Realistic 3D virtual world creation and simulation for training automated driving systems
A computer implemented method of creating a simulated realistic virtual model of a geographical area for training an autonomous driving system, comprising obtaining geographic map data of a geographical area, obtaining visual imagery data of the geographical area, classifying static objects identified in the visual imagery data to corresponding labels to designate labeled objects, superimposing the labeled objects over the geographic map data, generating a virtual 3D realistic model emulating the geographical area by synthesizing a corresponding visual texture for each of the labeled objects and injecting synthetic 3D imaging feed of the realistic model to imaging sensor(s) input(s) of the autonomous driving system controlling movement of an emulated vehicle in the realistic model where the synthetic 3D imaging feed is generated to depict the realistic model from a point of view of emulated imaging sensor(s) mounted on the emulated vehicle.
1 . A system for validating a driving system, comprising:
a data processor configured to:
obtain a sequence of images captured while traveling in a specific geographic area,
create an output feed by adjusting the sequence of images using at least one neural network and at least one type of environmental conditions selected from a group consisting of: lighting conditions, weather conditions, and timing attributes,
wherein creating the output feed comprises:
classifying a plurality of objects identified in the sequence of images to corresponding labels using one or more of the at least one neural network, and
synthesizing a corresponding visual image pixel data using the corresponding labels of one or more of the plurality of objects, inject the output feed into:
(a) the driving system, or
(b) one or more virtual drivers provided for the driving system, and
evaluate performance of the driving system based on a response of the driving system to the injected output feed.
2 . The system of claim 1 , wherein the at least one neural network comprises one or more generative neural networks applied in synthesizing the corresponding visual image pixel data for generating the output feed.
3 . The system of claim 1 , wherein the output feed is injected into one or more physical sensors of the driving system or one or more physical sensors of a vehicle connected to the driving system.
4 . The system of claim 1 , wherein the environmental conditions are selected by a user.
5 . The system of claim 1 , wherein the data processor is further configured to remove objects from the output feed.
6 . The system of claim 1 , wherein the data processor is further configured to insert objects into the output feed.
7 . The system of claim 1 , wherein the data processor is further configured to adjust the output feed based on a sensory ranging data captured with the sequence of images while traveling in the specific geographic area.
8 . The system of claim 7 , wherein the sensory ranging data is generated by applying one or more noise patterns associated with one or more range sensors.
9 . The system of claim 1 , wherein a software interface is used for injecting the output feed into the driving system.
10 . The system of claim 1 , wherein the data processor is further configured to adjust lighting conditions in the output feed, including simulating low sun conditions that produce glare.
11 . The system of claim 1 , wherein the data processor is further configured to modify objects in the output feed, including adding, removing, or changing lane markings.
12 . The system of claim 1 , wherein the data processor is further configured to allow a user to select specific weather and lighting conditions for adjusting the output feed.
13 . The system of claim 1 , wherein the data processor is further configured to insert one or more dynamic objects into the output feed, wherein the dynamic objects include at least one of: a ground vehicle, an aerial vehicle, a naval vehicle, a pedestrian, an animal, vegetation, or a dynamically changing road infrastructure object.
14 . The system of claim 13 , wherein the data processor is further configured to apply one or more driver behavior classes for controlling movement of one or more ground vehicles, wherein the driver behavior classes are adapted to the geographical area according to an analysis of typical driver behavior patterns identified in the geographical area.
15 . The system of claim 1 , wherein the data processor is further configured to inject simulated motion data to the driving system, wherein the simulated motion data is emulated by one or more emulated motion sensors associated with an emulated vehicle, the simulated motion data comprising one or more motion parameters including at least one of: a speed parameter, an acceleration parameter, a direction parameter, an orientation parameter, or an elevation parameter.
16 . The system of claim 1 , wherein the data processor is further configured to inject simulated transport data to the driving system, wherein the simulated transport data comprises Vehicle to Anything (V2X) communication between an emulated vehicle and one or more other entities.
17 . The system of claim 1 , wherein the data processor is further configured to adjust one or more mounting attributes of one or more emulated imaging sensors according to analysis of a visibility performance of the one or more emulated imaging sensors, wherein the mounting attributes include at least one of: a positioning on an emulated vehicle, a Field Of View (FOV), a resolution, or an overlap region with one or more adjacent imaging sensors.
18 . The system of claim 1 , wherein the output feed is generated using a virtual realistic model that includes one or more dynamic objects, and the data processor is further configured to apply movement to the one or more dynamic objects according to movement patterns predefined or learned for the geographical area.
19 . The system of claim 18 , wherein the data processor is further configured to:
obtain geographic map data of the specific geographic area;
obtain visual imagery data of the specific geographic area;
classify a plurality of static objects identified in the visual imagery data to corresponding labels using one or more of the at least one neural network; and
generate a virtual three dimensional (3D) realistic model by synthesizing a corresponding visual image pixel data using the corresponding labels of one or more of the plurality of labeled static objects.
20 . A method for validating a driving system, comprising:
obtaining, by a data processor, a sequence of images captured while traveling in a specific geographic area;
creating, by the data processor, an output feed by adjusting the sequence of images using at least one neural network and at least one type of environmental conditions selected from a group consisting of: lighting conditions, weather conditions, and timing attributes;
wherein creating the output feed comprises:
classifying a plurality of objects identified in the sequence of images to corresponding labels using one or more of the at least one neural network, and
synthesizing a corresponding visual image pixel data using the corresponding labels of one or more of the plurality of objects, injecting, by the data processor, the output feed into:
(a) the driving system, or
(b) one or more virtual drivers provided for the driving system; and
evaluating, by the data processor, performance of the driving system based on a response of the driving system to the injected output feed.