General research into autonomous parking, integrating industry standards, trend assessments and system introductions

Written in front

The automatic parking system means that the vehicle can independently find a parking space and complete accurate parking without manual intervention. It can also park according to the user's needs. The requirement is moved to the specified location. The maturity of this technology will greatly alleviate people's inconvenience during the parking process, such as wasting time looking for a parking space for a long time, as well as safety issues such as vehicle collision and friction that may occur when looking for a parking space.

Currently, many technology companies and university laboratories are exploring the field of autonomous parking. Companies such as Mercedes-Benz and Bosch have established parking infrastructure. NVIDIA is developing automatic driving algorithms and parking space detection methods for parking lots. BMW has also installed automatic parking modules on production vehicles.

Considering the widespread demand for automatic parking technology in the industry, we conducted a systematic survey on the development status of automatic parking systems. Including the industry standard trends of automatic parking systems, performance evaluation indicators, technologies designed into each sub-module of the automatic parking system, etc.

Current standard trends in the automatic parking industry

As the current automatic parking system has received widespread attention from industry and academia, various countries and companies are Continuously develop automatic parking technology. Therefore, it is particularly important to formulate standards for automatic parking technology. The technical standards that have been developed include Geographical Information, Outdoor Positioning, Indoor Space Positioning, Parking Standard and Vehicle Communication Standard Several aspects of content, specific standards are summarized in the following table.

Relevant standards for the automatic parking industry

• Geographic information and positioning standards establish principles and basis for the naming, definition and format of outdoor spaces , Such standards include ISO 14825, ISO 17572 and ISO TC204 177438, etc.
• The standardization of indoor space positioning is also being formulated, including ISO TC211 and OGC standards.
• ISO/DIS 16787 APS parking standard proposes the types of information required for parking and defines the technical name for controlling the vehicle. In addition, the standard defines the functions required to implement an assisted parking system and perform steering control functions of the vehicle.

Evaluation standards for automatic parking systems

The current evaluation of automatic parking systems includes two parts, namely evaluation of the automatic driving part and Review of the automatic parking part.

Evaluation Standards for Autonomous Driving Parts

The evaluation of autonomous driving technology follows the "Autonomous Driving Technology Stage" standards formulated by the Society of Automotive Engineering, which combines autonomous driving technology It is divided into six levels and classified according to technical level, control subject and driving ability. To make it easier to evaluate each level, the ISO/WD 34501 and ISO/WD 34502 standards under development were proposed. The ISO/WD 34501 standard applies to terms and definitions for test scenarios in Level 3 systems, while the ISO/WD 34502 standard provides guidance on test scenarios and a safety assessment process. The formulation of these standards aims to provide unified evaluation criteria for the development and application of autonomous driving technology and promote the development of autonomous driving technology.

Evaluation standards for the automatic parking part

Although automatic parking technology is currently attracting much attention, the formulation of international standards is still in its infancy. Unlike autonomous driving evaluation standards, the technical level of autonomous parking is mainly measured by developers’ evaluation standards.

The following table shows examples of autonomous parking scenarios in the traffic situation scenario system. Parking scenarios are divided according to performance levels.

Automatic parking level division

The traffic situation scenario system evaluation includes autonomous driving and parking scenarios, and uses the "automatic parking" defined by the Society of Automotive Engineering. Driving Skills Stage” to point out the ability to self-park. There are currently three levels in the system.

• The second level (Lv2): called the parking assist system is used to help people park more conveniently. At Lv2 level, vehicles are usually equipped with an obstacle distance warning system and a rearview camera.
• The third level (Lv3): Lv3 level can realize automatic parking in specified scenarios. For example, simple parking scenarios such as parking in a garage.
• The fourth level (Lv4): All scenes in Lv3 can be executed at Lv4 level. In addition, when the vehicle encounters an obstacle, the vehicle can return to its destination after stopping or avoiding the obstacle.

Detailed explanation of the automatic parking system

At present, the automatic parking system mainly includes three parts, as shown in the figure below, respectivelySearch Driving process , Automatic parking process and Return to driving process. We first give a general introduction to each process, and then introduce the current development trends of each process.

Workflow of the automatic parking system

Search driving process

The ultimate goal of the search driving process is to be in the parking lot The car can find a parking space independently, so the process requires positioning, anti-collision and parking space detection technology.

Introduction to positioning technology

In the automatic parking system, you can use GPS or the IMU sensor that comes with the vehicle to confirm the speed and speed of the vehicle. Attitude and correct the vehicle's position estimation error. However, in some indoor scenes, the GPS signal cannot be received. Therefore, self-driving vehicles need to be equipped with cameras, lidar and millimeter-wave radar for assistance. The picture below is an introduction to some sensor information equipped on the vehicle.

Introduction to sensor information equipped with automatic parking vehicles

• Lidar sensors use laser light to detect objects. The currently popular lidars are mainly divided into 16, 32, 64 and 128 lines of lidars. Its measurement range is approximately 200 meters and its vertical field of view is 30 to 50 degrees.
• Camera sensors can be roughly divided into monocular, binocular and fisheye types. Typically, autonomous vehicles are equipped with cameras with a horizontal field of view of 90 to 210 degrees and a vertical field of view of 90 to 180 degrees.
• Millimeter wave radar can be divided into short-range radar and long-range radar based on distance. Among them, the range of short-range radar is about 5 meters, the horizontal field of view is 5 to 20 degrees, and the vertical field of view is 10 to 35 degrees. The range of long-range radar can be about 200 meters, with a horizontal and vertical field of view of 35 to 80 degrees.

After obtaining the information collected by these different sensors, SLAM mapping technology can be used to reconstruct the environment around the autonomous vehicle, thereby positioning the vehicle. The SLAM mapping technology can be divided into the following two categories: Direct mapping method: tracking the intensity of the data that changes when the sensor moves to estimate the attitude of the sensor. However, because this type of method is easily affected by lighting changes and cannot achieve relocation, current mapping methods are rarely based on this type of method.

Feature-based mapping method: This method first obtains the feature points of surrounding objects from sensor information. Feature points received from the same object are projected to two different sensor coordinates, and the position of the target is estimated by calculating the geometric relationship of the projected points.

Introduction to anti-collision technology

Since the main application scenario of the automatic parking system is in the parking lot, and there will be There are a lot of cars parked, so anti-collision technology is very important. In anti-collision technology, ultrasonic sensors

,

short wave radar sensors, lidar sensors and camera sensors are mainly used. Acoustic and radar sensors are mainly used to achieve accurate ranging. Camera sensors mainly use the difference in position of the same object in consecutive images to estimate distance in depth.

Parking space detection technology

Parking space detection is continuously performed during the search and driving process, and usually includes traditional computer vision, Deep learning and hybrid implementations of the two methods.

Traditional computer vision mainly specifies and identifies the shape of parking spaces, such as parking space line detection and feature point detection, as shown in the figure below.

Example of parking space shape

Thanks to the rapid development of deep learning, methods based on CNN networks are currently widely used in parking space detection. The figure below represents a representative convolutional neural network structure for parking space detection. The convolutional layer learns the features of the input image, and the feature data is output through the fully connected layer. Since this is a fully supervised learning process, the output is directly determined by the labels in the training data.

Parking space detection process based on convolutional neural network

Automatic parking process

If the previous step Once an available parking space has been found during the search driving process, the system calls up the automated parking process. Generally, we use path generation methods to implement the autonomous parking process. Currently, the mainstream path generation methods mainly include algorithm-based methods and reinforcement learning-based methods.

The algorithm-based approach involves calculating the location and shape of the path to the parking space and the current location of the vehicle. In order to calculate a suitable parking path, the algorithm involves algorithms such as optimal control problems, grid-based path planning methods, and rapid exploration of random trees.

Reinforcement learning-based methods can generate optimal paths for autonomous parking processes in autonomous parking simulators. In an autonomous parking simulator, the vehicle learns a common parking procedure. This learning method continuously obtains the optimal path with the highest parking accuracy by repeating the path generation process and evaluation, as shown in the figure below.

Reverse parking process of automatic parking

Return to driving process

Return to driving process refers to the When the self-driving vehicle enters the parking space and is waiting, the user can call the vehicle to move it to the place designated by the user. In this process, we need to use path tracking technology.

Currently, path tracking technology includes methods of tracking traveling paths and methods of tracking modified paths. However, both methods use similar vehicle control algorithms. For details, please refer to the paper [1-2]. The general idea is to execute the control command to move the vehicle along the generated path, taking into account the vehicle's current position and steering angle conditions.

Conclusion

Due to the current increasing demand for automatic parking technology, various countries and manufacturers are vigorously developing automatic parking systems. Taking into account the rapid development of automatic parking technology, in this article, we summarize the standardization trends, evaluation standards and various components of automatic parking systems in automatic parking systems and introduce them in detail. We hope it can be helpful to everyone. ~

Original link: https://mp.weixin.qq.com/s/UPwW0E8LTX5V79GK12HF_Q

The above is the detailed content of General research into autonomous parking, integrating industry standards, trend assessments and system introductions. For more information, please follow other related articles on the PHP Chinese website!