With the maturity of computer vision technology, ecology and constantly travel intelligence, which is the typical application scenarios automobile autopilot.
1. automobile autopilot really come
2018, the world's first unmanned taxi was officially launched. This is a Silicon Valley start-up company Drive.ai in Frisco, Texas, achieved the first unmanned taxi travel.
China, Baidu is the automotive industry leader in autopilot. November 30 this year, Baidu in Guangzhou opened a trial operation RoboTaxi of service, this is Baidu autopilot taxi in Changsha trial operation, and the city began a trial operation of autonomous vehicles.
If you count the drops plans to launch at the end of automated driving a taxi in Shanghai, 2019 is likely to see the autopilot taxi figure in the three cities.
We can say that autonomous vehicles really want to come.
2. automobile autopilot support behind
In the process of automobile autopilot, the car itself needs to have the perception, planning, decision making, control and many other "skills", these skills can be referred to as "artificial intelligence."
However, the result of a so-called smart just want to let the car itself did algorithm to handle more and more complex scenes, behind on the need for vast amounts of real road data do support.
And this need to rely on the data label.
The meaning of existence data labeling machines is understood and appreciated in the world. In the field of automotive autopilot, labeling data processing typically includes a scene labeling overtaking lane changing, through the intersection, no unprotected left, right, and the long tail of complex scenes such as a traffic light control red light vehicle, pedestrian crossing the road, curb illegal parked vehicles and so on.
3. The common areas of several automobile autopilot Data Highlighter
Data processing typically labeling data content types of voice, text, images, etc., the classification types are marked, frame, comments, tags, etc., particularly to the field of automotive autopilot, frequently used tools 2D annotation box, 3D cube, lane lines, polygons, and so on semantic segmentation.
We marked the background Man Rover Technology data from a few selected examples diagram that can help us have a better understanding of the marked scene:
Man Fu Technology box marked 2D renderings
Man Fu Technology 3D cube marked renderings
Man Fu Technology lane line marked effect chart
Man Fu Technology polygon marked renderings
Man Fu Technology semantic segmentation mark renderings
It was with these data do support a good mark, autonomous vehicles can navigate in the fast lane of science and technology.
4. The high-quality data is the future of the industry
With autonomous vehicles start from the laboratory to reality, autonomous vehicles of security is more and more widespread concern in society.
As the basis for autopilot technology, it will directly affect the quality of the final mark data model is good or bad effects. Massive, high-quality, refinement of the data can improve the safety and practicality of the car autopilot to a large extent, boosting the autopilot landing process.
In fact, the field of automotive autopilot require labeling for new data quality data is the future direction of development of the industry marked an important microcosm. And Artificial Intelligence industry "sophisticated" different "high-tech" label, data label is still a labor-intensive industries, a long period of extensive state.
With artificial intelligence project began with automobile autopilot represented have landed, more and more companies realize AI, high-quality data set is the key to artificial intelligence project landing impact.
Future, fine, scene, customized label data will be an important direction of development of the industry, door industry changes already open, high-quality tagging data set will prop up the industry a new future of artificial intelligence.