Autonomous driving method and apparatus
View Patent ↗The present disclosure provides an autonomous driving method and an apparatus. The method includes: receiving a currently collected image transmitted by a unmanned vehicle, where the currently collected image is an image collected in a target scenario; acquiring current driving data according to the currently collected image and a pre-trained autonomous driving model, where the autonomous driving model is used to indicate a relationship between an image and driving data in at least two scenarios, and the at least two scenarios include the target scenario; and sending the current driving data to the unmanned vehicle. Robustness of the unmanned driving method is improved.
1. An autonomous driving method, comprising:
receiving a currently collected image sent by an unmanned vehicle, wherein the currently collected image is an image collected in a target scenario;
acquiring current driving data according to the currently collected image and one pre-trained autonomous driving model, wherein the one autonomous driving model is used to indicate a relationship between an image and driving data in at least two scenarios, and the at least two scenarios comprise a known scenario and the target scenario; and
sending the current driving data to the unmanned vehicle;
wherein the method requires before the acquiring of current driving data according to the currently collected image and one pre-trained autonomous driving model, comprising:
generating a sample image of the target scenario according to a sample image of the known scenario; and
acquiring the one autonomous driving model according to the sample image of the known scenario and the sample image of the target scenario;
wherein the generating a sample image of the target scenario according to a sample image of the known scenario comprises:
training one model with a generative adversarial network (GAN) technique to obtain one domain conversion model;
acquiring the sample image of the target scenario according to the sample image of the known scenario and the one pre-trained domain conversion model, wherein the one domain conversion model is used to indicate a relationship between an image of the known scenario and an image of the target scenario;
wherein the sample image of the known scenario is a daytime sample image, and the sample image of the target scenario is a night sample image, and wherein the sample image of the known scenario is further a sunny-day sample image, and the sample image of the target scenario is further a rainy-day sample image; and
the acquiring the sample image of the target scenario according to the sample image of the known scenario and the one pre-trained domain conversion model comprises:
acquiring the night sample image according to the daytime sample image and the one pre-trained domain conversion model, wherein the one domain conversion model is used to indicate a relationship between a daytime image and a night image; and
acquiring the rainy-day sample image according to the sunny-day sample image and the one pre-trained domain conversion model, wherein the one domain conversion model is used to indicate a relationship between a sunny-day image and a rainy-day image.
2. The method according to claim 1 , wherein the sample image of the known scenario is further a sample image in which there is no pedestrian and vehicle in front of a vehicle, and the sample image of the target scenario is further a sample image in which there is a pedestrian and a vehicle in front of the vehicle; and
the generating a sample image of the target scenario according to a sample image of the known scenario comprises: performing vehicle marking and pedestrian marking on the sample image in which there is no pedestrian and vehicle in front of the vehicle to obtain the sample image in which there is a pedestrian and a vehicle in front of the vehicle.
3. A nonvolatile memory having a computer program stored thereon, wherein the computer program implements the method according to claim 1 when executed by a processor.
4. The method according to claim 1 , wherein the known scenario differs from the target scenario in terms of time periods, weather, and whether there is a pedestrian and a vehicle in front of the unmanned vehicle.
5. An autonomous driving apparatus, comprising:
a processor; and
a memory, configured to store an executable instruction of the processor;
wherein the executable instruction, when executed by the processor, causes the processor to:
receive a currently collected image sent by an unmanned vehicle, wherein the currently collected image is an image collected in a target scenario;
acquire current driving data according to the currently collected image and one pre-trained autonomous driving model, wherein the one autonomous driving model is configured to indicate a relationship between an image and driving data in at least two scenarios, and the at least two scenarios comprise a known scenario and the target scenario; and
send the current driving data to the unmanned vehicle;
wherein before causing the processor to acquire current driving data according to the currently collected image and one pre-trained autonomous driving model, the executable instruction, when executed by the processor, further causes the processor to:
generate a sample image of the target scenario according to a sample image of the known scenario; and
acquire the one autonomous driving model according to the sample image of the known scenario and the sample image of the target scenario;
wherein the executable instruction, when executed by the processor, further causes the processor to:
train one model with a generative adversarial network (GAN) technique to obtain one domain conversion model;
acquire the sample image of the target scenario according to the sample image of the known scenario and the one pre-trained domain conversion model, wherein the one domain conversion model is used to indicate a relationship between an image of the known scenario and an image of the target scenario, and wherein the sample image of the known scenario is a daytime sample image, and the sample image of the target scenario is a night sample image, and wherein the sample image of the known scenario is further a sunny-day sample image, and the sample image of the target scenario is further a rainy-day sample image; and
wherein the executable instruction, when executed by the processor, further causes the processor to:
acquire the night sample image according to the daytime sample image and the one pre-trained domain conversion model, wherein the one domain conversion model is used to indicate a relationship between a daytime image and a night image; and
acquire the rainy-day sample image according to the sunny-day sample image and the one pre-trained domain conversion model, wherein the one domain conversion model is used to indicate a relationship between a sunny-day image and a rainy-day image.