Baidu Maps opens Shunde data center for the first time: automation level reaches over 80%

On the afternoon of July 28, Baidu Maps opened its data center in Shunde City, Guangdong Province to the media for the first time, showing in detail the entire process of Baidu Maps from field data collection to internal data processing, and fully displaying the process of a POI point and a road appearing on Baidu Maps; at the same time, Baidu Maps also explained for the first time the application of artificial intelligence, deep learning and other technologies in the field of map data production.

Gu Weihao, deputy general manager of Baidu Maps and general manager of Baidu Smart Car, said that Baidu Maps is using artificial intelligence technologies such as deep learning and image recognition to increase the proportion of automation in map data production. Currently, this figure has reached more than 80%.

Gu Weihao, Vice General Manager of Baidu Maps and General Manager of Baidu Smart Car, delivered a speech

| Multiple acquisition devices exposed

According to reports, the data production process of Baidu Maps includes four major processes: "data collection - data processing - quality inspection - release". This time, Baidu exposed several externally sourced devices for the first time in the data center in Shunde City. In addition to high-precision maps, all data in Baidu Maps currently used by users can be collected by these devices.

The first thing that caught people's attention was the dozens of Changan cars parked in the park. Baidu released its smart car strategy at the end of March, and Changan became its first automotive partner. That morning, Changan delivered 80 Changan cars to Baidu Maps for field data collection. After these 80 cars were put into use, the size of Baidu Maps' field data collection fleet has exceeded 250 vehicles.

These external acquisition teams are mainly responsible for the panoramic acquisition of common road sections. The core acquisition equipment used by most acquisition vehicles mainly includes GNSS+IMU and imaging systems, while the two acquisition vehicles closest to the entrance are also equipped with Velodyne 32-line laser radars, which means that the acquisition vehicles can be used for high-definition map acquisition. It is reported that all acquisition vehicles can complete ADAS-level map data acquisition.

In addition to the panoramic collection vehicle responsible for general road sections, Baidu also displayed collection equipment such as electric bicycles for narrow road sections, point cloud backpacks for indoor collection, and drone collection devices for special collection.

| Only 20% of the work needs to be done in-house

After the external procurement team completes data collection, the internal team needs to process the data accordingly before it can be put online. Using Baidu's artificial intelligence technology, more than 80% of the data collected by the external procurement team can be processed automatically, and manual processing accounts for less than 20% of the data processing in Baidu Maps, which greatly improves work efficiency and ensures the timeliness of data updates.

Entering the Shunde Data Center building, the relevant person in charge of Baidu Maps introduced the work flow of the internal team to the media. Since Baidu Maps uses image recognition, automatic differential fusion, automatic modeling and other technologies to automatically extract POI information from real-time panoramic images, produce panoramic intersection enlarged images, electronic eyes, 3D navigation and other data, it can be seen that most employees' work is mainly done by calibration. In the entire data processing link, manual workload accounts for less than 20%.

Baidu Maps Panoramic Image Automatic Recognition Technology

According to reports, Baidu Maps collects data once a quarter. At present, Baidu Maps has covered a total of 6.7 million kilometers of domestic roads, with a POI data scale of more than 45 million, and an average of 30 billion positioning requests responded to daily. Gu Weihao, deputy general manager of Baidu Maps and general manager of Baidu Smart Car, said that in the future, they will continue to explore the application of artificial intelligence in the field of map data, further improve efficiency, and make the data on Baidu Maps more vivid and practical.

| The battle for domestic maps has begun

Although Baidu Maps talked little about autonomous driving during the entire visit and stated that it was only a technology provider for its own driverless car plan, it is obvious that the ultimate goal of Baidu Maps is still autonomous driving, that is, high-precision maps.

Although autonomous driving is still in the semi-automatic stage, both semi-automatic and fully autonomous driving require the assistance of high-precision maps. At the same time, high-precision maps can be deeply integrated with vehicle functions to expand applications and services related to car driving. Therefore, the importance of maps to automakers is self-evident. The increasingly frequent "alliances" between automakers and map vendors in recent years have further illustrated this point. The first to start the map war abroad was Here Maps, which was fought over by major automakers last year. Not long ago, Ford Motor Company's three-dimensional map startup Civil Maps also attracted attention. In China, there are also undercurrents. Here is a brief summary of the cooperation between maps and automakers in the past two years:

In 2014, Tesla entered into a strategic partnership with NavInfo in China;

In 2015, Geely and Yitutong cooperated on some models; GAC announced at the end of 2015 that it would cooperate with Kaili on some models;

In March this year, Baidu and BAIC reached a high-precision map cooperation; also in March, AutoNavi Maps and Jaguar Land Rover launched navigation products; in April, AutoNavi released a car version of the map and cooperated with Geely Changan on multiple models; earlier this month, the YUNOS system installed in the Internet car Roewe RX5 jointly developed by SAIC and Alibaba was equipped with AutoNavi's high-precision map;

Behind the increasingly close interaction between car manufacturers and map vendors is the manifestation of the strategic position of map vendors in future cars. The data capabilities, professional technical capabilities and depth of cooperation with car companies of the map vendors they work with are the key to determining the success or failure of the competition landscape of car companies. After the "vertical and horizontal alliances", the current map vendors have become a game between BAT, and of course there is also Xiaomi, and the competition between these companies has been fierce, so we will see more competition for car company resources in the future. If map vendors want to win the favor of car companies, in addition to the amount of data, they also need to look at the data processing efficiency of artificial intelligence.