On the afternoon of July 28, Baidu Maps opened its data center in Shunde, Guangdong Province, for the first time. It displayed the entire process of Baidu maps collection from the field to the internal data processing, showing a complete POI and a road. The process on Baidu maps; at the same time, Baidu maps also expounded for the first time the application of artificial intelligence, deep learning and other technologies in the field of map data production.
Gu Weiyu, vice general manager of Baidu map and general manager of Baidu Wisdom Automotive, stated that Baidu Maps is using artificial intelligence technology such as deep learning and image recognition to increase the proportion of automation in the production of map data. At present, this figure has reached more than 80%.
Bai Wei map deputy general manager and general manager of Baidu wisdom car Gu Weikai speech
A variety of acquisition equipment exposureAccording to reports, Baidu map data production process includes "data collection - data processing - quality inspection - release" four processes. In the data center of Shunde City, Baidu first exposed more than one foreign mining equipment. Except for high-precision maps, all the data in Baidu maps currently used by users can be collected by these devices.
The most eye-catching is the Changan car parked in dozens of parks. Baidu released its smart car strategy at the end of March and Changan Automobile became its first automotive partner. In the morning, Changan Automobile just delivered 80 Changan vehicles to Baidu Maps for use in the field. After the 80 vehicles were put into use, the fleet of Baidu map mining teams has exceeded 250 vehicles.
These mining fleets are mainly responsible for the panoramic collection of common roads. Most of the acquisition vehicles use core acquisition equipment including GNSS+IMU and imaging systems, while the two collection vehicles closest to the entrance are additionally equipped with Velodyne 32-line laser radar. This means that the vehicle will be available for HD map acquisition. According to reports, all acquisition vehicles can complete ADAS level map data collection.
In addition to the panoramic vehicle that is responsible for the general road segment, Baidu also displayed electric bicycles for narrow road sections, point cloud backpacks for indoor collection, and drone collection devices for special collections.
The industry only needs to complete 20% of the workAfter the acquisition team completes the data collection, the internal team needs to process the data to complete the online process. Using Baidu's artificial intelligence technology, over 80% of the data collected by the mining team can be processed automatically. The manual processing only accounts for less than 20% of the data processing in Baidu maps, greatly improving the work. Efficiency ensures the timeliness of data updates.
Into the Shunde Data Center Building, Baidu map related responsible person introduced the work flow of the internal team for the media, due to Baidu map using image recognition, automatic differential fusion, automatic modeling and other technologies to achieve real-time panoramic image extraction POI information, making panoramic intersections With data such as enlarged views, electronic eyes, and 3D navigation, it can be seen that most of the employees' work mainly involves calibration. In the entire data processing session, manual labor accounts for less than 20%.
Baidu map industry panorama image automatic recognition technology
According to reports, Baidu Maps conducts data collection every quarter. At present, Baidu Maps has covered a total of 6.7 million kilometers of domestic roads. The scale of POI data has exceeded 45 million, and the average daily response has been requested 30 billion times. Bai Wei map deputy general manager and general manager of Baidu wisdom car Wei Wei said that in the future will continue to explore the application of artificial intelligence in the field of map data, and further improve efficiency, so that the data on Baidu map is more fresh and practical.
The battle of domestic maps has startedAlthough during the entire visit, Baidu Maps was less concerned with autopilot, and its own unmanned vehicle program also indicated that it was only its technical provider. However, it is clear that the ultimate goal of Baidu maps is still to drive autonomously, that is, high-precision maps. .
Although automatic driving is currently in a semi-automated stage, whether it is semi-automatic driving or full-automatic driving, it is necessary to rely on high-precision map assistance. At the same time, high-precision maps can be combined with vehicle functions in depth to expand the applications and services related to car driving. Therefore, the importance of maps for automakers is self-evident. The more and more frequent "coupons" between auto makers and map makers in recent years have also illustrated this point. The earliest map wars in foreign countries were the Here maps that were scrambled by big car companies last year. Ford Motor Company’s Civil Maps, a three-dimensional map startup, also attracted attention. In China, the same is the dark current surging, in this simple card for the next two years of cooperation between maps and car manufacturers:
Tesla’s strategic cooperation with NavInfo in China in 2014;
In 2015, Geely cooperated with Autopass in some models; at the end of 2015, GAC announced that some models will cooperate with Kay Rucker;
In March of this year, Baidu Beiqi reached a high-precision map cooperation. Also in March, the Gaode map and Jaguar Land Rover launched navigation products; in April, Gaode released a car version map in cooperation with a variety of Geely Changan models; at the beginning of this month, SAIC Motor Corporation The YUNOS system in the Roewe RX5, an Internet car jointly developed with Alibaba, is equipped with a high-definition map;
Behind the increasingly close interaction between depots and mappers is the emergence of mappers’ strategic position in the car of the future. The data capabilities, professional technical capabilities, and the depth of cooperation with car companies are the key factors that determine the competitive situation of car companies. After the current map dealers have gone through the game of “playing togetherâ€, they have become a game between BAT and of course there is also a millet, and the competition among these companies has been intensifying, so the future will see more in the future. Competition in car resources. Mappers want to win the favor of car companies, in addition to the amount of data also depends on the data processing efficiency of artificial intelligence.