Tesla uses lidar to build training set for vision? You deserve it!
Not long ago, a Tesla Model x with lidar was discovered in Los Altos, California.
△That's it.（picture source：https://www.teslarati.com/tesla-model-x-lidar-equipment-photos/）
Considering that Tesla is a staunch "visionary" at present, industry insiders speculate that lidars will not appear in mass-produced vehicles, and they are likely to be ground truth sensors used to build visual training sets.
What is a true value sensor?
The basic ability of automatic driving system is target perception.
△On the basis of object perception, the automatic driving function is further developed.
When testing and verifying the perception ability, sensors with outstanding performance are often selected as "true value sensors". The data collected by it will be used as the "reference answer" to evaluate the target perception.
Lidar has natural advantages in acquiring 3D contour of target, accurate ranging and real-time trajectory tracking. Lidar has been widely used in ranging calibration.
In Tesla patent named "estimating object propertiesusing visual image data", it is also mentioned that the laser radar / millimeter wave radar data associated with visual data is used to accurately evaluate the target information.
△The Tesla patent mentions lidar as a true value sensor
Therefore, the lidar armed on the roof of Tesla is likely to serve for training visual neural network and testing algorithm.
Thanks to years of experience in lidar application and development, bright channel intelligence launched the true value system mainly based on lidar in 2019.
△The table is for reference only. For specific configuration scheme, Liangdao intelligent will provide one-to-one service.
The customized true value system has been used in many Chinese and German main engine manufacturers, suppliers and institutional customers for perception evaluation and automatic driving function development.
At the same time, we also found that processing sensor data to establish true value is still a time-consuming and labor-consuming work. We want to save you the most time and labor costs.
Therefore, Liangdao intelligent has developed an automatic truth value establishment tool, which can label the original data of several lidars, and is equipped with semi-automatic manual correction tools, so that the labeling work can be automated as much as possible.
△Automatic driving test verifies off-line algorithm to help establish true value automatically
At present, the truth value building tool is available：
1）Identify and distinguish dynamic targets such as passenger cars, trucks, trucks / buses, motorcycles, pedestrians and bicycles；
2）Continuously output the position direction, velocity, acceleration, steering angle, time / distance between objects and other parameters of the dynamic target.
These data sets with precise annotation have become the high quality raw materials for the development of autopilot algorithm.
If you want to establish the true value for vision, the last link is the 3D bounding box generated by the fusion of lidar and vision. We have done this part for you.
△激光雷达与视觉融合生成视觉的 3D bounding box3D bounding box generated by fusion of lidar and vision
In the field of automatic driving, lidar has gradually become an indispensable core sensor. It can not be ignored that lidar has begun to play an irreplaceable role in the development and test verification stage.
For the truth system, you may have the following questions:
1）How to choose different truth systems for different perception evaluation objects?
2）How to ensure the integrity and reliability of the annotation results of the true value system?
3）What are the key indicators of the test KPI of the perception system? What value can it provide for our perception algorithm upgrade?
If you want to know more about the solution of Brightway intelligence, please contact us email@example.com We will introduce the details of the truth system to you.