Hamburg-based automotive LiDAR specialist, Ibeo Automotive Systems has developed a new automated solution for the validation of sensors necessary for ADAS and autonomous driving.
The ibeo.Reference toolchain also automates and specifies the manual data labelling process, which is personnel and time-intensive.
ADAS and AD must be validated in a complex process before they ultimately can be installed in the vehicle, ready for series production. Regardless of the technology of the sensor system (camera, radar, or LiDAR), reference data is required for each system and must be compared with the data acquired during the test process.
Ibeo’s recording system is installed on the vehicle alongside the manufacturer’s prototype system and records data in parallel. Furthermore, it can also be used to create scenarios for ADAS or AD algorithm-development.
The recorded data collected in validation drives can be in the low-to-mid thousands of hours. And for every hour of test driving, manual evaluation results in up to 300 hours of labelling activity.
“Manual reconciliation of this data is currently a time-consuming and cost-intensive factor in many new developments,” said ibeo.Reference product manager, Philip Neitzel.
To classify the large amounts of data recorded, the industry traditionally uses outsourced service providers. In many cases, hundreds of employees manually evaluate every metre of road to determine whether the sensor being tested has detected cars, people, lane markings or guard rails, for example, to reconcile the data.
The ibeo Validation Bundle in Ibeo’s new toolchain tackles this task: “As part of the ibeo Validation Bundle, we offer an automated overall solution in the form of the ibeo Auto Annotation Software, which algorithmically generates reference data from the incoming data,” added Neitzel.
“If required, these can then be manually processed to produce ground-truth data, that is, data that fully represent the vehicle environment in a realistic manner. In this way, we are solving one of the main challenges we currently face on the way to a series product.”
With the new ibeo.Reference toolchain, the user receives the necessary building blocks for sensor validation from a single source:
- The recording system, consisting of hardware and software, for running in the reference data, which now also supports selected third-party LiDAR sensors for the first time
- Ibeo Auto Annotation as scalable, cloud-compatible software for automated creation of labelled reference data Editors for further processing of reference data to ground-truth data
“It was very important for us to offer a complete system that covers almost all challenges on the customer side in terms of how the LiDAR components can be combined and that is regularly expanded technically,” noted Neitzel.
“In this way, we help our customers to develop faster, more precisely and more cost-effectively and thus come a great deal closer to the overall vision of fully automated mobility.
The ibeo.Reference toolchain is available now. In addition, Ibeo offers a test licence, which can be requested at firstname.lastname@example.org.