truth system LDCompass®
When developing autonomous driving perception, we need to answer the following questions:
How to verify that a new sensor meets the functional requirements during the sensing function definition stage?
In the development and verification of perception algorithms, how to objectively evaluate the shortcomings of existing algorithms and quickly optimize iterative algorithms?
The truth value system launched by Liangdao Intelligence LDCompass®, including two major modules of true value acquisition, data management and post-processing tool chain, support rapid test verification, evaluation, training perception ability, and ensure the delivery of autonomous driving mass production projects on schedule。
At the same time, the ground truth system can also be used for the construction of automatic driving natural driving scene library, and supports the definition, development and verification of automatic driving functions based on the scene.
LDCompass® OnBoard truth acquisition system
Roof box design makes installation and disassembly more convenient
Professional team provides engineering adaptation, sensor calibration and time synchronization services
LDCompass®ToolChain Data management and post-processing toolchain
According to the level of data storage, three toolchain versions are provided: PC, server and cloud platform
Support key functions such as data work order management, truth value establishment, scene mining, and perception KPI report generation
1 LD Big Data Manager
Visual data and work order management platform. Supports test big data management, collection task management, labeling task management, submitting service work orders, viewing scene mining results and KPI statistical analysis reports.
2 LD Ground Truth Automatic truth value establishment
Automatically label dynamic and static objects in point clouds and video data, supporting:
Annotation of dynamic objects such as cars, trucks, pedestrians, motorcycles, bicycles, etc.
Continuously output the position, direction, speed, acceleration, steering angle, time distance/spacing and other parameters of dynamic objects
Annotate static target information such as road edges, road signs, tunnels, high-speed entrances and exits
3 LD Scenario Automated scene extraction and analysis
Automatically detect the interaction events of traffic participants in the driving scene (such as cutting, following, changing lanes, etc.), and save it in the internationally accepted Open X series standard data format. Scene retrieval supports a single condition and condition combination, and can filter data according to specific values and value ranges, and output data structure charts of multiple dimensions.
4 LD KPI Perception assessment KPI analysis
Automatically compare the perception result of the sensor under test with the true value, and output the KPI report of perception performance. Report dimensions include but are not limited to target recognition rate, target missed recognition rate, target position error, target contour error, target motion information error, etc. The tool can also be used to customize a defect analysis report for the identification performance of the sensor under test.
The first Third-party LiDAR Testing Joint Laboratory in China
Controlled Environment Performance Testing
Closed-field Performance Testing
Environmental Interference Testing
Rain and fog simulation system
Dynamic model car and standard dummy
A variety of common traffic signs