博客

无人驾驶车需要更详细、精确的地图


Apollo 高精度地图源码

Navigation Map vs. HD Map

A high definition map contains a huge amount of driving assistance information

• the accurate three-dimensional representation of the road network.
  • the layouts of intersections and the locations of signposts.
• a lot of
  
  semantic
  
  information.
  • The map may report what different colors of
    
    traffic lights
    
    mean.
  • It might also indicate the
    
    speed limit
    
    for a road.
  • and where the
    
    left turn lane begins.
    
• precision
  • Navigation Map:
    
    meter-level
    
    precision.
  • HD Map:
    
    centimeter-level
    
    precision.

Localization, Perception, Planing

video 1

无人驾驶车辆需要知道它在地图上的确切位置

首先，车辆可能会寻找

地标

。我们可以使用从各类

传感器

收集的数据，如摄像机图像数据，以及激光雷达收集的三维点云数据来

查找地标

。车辆将其收集的数据与其高精度地图上的已知地标进行

比较

。这一匹配过程是需要

预处理、坐标转换和数据融合

的复杂过程。

• Preprocessing eliminates inaccurate or poor quality data.
• Coordinate transformation converts the data from different perspectives(不同视角) into a uniform coordinate system(统一的坐标系).
• Data fusion merges data from different vehicle and different types of sensors.
  
  一旦无人驾驶车高度精确地确定了其位置，定位任务也就完成了。

video 2


perception

software relies on the high-definition map.

• HD Map can feed
  
  traffic light positions
  
  to the rest of the software stack even if the sensors can’t detect the traffic light yet.
• HD Map will help the sensor
  
  narrow it’s detection scope
  
  . region of interest(ROI) can help us improve both detection accuracy and speed, saving computing resources for other tasks in the vehicle.

video 3


planning

software relies on the high-definition map.

Apollo HD Map

Apollo HD Map Construction

HD Maps Service

https://data.apollo.auto/hd_map_intro?locale=en-us