Ground Truth System
During series production development of autonomous driving perception system, we often have to answer the following questions:
In the autonomous driving function deﬁnition phase, how to validate that the sensors to be selected meet the sensing requirements?
How to evaluate sensor perception capability and performance, to update the algorithms?
LDCompass™ consists of two parts: raw data acquisition system and big data processing and validation software toolchain system. It helps the customer to quickly test and validate, evaluate and train the perception performance to ensure the products hit the market on time.
In addition, LDCompass ™ can also be used to build a natural driving scenario database for autonomous driving. It supports the definition, development and validation of scenario-based autonomous driving functions.
LDCompass™ OnBoard: Ground Truth acquisition system
The product uses the roof box design for easy installation and removal.
LiangDao Intelligence has a professional engineering team that provides modification, sensor spatial calibration and time synchronization services based on the model.
LDCompass™ToolChain: Big data processing and validation software toolchain system
Depending on the magnitude of the data storage, the customer can choose three versions of toolchain software system: Workstation-Version, Server-Version and Cloud-Platform-Version.
Supporting key functions such as data order management, Ground Truth generation, scenario mining, and perception KPI report generation.
1. LD Big Data Manager
Data visualization and order management platform. It supports the management for big data, acquisitions, labelling, sending orders, viewing scenario mining results and statistical KPI analysis reports.
2. LD Ground Truth
LD Ground Truth is a software that could generate Ground Truth automatically. The tool processes raw data collected by the LDCompass™ OnBoard, to annotate dynamic and static objects in a continuous driving environment.
Dynamic objects such as car, truck, pedestrian, motorcycle and bicycle.
Parameters such as position, direction, speed, acceleration, steering angle, time distance/spacing between objects.
Bridge detection, tunnel detection, lane detection, roadside edge detection.
3. LD Scenario
LD Scenario can automaticly detect the interaction scenarios between traffic participants in driving condition, such as cut-in, car following, lane change, etc. The scenario data is saved in the Open X standards data format.Scenario recall system supports single and combined conditions, and data can be filtered by specific numerical values and ranges of numbers. The toolchain can also quickly generate dozens of dimensions of data structure charts.
4. LD KPI
Comparing the results and Ground Truth of DuT automatically, outputting KPI report about the perception performance. LD KPI can output KPI reports of camera, radar, LiDAR, sensor fusion system, including information of object detection rate,object detection false positive rate, object location errors, object outline error, object motion information error and other dimensions. Based on customer defined requirements, the tool can also provide performance defect analysis reports for the DuTs.