Virtual testing of autonomous driving systems with thermal cameras
In one embodiment, use of virtual thermal cameras in simulations systems can rapidly increase the deployment of vehicles with autonomous driving systems that include thermal cameras as a part of the sensor systems that detect people, other cars, etc. This use can be achieved by calibrating virtual thermal cameras using a virtual object or scenes with predetermined temperatures of objects (humans) in the scenes. The virtually calibrated thermal camera can then be used in virtual simulations of driving scenarios to generate learning data, such as virtual thermal images, to test autonomous driving systems and autonomous driving algorithms. The calibration can use a temperature conversion process, and the generation of the learning data can also use this temperature conversion process that includes the addition of noise, represented by temperature values.
1 . A non-transitory machine readable medium storing executable program instructions which when executed by a data process system cause the data processing system to perform a method, the method comprising:
calibrating a virtual thermal camera using a known virtual object that has a predetermined temperature, the calibrating comprising calculating a set of sensor parameters of the virtual thermal camera based on energy data of a calibration scene containing the virtual object to produce calibrated sensor parameters that cause the virtual thermal camera to provide an output of temperature for one or more pixels in the calibration scene containing the virtual object, the temperature for the one or more pixels derived from the energy data for the one or more pixels in the calibration scene; and
generating a set of simulated virtual thermal images of a virtual scene surrounding the calibrated virtual thermal camera on a simulated vehicle that is driving through the virtual scene, the generating using the calibrated virtual thermal camera to compute a set of temperature maps of images from energy data of the virtual scene, wherein the energy data of the virtual scene is converted into pixel temperature data, wherein random noise data is added to the pixel temperature data, the set of temperature maps based on the pixel temperature data, and the set of simulated virtual thermal images being derived from the set of temperature maps.
2 . The medium as in claim 1 wherein the calculating of the set of sensor parameters comprises repeatedly calculating the set of sensor parameters in an optimization loop until a converged optimum solution for the set of sensor parameters is achieved.
3 . The medium as in claim 1 , wherein the method further comprises:
converting, as part of the calibrating, the energy data for the one or more pixels in the calibration scene into first pixel temperature data to derive first temperature data for each of the one or more pixels in the calibration scene.
4 . The medium as in claim 3 , wherein the method further comprises:
converting, as part of the generating of the set of simulated virtual thermal images, the energy data for the one or more pixels in the virtual scene into second pixel temperature data and adding the random noise data, represented by one or more temperatures, to the second pixel temperature data to derive second combined temperature data for each of the one or more pixels in the virtual scene.
5 . The medium as in claim 4 , wherein the calibrated sensor parameters are used to convert the energy data for the one or more pixels in the calibration scene into the first pixel temperature data.
6 . The medium as in claim 5 , wherein the method further comprises:
converting the second combined temperature data into energy data with added noise for the one or more pixels in the virtual scene.
7 . The medium as in claim 6 , wherein the method further comprises:
propagating virtual images through a lens model of the calibrated virtual thermal camera to generate the set of simulated virtual thermal images.
8 . The medium as in claim 6 , wherein the set of sensor parameters are camera specific parameters that are used in one or more equations to convert between energy data and temperature.
9 . The medium as in claim 8 , wherein the method further comprises:
storing the set of simulated virtual thermal images for use in testing one or more autonomous driving algorithms.
10 . The medium as in claim 9 , wherein the virtual thermal camera is a long wave infrared virtual thermal camera, and wherein the energy data is a radiative energy flux through a surface area.
11 . A machine implemented method comprising:
calibrating a virtual thermal camera using a known virtual object that has a predetermined temperature, the calibrating comprising calculating a set of sensor parameters of the virtual thermal camera based on energy data of a calibration scene containing the virtual object to produce calibrated sensor parameters that cause the virtual thermal camera to provide an output of temperature for one or more pixels in the calibration scene containing the virtual object, the temperature for the one or more pixels derived from the energy data for the one or more pixels in the calibration scene; and
generating a set of simulated virtual thermal images of a virtual scene surrounding the calibrated virtual thermal camera on a simulated vehicle that is driving through the virtual scene, the generating using the calibrated virtual thermal camera to compute a set of temperature maps of images from energy data of the virtual scene, wherein the energy data of the virtual scene is converted into pixel temperature data, wherein random noise data is added to the pixel temperature data, the set of temperature maps based on the pixel temperature data, and the set of simulated virtual thermal images being derived from the set of temperature maps.
12 . The method as in claim 11 wherein the calculating of the set of sensor parameters comprises repeatedly calculating the set of sensor parameters in an optimization loop until a converged optimum solution for the set of sensor parameters is achieved.
13 . The method as in claim 11 , wherein the method further comprises:
converting, as part of the calibrating, the energy data for the one or more pixels in the calibration scene into first pixel temperature data to derive first temperature data for each of the one or more pixels in the calibration scene.
14 . The method as in claim 13 , wherein the method further comprises:
converting, as part of the generating of the set of simulated virtual thermal images, the energy data for the one or more pixels in the virtual scene into second pixel temperature data and adding the random noise data, represented by one or more temperatures, to the second pixel temperature data to derive second combined temperature data for each of the one or more pixels in the virtual scene.
15 . The method as in claim 14 , wherein the calibrated sensor parameters are used to convert the energy data for the one or more pixels in the calibration scene into the first pixel temperature data.
16 . The method as in claim 15 , wherein the method further comprises:
converting the second combined temperature data into energy data with added noise for the one or more pixels in the virtual scene.
17 . The method as in claim 16 , wherein the method further comprises:
propagating virtual images through a lens model of the calibrated virtual thermal camera to generate the set of simulated virtual thermal images.
18 . The method as in claim 16 , wherein the set of sensor parameters are camera specific parameters that are used in one or more equations to convert between energy data and temperature.
19 . The method as in claim 18 , wherein the method further comprises:
storing the set of simulated virtual thermal images for use in testing one or more autonomous driving algorithms.
20 . The method as in claim 19 , wherein the virtual thermal camera is a long wave infrared virtual thermal camera, and wherein the energy data is a radiative energy flux through a surface area.