In development projects for advanced driver-assistance systems (ADAS) and autonomous driving (AD), it is crucial to record large amounts of data in order to validate the entire function.
For example, in an L2+ project, 2 GB/s or more of data must be recorded, depending on the test task. This data comes from sensors such as LIDAR and radar, as well as from cameras and in-vehicle communication between control units.
Whether you are developing your ADAS/AD applications in a real or virtual environment: We offer comprehensive solutions consisting of software and hardware tools as well as embedded components.
Highlights
- Coordinated ADAS/AD toolchain: Suitable tools, software components, and hardware for your specific use case
- A single test tool for all development phases: Use test definitions for applications in MIL, SIL, or HIL environments throughout the entire process
- High-performance sensor and ECU integration: Acquire and log multisensor data with high data rates and optional data compression for Linux and Windows
- Efficient use of measurement data: Analysis and Rapid Prototyping of driving functions using real measurement results
- Verification of sensor objects: Both in video images of the environment and in 3D scenes via object overlay
Logging
Functional testing is an integral part of ADAS/AD development. To avoid many time-consuming and costly test drives with new software versions, sensor data is recorded in the vehicle for later MiL/SiL/HiL runs and analysis. However, this also presents challenges: terabytes of data must be stored in a harsh automotive environment. This is caused, among other things, by the wide variety of sensor types (based on radar, LIDAR, and video technology) and the highly complex and extensive ECU data that must be logged in a time-synchronized manner. Depending on the area of application, selectable compression methods ensure memory- and thus cost-optimized operation of the entire data pipeline.
Data Analysis and Function Validation
Effective data analysis plays a important role in determining a system's performance. Metadata is required to select the appropriate measurement data for the next steps. It contains information about the vehicle, the driving situation, and the software status. Once the relevant measurement data sets have been identified, they are fed into the actual analysis and validation process in the second step. The analysis itself can be performed manually, semi-automatically, or fully automatically using data mining.
Video Anonymization
Not only the collection and labeling of data, but also its proper storage and processing are relevant. Data must be securely stored, transmitted, and analyzed. In addition to technical challenges associated with recording large amounts of data, data protection requirements also play a key role. If personal data – such as faces or license plates – is being filmed, it must be additionally anonymized before it can be stored or further processed.
Visualization and Data Assessment
Data from ADAS sensors can be transmitted either as raw data via special, high-bandwidth sensor interfaces or as object data via vehicle-specific networks (e.g., CAN or Automotive Ethernet).
For the development of driver assistance systems and autonomous vehicles, sensor data is recorded for direct analysis or to be used later in a re-simulation to test new software versions. Suitable software uses sensor data to visualize a graphical representation of the detected objects in a reference image and a 2D/3D scene, which are then used for evaluation.
Setup Examples
Depending on the complexity of the functions to be measured, vehicle setups differ primarily in the number of sensors used and the resulting complexity of the networks and sensor interfaces. They all share a measurement setup that is identical at its core and can be used with different scales depending on the characteristics.
Low complexity: L2 SmartCam with front and side radars
When developing a new front camera (e.g., for AEB or LKA), the camera itself and several radar sensors are instrumented in the test vehicle to confirm proper function. In addition to this sensor data, data from the CAN buses is also recorded.
Medium complexity: 5V5R, e.g., for automated parking
A parking assistant brings together several functions: Detecting the surroundings, evaluating the situation, and executing the parking function. To test these, significantly more data must be collected. Typically, all installed sensors are acquired, meaning a front camera, a front radar, four surround cameras, and four side radars. If Automotive Ethernet is also used, this results in a significantly higher data rate during logging than with CAN.
Very high complexity: On the path to autonomous driving
Autonomous or semi-autonomous driving must function reliably and without errors, as the driver gradually hands over full control to the vehicle. For test drives to validate the function, a large number of sensors and cameras are therefore used: At least five radar sensors, eleven cameras (1x wide-angle front camera, 1x front camera, 4x side cameras, 4x surround cameras, 1x rear camera), and a LIDAR sensor on the top of the vehicle. Together with the vehicle communication (Automotive Ethernet), this results not only in a complex measurement setup but also in a very high data rate. This can be reduced using hardware compression.
