He came to the second lesson, open dry mode!
1. Introduction & 2.Sebastian introduce high-precision map map
slightly
3. High-precision map vs traditional map
General maps is certainly not possible to meet the needs of unmanned, high-precision map has gradually become unmanned just need to get high-precision surveying and mapping of business in recent years is gradually increasing, so this really is a big trend ah!
Compared with the traditional high-precision map map, first of all is the "precision" of the ordinary maps can only achieve meter level, and high-precision map, you can reach centimeter level, this unmanned vehicle safety is to guarantee essential. Second, high accuracy maps also includes " semantic information " (e.g., speed limit, go straight, turn, etc.).
4. The map and positioning, the relationship between perception and planning _a
High-precision map is one of the core unmanned, because other operations of the respective modules are required to support high-precision map.
An important feature is the highly accurate map positioning, positioning and puzzles to take courses analogy, that is, matching puzzle. Unmanned itself need to know where the location map, the vehicle is first need to find landmarks, information of the vehicle obtained through the camera, radar and other sensors known landmarks on the same map accurately compared.
The matching process requires pretreated, coordinate conversion, data fusion these complex processes. Pretreatment to remove inaccurate or poor quality data, the coordinate conversion of the data conversion from different perspectives into a unified coordinate system, and finally data fusion, to form a "landmark", compared with the high-precision map, thereby completing the positioning jobs.
The map and positioning, the relationship between perception and planning _b
The human senses have limitations, sensors of the vehicle is no exception. Affected by external conditions, such as bad weather, at night, etc. In addition, if an obstacle is encountered, the perception is unable to understand the objects behind obstacles.
Because of this, high-precision sensor map can help predict in advance the location of things, can also help reduce the sensor detection range that allows the sensor detects only the region of interest (ROI), thereby improving the detection accuracy and speed sensor, saving computing resources, improve real-time sex.
6. The map and positioning, the relationship between perception and planning _c
High-precision map can help plan the path of the vehicle, such as the vehicles traveling along the lane centerline; in the crosswalk area, speed bumps, etc., can be predicted in advance by high-precision map, slow down in advance; the need to change lanes in front of an obstacle, high map accuracy can help narrow the selection, best channel change program.
7.Apollo high-precision map
Designed for unmanned high-precision map designed to provide a definition of the road, intersections, traffic signals, lane rules, and other elements of the car navigation can provide a wide range of help for the driverless car.
E.g. accurately map the exact position and height of the recording information light, thereby reducing the perceived difficulty.
Map updates is an important project for high-precision map, you need a lot of work vehicles, high-precision map constantly verified and updated at the same time to achieve centimeter-level accuracy, which requires a very high mapping technology.
There are many high-precision map format, will lead to a different format is not compatible with the system, in order to facilitate sharing, Apollo uses OpenDrive format, which is the industry standard drawing format.
Baidu Apollo has also developed a comprehensive mining map drawing system, of which 90% of the mapping process automated.
8.Apollo high-precision map building
Construction of a high-precision map five processes: data acquisition, data processing, object detection, manual validation, and mapping .
Data collection : Baidu Apollo has nearly 300 professional vehicle for gathering the data source, of course, these cars are not just for data acquisition data acquisition, maintenance and updates to the map is also very important, he can timely update the map data. Collection vehicle having an angle sensor, such as GPS, inertial measurement unit (an IMU), camera and the laser radar , Apollo defines a framework for hardware, these sensors integrated into a single autonomous system, by the fusion of these data, the final precision generated map.
Data processing : data sorting, classification and washing to obtain without any semantic information or comments initial map template. For example, FIG initial point cloud data of the Beijing Zhongguancun process.
Object detection : Apollo team uses artificial intelligence (deep learning) to detect and classify static objects, including lane markings, traffic signs and even telephone poles.
Manual Verification : This step is the investigation of artificial correction, automatic map creation process in order to ensure proper and timely detection of problems.
Map released : After processing the above process, you can publish precision maps. In addition to publishing high accuracy maps , Apollo also released a top-down view of the relative positioning of the map and the three-dimensional point cloud map .
In the process of building and updating the map in, Apollo uses crowdsourcing , which means that anyone can participate in the task of making highly accurate maps published by the tool Baidu, crowdsourcing can also smart phones, smart infotainment system and even other driverless car to achieve. This crowdsourcing , feels a bit like " open source " thinking ah ha ha ha ha.
9. Course Overview
High-precision positioning map is critical to performance, but also plays an important role in planning and perceptual systems. At the same time we understand the Apollo high-precision map of the production process, the above study, we should have a more profound understanding of the works unmanned.
The second lesson is probably the case.
