An autonomous driving drivable area detection method, system and vehicle

Abstract

The present invention provides a method for detecting a drivable area of an autonomous vehicle. The method uses a lidar to obtain original point cloud information and performs coordinate transformation; longitudinally stratifies the point cloud information to filter out ground point clouds and impurities; longitudinally adds virtual point clouds according to the obtained point cloud distance resolution to obtain second point cloud information; maps the point cloud into a polar coordinate grid, determines whether there is point cloud in the grid, if not, adds virtual points as point cloud data, selects the point closest to the vehicle in the grid as a candidate point; connects the candidate points to form a drivable area. The present invention can realize the detection of the drivable area in the case of weak light, and at the same time, the obtained boundary is all reliable boundaries.

Description

Technical Field:

[0001] The present invention relates to the technical field of automated detection, and particularly to a method, device, and system for detecting a drivable area based on road surface images. Background Art:

[0002] In recent years, with the rapid development of the automotive industry, traffic accidents have become a global problem. Therefore, driverless applications integrating technologies such as automatic control, artificial intelligence, and pattern recognition have emerged. In the field of driverless, detecting the drivable area is one of the key parts of driverless technology.

[0003] In the prior art, a monocular solution based on deep learning is mainly used to detect the drivable area. Since the monocular solution based on deep learning inevitably uses a GPU, real-time detection cannot be performed. Moreover, due to the lack of chip platforms and cost limitations, there is currently no product using this technology on the market. In addition, for the monocular solution based on deep learning, the more complex the model and the better the training data, the better the detection effect, which easily leads to slow detection speed and strong dependence on data. Summary of the Invention:

[0004] In view of the problems existing in the above prior art, the present invention provides a method, system, and vehicle for detecting the drivable area of an autonomous vehicle. The present invention can perform detection in low-light conditions and improve the detection accuracy through the method, system, and vehicle for detecting the drivable area of an autonomous vehicle, including:

[0005] Obtain the original point cloud information through a lidar and perform coordinate transformation to transform the data into the same coordinate system.

[0006] Vertically layer the point cloud information to filter out the ground point cloud and impurities;

[0007] According to the obtained point cloud distance resolution, vertically add virtual point clouds to obtain second point cloud information;

[0008] Map the point cloud to a polar coordinate grid, determine whether there is a point cloud in the grid. If not, add a virtual point as point cloud data, and select the point closest to the vehicle in the grid as a candidate point;

[0009] Connect the candidate points to form a drivable area.

[0010] Preferably, the point cloud information obtained in different coordinate systems is transformed into the same coordinate system,

[0011] where the different coordinate systems include point cloud data obtained by two or more lidar scanning devices.

[0012] Preferably, vertically stratifying the point cloud information includes vertically fan-shaped stratification with the radar as the coordinate origin;

[0013] Among them, the ground point cloud and impurities are filtered according to the set parameters.

[0014] Preferably, mapping the point cloud into a polar coordinate grid, where the grid is equally divided horizontally in polar coordinates.

[0015] Preferably, select the point closest to the vehicle in the grid as the candidate point, where the connection line of the closest candidate points and the connection line from the point cloud to the vehicle itself form a closed drivable area.

[0016] Preferably, map the closed drivable area into a driving map, and the autonomous driving vehicle controls the vehicle to drive according to the drivable area.

[0017] Preferably, an autonomous driving system includes the above-mentioned method for detecting the drivable area of a vehicle.

[0018] Preferably, according to the autonomous driving system described in the present invention, it further includes:

[0019] Sensors, including lidar sensors, for obtaining environmental information and parsing the point cloud information through the SDK provided by the lidar.

[0020] A processor, which can be a computer, for processing the point cloud information obtained by the radar sensor and processing the point cloud information.

[0021] Preferably, the present invention includes an autonomous driving vehicle, including an autonomous driving system that can detect the drivable area.

[0022] The present invention has the following advantages compared with the prior art:

[0023] First, the present invention improves the method for detecting the drivable area based on images, and the detection of the drivable area can be completed even when the light is poor.

[0024] Second, the present invention utilizes lidar point cloud information, considers information such as vertical viewing angle, range resolution, and adding virtual candidate points, obtains and determines reliable candidate points, and obtains all the boundaries of the drivable area, avoiding the inability to determine all the boundaries due to sparse point clouds. Brief Description of the Drawings:

[0025] Figure 1 Illustrates a flowchart of laser point cloud data processing according to an embodiment of the present application.

[0026] Figure 2 Illustrates a schematic diagram of the installation position of a laser scanner according to an embodiment of the present application.

[0027] Figure 3Illustrates the vertical scanning area and layered schematic diagram of the lidar according to an embodiment of the present application

[0028] Figure 4 Illustrates the pulse echo width distribution histogram according to an embodiment of the present application.

[0029] Figure 5 Illustrates the schematic diagram of longitudinal growth candidate points according to an embodiment of the present application.

[0030] Figure 6 Illustrates the test result map of the obtained closed drivable area according to an embodiment of the present application. Detailed implementation manners:

[0031] The following combines Figures 1-6 and embodiments to further describe the present invention in detail: The present invention processes the data obtained from the lidar sensor to obtain the required data:

[0032] The system devices involved in this method include: lidar and computer. The lidar obtains environmental data, processes the data in the Ethernet processor, parses and obtains the original information of the point cloud, and transmits the original point cloud information into the drivable area for data processing, maps the drivable area into the map module, and the autonomous driving system controls the driving of the vehicle according to the drivable area.

[0033] As Figure 1 shown in the lidar point cloud data processing flow chart.

[0034] Step 101 is the start step.

[0035] In step 102, the original point cloud information is obtained through the lidar, and coordinate transformation is performed to convert different data into the same coordinate system.

[0036] In step 103, the point cloud information is longitudinally layered to filter out the ground point cloud and impurities

[0037] In this step, the point cloud information is longitudinally layered as shown in the vertical scanning area and layered schematic diagram of the lidar scanner, and the ground point cloud and impurities are filtered out. The longitudinal layering refers to dividing into 0, 1, 2... n layers from the ground upward at a certain angle as shown in the figure. The purpose is to reduce ground clutter.

[0038] In step 104, virtual point clouds are longitudinally added according to the obtained point cloud distance resolution to obtain the second point cloud information;

[0039] Due to the characteristic that the point cloud is a sparse matrix, the present invention utilizes the range resolution. When the trailing edge (falling edge) of the echo pulse of a closer target coincides exactly with the leading edge (rising edge) of the echo of a farther target, it is regarded as the limit of resolvability. At this time, the distance between the two targets is the range resolution. Virtual point clouds are added in the longitudinal direction. The solid points shown in the figure are the original points, and the dotted points are the point clouds added according to the range resolution.

[0040] Step 105 filters out the spurious points by using pulse width modeling;

[0041] In addition to filtering out the ground point clouds, it is also necessary to filter out the impurity points, and the spurious points are filtered out by pulse width modeling.

[0042] Step 106 maps the point cloud into a polar coordinate grid, determines whether there is a point cloud in the grid. If not, virtual points are added as point cloud data, and the point closest to the vehicle in the grid is selected as the candidate point;

[0043] The polar coordinates are equally divided into grids. By equally dividing the polar coordinates, it is ensured that there is a point closest to the vehicle itself in each sector. If there is no point cloud in the sector, virtual points will be created to obtain a closed drivable area.

[0044] Step 107 determines whether there is a candidate point in the equally divided grid.

[0045] Step 108 creates virtual points if there is no candidate point. Virtual points are added according to the range resolution of the lidar and the original point cloud.

[0046] Step 109 connects the candidate points to form a drivable area;

[0047] The point closest to the vehicle in the grid is used as the candidate point, and the candidate points are connected to form a drivable area.

[0048] Step 110 ends the program operation. The drivable area will be cyclically detected during vehicle driving.

[0049] As Figure 2 shown, it is a schematic diagram of vehicle radar installation. Generally, it is installed on both the left and right sides in front of the vehicle.

[0050] Environmental data information is obtained through the lidar, and the original information of the point cloud is parsed and obtained. Coordinate transformation needs to be performed on the point cloud information. The transformation process of the point cloud information is as follows:

[0051] 1)

[0052] 2)

[0053] x raw1 ,y raw1 ,z raw1 ,x raw2, y raw2 , z raw2 are the original point cloud position information, and θ1, θ2 are yaw, x offset , y offset , z offset is the installation position of the sensor.

[0054] Such as Figure 3 the vertical scanning area and the schematic diagram of layered points of the laser scanner. As shown, the point cloud information is longitudinally layered, and the ground point cloud and impurities are filtered out. The longitudinal layering refers to dividing into 0, 1, 2... n layers from the ground upward at a certain angle as shown in the figure. The purpose is to reduce ground clutter points.

[0055] For example, a certain degree is set to 0.8°, that is, the difference between each layer is 0.8°, and the position of the lidar is used as the origin of the coordinate axis.

[0056] For the point cloud data of the 0th and 1st layers, the following conditions are met for obstacle target points:

[0057] The 0th layer: If r ≤ |z mount / sin(1.6*π / 180)| - 3, it is non-road surface information;

[0058] The 1st layer: If r ≤ |z mount / sin(0.8*π / 180)| - 3, it is non-road surface information;

[0059] Among them, r represents the distance from the point cloud to the coordinate origin, and z mount represents the installation height of the laser scan.

[0060] After filtering out the ground point cloud from the point cloud, the echo pulse number and pulse width in the original data are used to filter out the clutter points. The reflectivity is affected by the material, color, etc. of the object. The laser radar echo pulse width characteristic is determined by the reflectivity. The present invention filters out the clutter points by using the echo pulse difference. Through the analysis of the original data, it can be known that the pulse width range is 0 - 5 cm. From the echo histogram, it can be seen that the echo presents a Gaussian distribution. At the same time, it can be analyzed that for larger or smaller pulse widths, they are interference points. In order to eliminate these interference points, the dynamic threshold is obtained by using the error analysis 3σ principle. The points within the range of (μ - 3σ, μ + 3σ) are reliable target points.

[0061] (1)

[0062] (2)

[0063] Among them, (T1, T2, T3......) are the sample values within all pulse width ranges after the ground points are proposed.

[0064] Figure 5As shown in the figure, it is a schematic diagram of the longitudinal growth candidate points in the embodiment of the present application. According to the obtained point cloud distance resolution, virtual point clouds are added longitudinally to obtain the second point cloud information.

[0065] At the same time, according to the distance resolution of the laser scanner, longitudinal growth candidate points are added; due to the characteristic that the point cloud is a sparse matrix, the present invention utilizes the distance resolution. When the trailing edge (descending edge) of the echo pulse of a nearer target coincides exactly with the leading edge (ascending edge) of the echo of a farther target, it is regarded as the limit of distinguishability. At this time, the distance between the two targets is the distance resolution. Virtual point clouds are added in the longitudinal direction. The solid points shown in the figure are the original points, and the dotted points are the point clouds added according to the distance resolution.

[0066] The specific addition method is as follows:

[0067] Due to the characteristic of sparse point clouds, the present invention adds virtual points according to the distance resolution. Therefore, the coordinates of the virtual points are:

[0068]

[0069] x row ,y row ,z row is the point cloud of the original measurement, and 0.04 is the distance resolution of the lidar.

[0070] As Figure 6 shown, the polar coordinates are equally divided into grids. The polar coordinates are equally divided to ensure that there is a point closest to the vehicle itself in each sector. If there is no point cloud in the sector, virtual points will be created to obtain a closed drivable area.

[0071] The point cloud is mapped into the polar coordinate grid, and it is judged whether there is a point cloud in the grid. If not, virtual points are added as point cloud data, and the point closest to the vehicle itself in the grid is selected as the candidate point;

[0072] The polar coordinates are equally divided to ensure that there is a point closest to the vehicle itself in each sector. If there is no point cloud in the sector, virtual points will be created to obtain a closed drivable area. The manifestation is as follows: Connect the candidate points to form a drivable area.

[0073] An autonomous vehicle, the autonomous driving system of the above method for detecting a drivable area, can control the operation of the vehicle according to the drivable area.

[0074] The autonomous driving system of the present invention includes a sensor, including a lidar sensor, which is used to obtain environmental information and parse the obtained point cloud information through the SDK provided by the lidar. A processor, including a computer, which is used to process the point cloud information obtained by the radar sensor and process the point cloud information. A memory; and computer program instructions stored in the memory, which, when run by the processor, cause the processor to execute the method for detecting drivable areas in the present invention.

[0075] The device of the present invention can be applied to autonomous driving vehicles, which are provided with the above-mentioned autonomous driving system, and control the vehicle operation through drivable area detection.

[0076] The control system includes a computer-readable storage medium, on which computer program instructions are stored. The computer program instructions, when run by a processor, cause the processor to execute the method for detecting drivable areas in the present invention.

[0077] The method of the present invention improves the method for detecting drivable areas based on images, and the detection of drivable areas can also be completed in case of poor light.

[0078] At the same time, using lidar point cloud information, considering information such as vertical viewing angle, distance resolution, and adding virtual candidate points, reliable candidate points are obtained and determined, and all boundaries of the drivable area are obtained, avoiding the inability to determine all boundaries due to sparse point clouds.

Claims

1. A method for detecting a drivable area, characterized in that Including: Obtain the original point cloud information and perform coordinate transformation; Vertically layer the point cloud information to filter out the ground point cloud and impurities, including: taking the lidar as the coordinate origin, vertically fan-layer the point cloud information, and divide it into 0, 1, 2... n layers from the ground upwards at a preset angle; filter out the ground point cloud and impurities according to the set parameters; According to the obtained point cloud distance resolution, vertically add virtual point clouds to obtain the second point cloud information; When the falling edge of the echo pulse of a closer target just coincides with the rising edge of the echo of a farther target, it is regarded as the limit of distinguishability. At this time, the distance between the two targets is the distance resolution; according to the distance resolution, add virtual points, and the coordinate calculation method of the virtual points is: x virtual = x row+ Range resolution * x row y virtual = y row Z virtual= z row x row , y row , z row are the point cloud coordinates of the original measurement, x virtual， y virtual， Z virtual are the coordinates of the added virtual point cloud; Horizontally and equally divide the polar coordinates into grids, map the point cloud into the polar coordinate grids, judge whether there is a point cloud in the grids. If not, add virtual points as point cloud data, and select the point closest to the vehicle in the grids as the candidate point; Connect the candidate points to form a drivable area.

2. The detection method according to claim 1, wherein, Including: Transform the point cloud information obtained under different coordinate systems into the same coordinate system, where the different coordinate systems include point cloud data obtained by two or more lidar scanning devices.

3. The detection method according to claim 1, wherein Including: Select the point closest to the vehicle in the grids as the candidate point, where the connection of the closest candidate points and the connection from the point cloud to the vehicle itself form a closed drivable area.

4. The detection method according to claim 3, wherein Including: Map the closed drivable area into the driving map, and the autonomous driving vehicle controls the vehicle to drive according to the drivable area.

5. An autonomous driving system, characterized in that, Including the drivable area detection method according to any one of claims 1-4.

6. The autonomous driving system according to claim 5, characterized in that, Including: Sensors, including lidar sensors, for obtaining environmental information and parsing the point cloud information through the SDK provided by the lidar; Processors, including computers, for processing the point cloud information obtained by the radar sensors and processing the point cloud information; Memories; And computer program instructions stored in the memories, the computer program instructions, when run by the processors, cause the processors to execute the method according to any one of claims 1-4.

7. An autonomous vehicle, characterized in that, Including the autonomous driving system according to claim 5.

8. A computer-readable storage medium, characterized in that, On which computer program instructions are stored, the computer program instructions, when run by the processor, cause the processor to execute the method according to any one of claims 1-4.

Images