In this paper camera raw data injection and scenery simulation are combined to realize intrinsic and extrinsic calibration to improve confidence level and put forward a standard debugging process proposal for camera simulation. The same view perception board are tested in hardware-in-the-loop (HIL) simulation system and real vehicle, while lens and image sensor are simulated in HIL system. Static object range measurement results reported by the camera system are compared to get the similarity between camera raw data injection simulation and real vehicle test. We found the average range similarity is over 90% when the object stays in nearest three lanes.
We used to try to set all parameters exactly the same with the device under test (DUT). The old way requires simulation software support perfect parameter settings of camera and costs great effort debugging. In this paper, we consider the virtual camera (virtual lens and image sensor with real view perception board) as one of the same production batch products with the real camera. As a result, the virtual camera should meet the manufacturing tolerances, which is satisfied by using designed parameters. The corresponding onboard software is supposed to be calibrated before doing HIL test. This method reduces requirements for simulation software. Validation engineers only need to fix the virtual camera on virtual vehicle as the specification required and finish the calibration, just same with what they do on real vehicle.
This camera raw data injection calibration method can be reproduced on most commercial simulation platforms and can work for most monocular camera products. High confidence level and standard operation will make camera raw data injection more popular and reliable for camera and autonomous driving simulation test.