the traffic signs, the physical changes including at least one of vegetation spaced between the respective traffic sign and a camera recording the respective initial image, graffiti on the traffic signs, physical bending of the traffic signs limited to subparts of the initial images including the traffic signs, erasure of symbols on the traffic signs limited to the subparts of the initial images including the traffic signs, or corrosion of the traffic signs limited to the subparts of the initial images including the traffic signs;
- and
- train a machine-learning model on training data, wherein the training data includes the plurality of new images;
- wherein the physical changes include corrosion of the traffic signs limited to the subparts of the initial images including the traffic signs.
3 more elements
Dependencies are inferred from the claim text.