Guardrail detection method and device, electronic equipment and storage medium

Abstract

Embodiments of the present disclosure disclose a guardrail detection method and device, electronic equipment and storage medium, which obtains a current frame point cloud, determines at least one obstacle target in the current frame point cloud, and for each obstacle target, determines final node statistical information through the corresponding point cloud block of the obstacle target, and then judges whether the obstacle target meets a length continuity condition. For the obstacle target that does not meet the length continuity condition, it is judged whether the obstacle target meets a height continuity condition according to the final node statistical information. The obstacle target that meets the length continuity condition and the obstacle target that meets the height continuity condition are determined as candidate guardrail targets of the current frame. The official guardrail targets of the current frame are determined in combination with the official guardrail targets of the historical frames, so as to determine the guardrail detection result of the current frame, realize the guardrail detection based on the laser radar point cloud, and solve the problem of missing detection of the guardrail caused by missing of the guardrail point cloud due to shielding or other factors.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of automatic driving environment perception, and in particular, to a guardrail detection method and device, an electronic device, and a storage medium. BACKGROUND

[0002] Current guardrail detection methods mainly include traditional rule-based methods and deep learning methods. Among them, the traditional rule-based method only uses a single height and width threshold for judgment. If a segment of guardrail point cloud is missing due to occlusion or other factors, this segment of guardrail will be missed.

[0003] In addition, the deep learning method trains a corresponding model through data-driven to detect guardrail targets specifically, but the calculation resource occupation is relatively high. For the limited calculation resources of the automatic driving vehicle platform, it is not suitable to add another model to detect guardrail targets specifically. Even if the guardrail targets are added to the conventional obstacle detection model, the number of guardrail targets is relatively small compared to other targets, which will cause the long tail effect, resulting in an unsatisfactory effect of guardrail detection. SUMMARY

[0004] To solve the above technical problems or at least partially solve the above technical problems, the embodiments of the present disclosure provide a guardrail detection method, device, electronic device and storage medium, which solve the problem of missing guardrail caused by occlusion or other factors of guardrail point cloud, and quickly, effectively and stably detect guardrail targets without increasing excessive calculation load.

[0005] In a first aspect, the embodiments of the present disclosure provide a guardrail detection method, which comprises:

[0006] obtaining a current frame point cloud, and determining at least one obstacle target in the current frame point cloud, for each obstacle target, determining the final node statistical information of the obstacle target based on the point cloud block corresponding to the obstacle target;

[0007] determining whether each obstacle target meets a length continuity condition based on the final node statistical information of each obstacle target, and determining whether the obstacle target meets a height continuity condition based on the final node statistical information of the obstacle target for the obstacle target that does not meet the length continuity condition;

[0008] determining the obstacle target meeting the length continuity condition and the obstacle target meeting the height continuity condition as candidate guardrail targets of the current frame;

[0009] determining each official guardrail target of the current frame based on each candidate guardrail target of the current frame and each official guardrail target of the historical frame, and determining a guardrail detection result of the current frame according to each official guardrail target of the current frame.

[0010] In a second aspect, the embodiments of the present disclosure further provide a guardrail detection device, which comprises:

[0011] a node statistics module, configured to acquire a current frame point cloud, and determine at least one obstacle target in the current frame point cloud, and for each of the obstacle targets, determine final node statistics information of the obstacle target based on a point cloud block corresponding to the obstacle target;

[0012] a continuous judgment module, configured to judge whether each of the obstacle targets meets a length continuity condition based on the final node statistics information of each of the obstacle targets, and for the obstacle target that does not meet the length continuity condition, judge whether the obstacle target meets a height continuity condition based on the final node statistics information of the obstacle target;

[0013] a candidate determination module, configured to determine the obstacle target that meets the length continuity condition and the obstacle target that meets the height continuity condition as a candidate guardrail target of the current frame;

[0014] an official determination target, configured to determine each official guardrail target of the current frame based on each of the candidate guardrail targets of the current frame and each official guardrail target of a historical frame, and determine a guardrail detection result of the current frame according to each official guardrail target of the current frame.

[0015] In a third aspect, the embodiments of the present disclosure further provide an electronic device, which comprises: one or more processors; a storage device configured to store one or more programs; and when the one or more programs are executed by the one or more processors, the one or more processors implement the guardrail detection method as described above.

[0016] In a fourth aspect, the embodiments of the present disclosure further provide a computer readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the guardrail detection method as described above is implemented.

[0017] The guardrail detection method provided in the embodiments of the present disclosure comprises: acquiring a current frame of point clouds, determining at least one obstacle target in the current frame of point clouds, and for each obstacle target, determining final node statistical information of the obstacle target through a corresponding point cloud block of the obstacle target, and then determining whether the obstacle target meets a length continuity condition according to the final node statistical information. For the obstacle target that does not meet the length continuity condition, it is determined whether the obstacle target meets a height continuity condition according to the final node statistical information. The obstacle target that meets the length continuity condition and the obstacle target that meets the height continuity condition are determined as candidate guardrail targets of the current frame, and each official guardrail target of the current frame is determined in combination with each official guardrail target of a historical frame, so as to determine a guardrail detection result of the current frame. The guardrail detection based on the laser radar point cloud is realized. In the detection process of the guardrail target, the length continuity and the height continuity of the guardrail are considered, and the result of the detection of the historical frame is combined for guardrail identification, so as to avoid the missed detection of the guardrail as much as possible, improve the robustness of the guardrail detection, solve the problem of the missed detection of the guardrail caused by the missing of the guardrail point cloud due to the shielding or other factors, realize the fast, effective and stable detection of the guardrail target, and do not increase too much calculation load and are not affected by the use scene and the sensor configuration. The real-time operation can be realized in the case that the computing resources of the vehicle-mounted platform are limited. BRIEF DESCRIPTION OF DRAWINGS

[0018] The above and other features, advantages, and aspects of the present disclosure will become more apparent when considered in connection with the following detailed description, taken in conjunction with the accompanying drawings. Throughout the drawings, similar or same reference numerals are used for similar elements or features. It should be understood that the drawings are diagrammatic and schematic representations of elements and features, not necessarily to scale.

[0019] Figure 1 A flowchart of a guardrail detection method in the embodiments of the present disclosure;

[0020] Figure 2 A schematic diagram of a transverse point cloud block in the embodiments of the present disclosure;

[0021] Figure 3 A schematic diagram of a longitudinal point cloud block in the embodiments of the present disclosure;

[0022] Figure 4 A schematic diagram of a traversal process in the embodiments of the present disclosure;

[0023] Figure 5 A schematic diagram of an obstacle target in the embodiments of the present disclosure

[0024] Figure 6 A structural schematic diagram of a guardrail detection device in the embodiments of the present disclosure;

[0025] Figure 7 A structural schematic diagram of an electronic device in the embodiments of the present disclosure. Detailed Implementation

[0026] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0027] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0028] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0029] Before providing a detailed description of the guardrail detection method provided in this disclosure, the technical problem solved by this method will be explained. In the prior art, guardrail detection methods mainly include traditional rule-based methods and deep learning methods. For example, the rule-based method and deep learning method can be found in the following solutions:

[0030] Patent 1 (Publication No. CN111881752A) describes a method based on image and point cloud data. It projects the point cloud onto the image using an extrinsic parameter matrix, establishing a correspondence between image pixels and the point cloud. Then, it uses deep learning to detect guardrail targets in the image, distinguishing whether the guardrail is located on the left or right side of the vehicle based on the detection results. Simultaneously, a point cloud semantic segmentation network is used to segment the point cloud data, obtaining the segmentation results for the guardrail targets. Finally, the detection and segmentation results are combined to derive the final detection result and attributes of the guardrail targets.

[0031] Patent 2 (Publication No. CN115327539A) describes a method that leverages the sensitivity of millimeter-wave radar to guardrail targets, utilizing millimeter-wave radar point clouds for guardrail detection. During detection, the millimeter-wave radar data is first processed into a grid, then the gridded point cloud data undergoes a depth-first search to find the longest guardrail target. Finally, the detected guardrail target is fitted to obtain the guardrail's curve equation.

[0032] Patent 3 (publication number CN116129359A), this method proposes a method based on image data, using deep learning method offline detection guardrail. By using canny algorithm to detect the edge in the image, and the edge detection image with the original image in the channel dimension splicing, then using the improved Resnet50 guardrail target detection, the target result is used for the maintenance analysis of guardrail.

[0033] Patent 4 (US20230069618A1), this method proposes a method based on point cloud guardrail detection, this method first uses plane fitting method to remove ground point cloud, then based on the characteristics of mechanical laser radar rotating scanning to construct 3D spherical voxel space, the point cloud is divided into each voxel, and the voxel containing point cloud is distinguished. Further based on the characteristics of the guardrail parallel to the car, the voxel that may be guardrail is selected out, the points in this part of the voxel are extracted and fitted to get the candidate guardrail target, finally the preset guardrail target width, height is used to check the guardrail candidate target.

[0034] The above patent 1, patent 3 uses the method based on deep learning, this kind of method through data driven training corresponding model special for guardrail target detection, the advantage is high accuracy, model can be updated, but the shortcoming is that the calculation resource occupies more, because for the automatic driving vehicle platform limited computing resources, it is not suitable to increase a model to detect guardrail target, even if the guardrail target is added to the conventional obstacle detection model, the number of guardrail target is less than other targets, which will cause the long tail effect, resulting in the effect of guardrail detection is not ideal.

[0035] The above patent 2 and patent 4 adopt the traditional rule-based method. Patent 2 is based on the characteristics of millimeter wave radar sensitive to metal targets to quickly locate the guardrail, and then determines whether the target is a guardrail target according to the length of the fitted point cloud. This method has a faster processing speed, but the disadvantage is that millimeter wave radar is sensitive to metal targets. Various metal targets are easy to form continuous point clouds, and general millimeter wave radar lacks target height information, making it difficult to verify the target from the height aspect. In addition, it is difficult to select the length threshold for guardrail judgment, so this method is more likely to produce false positives. Patent 4 uses laser radar point cloud data to detect guardrails. Based on the characteristics of high coordinate position accuracy of laser radar point cloud data, the guardrail target can be effectively detected. However, the 3D spherical voxel division method used in this method is time-consuming and not suitable for automatic driving platforms with high real-time requirements. In addition, this method relies on the judgment of the flatness and straightness of the guardrail. For non-planar guardrails or guardrails at turning points, the detection effect will be greatly affected. In addition, this method only uses a single height and width threshold to judge the guardrail target at the end, without considering the continuity of the guardrail in the height and width directions. If a section of the guardrail point cloud is missing due to obstruction or other factors, the guardrail will be missed.

[0036] Therefore, in order to solve the above problems, the embodiment of the present disclosure provides a guardrail detection method. In order to solve the problem of high resource occupation and long time consumption of deep learning method, the embodiment of the present disclosure adopts the traditional rule-based idea to design the algorithm, without increasing too much computing load, so that the real-time operation can be realized under the limited computing resources of the vehicle-mounted platform. In addition, in order to solve the problems of long time consumption of 3D space voxel division, reliance on flatness and straightness judgment, and single height and width threshold judgment in the above patent 4, the embodiment of the present disclosure quickly and effectively and stably detects the guardrail target through the processes of identifying the obstacle target, determining the final node statistical information, judging whether the length continuity condition is met, and judging whether the height continuity condition is met, thereby further improving the robustness of the guardrail detection.

[0037] Figure 1 A flowchart of a guardrail detection method in the embodiment of the present disclosure. The method can be applied to point cloud data collected by a vehicle-mounted laser radar to determine the corresponding guardrail detection result, so as to plan the path of the vehicle or control the obstacle avoidance of the vehicle through the guardrail detection result. The method can be executed by a guardrail detection device, which can be realized in the form of software and / or hardware, and the device can be configured in an electronic device. As shown in the figure, the method can specifically include the following steps: Figure 1

[0038] ​S110, acquire a current frame point cloud, and determine at least one obstacle target in the current frame point cloud, for each obstacle target, determine final node statistical information of the obstacle target based on a point cloud block corresponding to the obstacle target.

[0039] The current frame point cloud can be a point cloud in a vehicle body coordinate system collected by a laser radar carried by a vehicle at a current time. Specifically, at least one obstacle target in the current frame point cloud can be determined by obstacle point separation, grid map division, grid state determination, and target clustering on the current frame point cloud.

[0040] In a specific embodiment, determining at least one obstacle target in the current frame point cloud comprises the following steps:

[0041] Step 11, screening the current frame point cloud to obtain point clouds in the current frame point cloud located in a region of interest;

[0042] Step 12, based on a preset vehicle body size parameter, a minimum height threshold, and a maximum height threshold, removing vehicle body point clouds, ground point clouds, and suspended point clouds in the point clouds located in the region of interest to obtain obstacle point clouds;

[0043] Step 13, dividing the region of interest into a grid map, and determining the state of each grid according to the corresponding obstacle points of each grid in the obstacle point clouds, wherein the state of the grid is an occupied state or a non-occupied state;

[0044] Step 14, determining at least one obstacle target according to the state of each grid.

[0045] The region of interest can be a pre-set three-dimensional region without height restriction. Specifically, the point clouds located in the region of interest can be screened from the current frame point cloud through the region of interest. Further, the vehicle body point clouds in the point clouds located in the region of interest are removed through the preset vehicle body size parameter, and the points less than the minimum height threshold are removed from the point clouds located in the region of interest, and the points greater than the maximum height threshold are removed from the point clouds located in the region of interest. Finally, the point clouds located in the region of interest after removal are used as obstacle point clouds.

[0046] Further, the region of interest can be divided into a grid map of HxW according to a set size, and for each grid in the grid map, a corresponding relationship between the grid and the obstacle points in the obstacle point clouds located in the grid is established, that is, the obstacle points in the grid can be indexed through the grid coordinates.

[0047] Further, the height information of all obstacle points in each grid can be counted, the highest obstacle point and the lowest obstacle point in each grid are recorded, and the state of the grid is determined according to the height of the highest obstacle point and the lowest obstacle point. The state of the grid can be an occupied state or a non-occupied state, and the occupied state indicates that the grid is occupied by an obstacle, and the non-occupied state indicates that the grid is not occupied by an obstacle.

[0048] For example, for each grid, if the height of the highest obstacle point in the grid is greater than a set height threshold, and the relative height difference between the highest obstacle point and the lowest obstacle point is greater than a set height difference threshold, it can be determined that the state of the grid is the occupied state (such as NO-PASS); if the height of the highest obstacle point in the grid is less than the set height threshold, or there is no obstacle point in the grid, or the relative height difference between the highest obstacle point and the lowest obstacle point is not greater than the set height difference threshold, it can be determined that the state of the grid is the non-occupied state (such as PASS).

[0049] Further, after determining the state of each grid, the connected region of each obstacle target can be obtained by performing connected domain calculation on each grid in the occupied state, the coordinates, length, width and maximum height of the occupied grid of the obstacle target are determined through the connected region, and the identification of at least one obstacle target in the current frame point cloud is realized. For example, for each grid in the occupied state, if the state of the adjacent other grid of the grid is the occupied state, it can be determined that the grid and the other grid are connected, and in this way, the connected region of each obstacle target can be determined in the grid map.

[0050] Through steps 11-14, the vehicle body point cloud, the ground point cloud and the suspended target point cloud that does not affect the passage can be filtered out, only the obstacle point cloud in the region of interest that affects the passability is retained, the efficiency of obstacle target identification is improved, and the determination of the grid state and the target clustering can be realized, and the accuracy of the obstacle target is ensured.

[0051] In the embodiments of the present disclosure, after determining at least one obstacle target in the current frame point cloud, for each obstacle target, the final node statistical information can be determined according to the corresponding point cloud block. The point cloud block can be composed of all point clouds corresponding to the obstacle target, and the final node statistical information can include a guardrail node list, a guardrail node number and a boundary node number. The guardrail node list is composed of at least one guardrail node, the guardrail node can be a local point cloud block detected as a point cloud block that meets the guardrail width feature, and the boundary node can be a guardrail node detected to be located on the boundary of the guardrail detection area.

[0052] In a specific embodiment, for each obstacle target, the final node statistical information of the obstacle target is determined based on the point cloud block corresponding to the obstacle target, including the following steps:

[0053] Step 21, for each obstacle target, based on each grid corresponding to the obstacle target, determine each horizontal point cloud block corresponding to the obstacle target in the horizontal coordinate direction, wherein the horizontal coordinate direction is the direction perpendicular to the vehicle head, and the horizontal point cloud block is each row in the point cloud block of the obstacle target;

[0054] Step 22, starting from the first horizontal point cloud block in each horizontal point cloud block, determine whether there is a horizontal guardrail node and a horizontal boundary node according to the width of the horizontal point cloud block and the boundary of the guardrail detection area, until the width of the horizontal point cloud block is not less than the preset guardrail width threshold, to obtain the horizontal guardrail node list, the number of horizontal guardrail nodes and the number of horizontal boundary nodes;

[0055] Step 23, based on each grid corresponding to the obstacle target, determine each vertical point cloud block corresponding to the obstacle target in the vertical coordinate direction, wherein the vertical coordinate direction is the direction of the vehicle head;

[0056] Step 24, starting from the first vertical point cloud block in each vertical point cloud block, determine whether there is a vertical guardrail node and a vertical boundary node according to the width of the vertical point cloud block and the boundary of the guardrail detection area, until the width of the vertical point cloud block is not less than the preset guardrail width threshold or all vertical point cloud blocks are determined, to obtain the vertical guardrail node list, the number of vertical guardrail nodes and the number of vertical boundary nodes;

[0057] Step 25, if the number of horizontal guardrail nodes is greater than the number of vertical guardrail nodes, then determine the final node statistical information of the obstacle target according to the horizontal guardrail node list, the number of horizontal guardrail nodes and the number of horizontal boundary nodes, otherwise, determine the final node statistical information of the obstacle target according to the vertical guardrail node list, the number of vertical guardrail nodes and the number of vertical boundary nodes.

[0058] Wherein, the horizontal coordinate direction and the vertical coordinate direction can construct a plane parallel to the ground, the horizontal coordinate direction can be understood as the direction perpendicular to the vehicle head in the plane, and the vertical coordinate direction can be understood as the direction of the vehicle head. In the above step 21, the horizontal point cloud block can be a horizontal local point cloud block in the point cloud block of the obstacle target, that is, each row in the point cloud block of the obstacle target.

[0059] Exemplary, Figure 2 A schematic diagram of a horizontal point cloud block in an embodiment of the present disclosure is as follows, Figure 2As shown in the figure, the point cloud block of the upper left obstacle target can be divided into four horizontal point cloud blocks, i.e., the four horizontal point cloud blocks are the grids occupied by the obstacle target in the 6th row to the 9th row, and the point cloud block of the lower right obstacle target can be divided into four horizontal point cloud blocks, i.e., the four horizontal point cloud blocks are the grids occupied by the obstacle target in the 3rd row to the 6th row.

[0060] Specifically, in the step 22, the first horizontal point cloud block in all horizontal point cloud blocks can be started, whether the horizontal point cloud block meets the guardrail width feature is determined according to the width of the horizontal point cloud block, if yes, the horizontal guardrail node is determined according to the horizontal point cloud block, and whether the horizontal guardrail node is the horizontal boundary node is further determined in combination with the guardrail detection region boundary.

[0061] For example, for the upper left obstacle target, whether the horizontal guardrail node and the horizontal boundary node exist in each horizontal point cloud block can be determined in turn from bottom to top. Figure 2

[0062] For the step 22, optionally, the first horizontal point cloud block in each horizontal point cloud block can be started, whether the horizontal guardrail node and the horizontal boundary node exist are determined according to the width of the horizontal point cloud block and the guardrail detection region boundary, until the width of the horizontal point cloud block is not less than the preset guardrail width threshold, the horizontal guardrail node list, the horizontal guardrail node number and the horizontal boundary node number are obtained, including:

[0063] The first horizontal point cloud block in the horizontal coordinate direction is taken as the current point cloud block, each grid in the current point cloud block is traversed along the horizontal coordinate direction, the width of the current point cloud block is obtained, and whether the width of the current point cloud block is less than the preset guardrail width threshold is determined;

[0064] If yes, the center point of the current point cloud block is determined as the horizontal guardrail node, the horizontal guardrail node number is updated, and the horizontal guardrail node is determined as the horizontal boundary node in the case that the horizontal guardrail node is located on the guardrail detection region boundary, and the horizontal boundary node number is updated;

[0065] The next horizontal point cloud block of the current point cloud block is taken as a new current point cloud block, and the step of traversing each grid in the current point cloud block along the horizontal coordinate direction is returned to be executed until the width of the horizontal point cloud block is not less than the preset guardrail width threshold, the horizontal guardrail node list, the horizontal guardrail node number and the horizontal boundary node number are obtained.

[0066] ​Specifically, starting from the first horizontal point cloud block, the first horizontal point cloud block is taken as a current point cloud block, each grid in the current point cloud block is traversed along the horizontal coordinate direction, the width of the current point cloud block is determined according to the number of the grids in the current point cloud block that are traversed, and then it is judged whether the width is less than the preset guardrail width threshold. If yes, it indicates that the current point cloud block meets the guardrail width feature, that is, the guardrail has the feature of smaller width compared with other targets, so it can be determined that the current point cloud block is a guardrail, the center point of the current point cloud block is taken as a horizontal guardrail node, the number of horizontal guardrail nodes is incremented by one, and the global coordinates and the maximum height of the horizontal guardrail node are recorded. The maximum height of the horizontal guardrail node can be the height value of the highest point in the corresponding horizontal point cloud block.

[0067] Further, it can be judged whether the horizontal guardrail node is located on the boundary of the guardrail detection area. If yes, the horizontal guardrail node is determined as a horizontal boundary node, and the number of horizontal boundary nodes is incremented by one. The next horizontal point cloud block of the current point cloud block is taken as a new current point cloud block, and the above process is repeated until the width of the horizontal point cloud block is not less than the preset guardrail width threshold. At this time, the iteration process can be stopped, and the horizontal guardrail node list, the number of horizontal guardrail nodes and the number of horizontal boundary nodes are obtained. In this way, the horizontal traversal of each obstacle target is realized, and then the horizontal guardrail node list, the number of horizontal guardrail nodes and the number of horizontal boundary nodes of each obstacle target in the horizontal coordinate direction are obtained.

[0068] Specifically, in the above step 23, if the width of the horizontal point cloud block of the obstacle target is not less than the preset guardrail width threshold, the horizontal traversal is switched to vertical traversal to count the vertical guardrail node list, the number of vertical guardrail nodes and the number of vertical boundary nodes of the obstacle target in the vertical coordinate direction.

[0069] Specifically, starting from the first vertical point cloud block of the obstacle target, it is judged whether the vertical point cloud block meets the guardrail width feature according to the width of the vertical point cloud block. If yes, the vertical guardrail node is determined according to the vertical point cloud block, and it is further judged whether the vertical guardrail node is a vertical boundary node in combination with the boundary of the guardrail detection area, until the width of the vertical point cloud block is not less than the preset guardrail width threshold or all vertical point cloud blocks are judged.

[0070] Exemplarily, Figure 3 A schematic diagram of a vertical point cloud block in an embodiment of the present disclosure is shown in FIG. 6. As shown in FIG. 6, the vertical point cloud block is a grid in the vertical coordinate direction, and the width of the vertical point cloud block is the number of grids in the vertical coordinate direction. Figure 3As shown in the figure, the point cloud block of the upper left obstacle target can be divided into five longitudinal point cloud blocks, i.e., five longitudinal point cloud blocks are respectively the grids occupied by the obstacle target in columns 4-8. The point cloud block of the lower right obstacle target can be divided into one longitudinal point cloud block, i.e., the grid occupied by the obstacle target in column 7. For the upper left obstacle target, it can be determined in sequence whether each longitudinal point cloud block has a longitudinal guardrail node and a longitudinal boundary node from left to right. The specific determination process can refer to the determination steps of the transverse point cloud block.

[0071] In step 24, in the case that there is a longitudinal point cloud block with a width not less than the preset guardrail width threshold, or in the case that all longitudinal point cloud blocks are determined, the longitudinal traversal can be stopped, and the longitudinal guardrail node list, the longitudinal guardrail node number, and the longitudinal boundary node number are obtained.

[0072] Further, the number of transverse guardrail nodes and the number of longitudinal guardrail nodes can be compared. If the number of transverse guardrail nodes is greater than the number of longitudinal guardrail nodes, it indicates that the guardrail direction of the obstacle target is the longitudinal coordinate direction. At this time, the transverse guardrail node list, the transverse guardrail node number, and the transverse boundary node number can be taken as the final node statistical information, and the guardrail recognition is performed on the obstacle target. Otherwise, it indicates that the guardrail direction of the obstacle target is the horizontal coordinate direction. The longitudinal guardrail node list, the longitudinal guardrail node number, and the longitudinal boundary node number can be taken as the final node statistical information, and the guardrail recognition is performed on the obstacle target.

[0073] Through steps 21-25, the node statistics of each obstacle target are realized, and then it is convenient to detect whether each obstacle target is a guardrail according to the statistical information, and the accuracy of the guardrail detection is ensured.

[0074] In order to further ensure that the final node statistical information is obtained after the transverse traversal and the longitudinal traversal are completed, in an example, after the transverse guardrail node list, the transverse guardrail node number, and the transverse boundary node number are obtained, it further includes determining the current traversal row number. Correspondingly, after the longitudinal guardrail node list, the longitudinal guardrail node number, and the longitudinal boundary node number are obtained, it further includes determining the current traversal column number. It is determined whether the current traversal row number and the current traversal column number are equal to the preset value. If not, the step of determining the final node statistical information of the obstacle target is executed.

[0075] In this example, when the horizontal traversal ends (i.e., when the width of the horizontal point cloud block is not less than the preset guardrail width threshold), the horizontal traversal can be considered complete, and the current traversed row number can be determined. Similarly, when the vertical traversal ends (i.e., when the width of the vertical point cloud block is not less than the preset guardrail width threshold or all vertical point cloud blocks have been determined), the vertical traversal can be considered complete, and the current traversed column number can be determined. Furthermore, if the current traversed row number is not equal to a preset value, and the current traversed column number is not equal to a preset value, then both the horizontal and vertical traversals can be considered complete. At this point, the final node statistics can be determined based on the comparison between the number of horizontal guardrail nodes and the number of vertical guardrail nodes. This example avoids obtaining the final node statistics when the horizontal and vertical traversals are not complete, ensuring the accuracy of the final node statistics.

[0076] For example, with Figure 2 Taking the top-left target as an example, the preset guardrail width threshold can be 3. First, start the horizontal traversal from row=6. When traversing to row=9, the width of the horizontal point cloud block exceeds the preset guardrail width threshold. At this point, we can determine that the current traversal row number is 9, and switch to vertical traversal. Then, start the vertical traversal from col=4. When col=4, the width of the vertical point cloud block exceeds the preset guardrail width threshold. At this point, we can determine that the current traversal column number is 4. Finally, check whether the current traversal row number and the current traversal column number are equal to the preset value (such as -1). If they are not equal, it means that the target has completed the horizontal and vertical traversal, and obtain the horizontal guardrail node list, the number of horizontal guardrail nodes, the number of horizontal boundary nodes, as well as the vertical guardrail node list, the number of vertical guardrail nodes, and the number of vertical boundary nodes.

[0077] In this embodiment of the disclosure, during the lateral traversal of the obstacle target, considering that there is a situation where the width of all lateral point cloud blocks is less than the preset guardrail width threshold, that is, when the last lateral point cloud block is traversed, the width of the lateral point cloud block is less than the preset guardrail width threshold. At this time, the guardrail direction of the obstacle target is highly likely to be the vertical coordinate direction, so the longitudinal traversal of the obstacle target is no longer required.

[0078] For example, optionally, the method provided in this disclosure embodiment further includes: if the width of all horizontal point cloud blocks is less than a preset guardrail width threshold, then the final node statistics of the obstacle target are determined based on the horizontal guardrail node list, the number of horizontal guardrail nodes, and the number of horizontal boundary nodes.

[0079] That is, if the width of all horizontal point cloud blocks is less than the preset guardrail width threshold, the list of horizontal guardrail nodes, the number of horizontal guardrail nodes, and the number of horizontal boundary nodes can be directly used as the final node statistics information without vertical traversal, thus improving the efficiency of node statistics.

[0080] Exemplary, Figure 4 A schematic diagram of a traversal process in an embodiment of the present disclosure is shown in FIG. 1. Figure 4 As shown, for each obstacle target, it can be first determined whether the search direction (search_dir, which can be preset as 0, 0 representing horizontal traversal and 1 representing vertical traversal) is 0. If yes, it means that horizontal traversal is needed, and then it is determined whether the width of each horizontal point cloud block is less than a preset guardrail width threshold. If the width is less than the preset guardrail width threshold, the guardrail node is recorded and the number of guardrail nodes is updated. It is further determined whether the guardrail node is located at the boundary. If yes, the boundary node is recorded and the number of boundary nodes is updated. Then it is determined whether the number of guardrail nodes (denoted as pos_cnt) is equal to the number of horizontal nodes (denoted as row_cnt, i.e., the number of horizontal point cloud blocks contained in the point cloud block corresponding to the obstacle target). If no, the traversal of the next horizontal point cloud block is continued. If yes, the traversal is ended, and the guardrail node list obtained by horizontal traversal, the number of guardrail nodes, and the number of boundary nodes are taken as the final node statistical information.

[0081] If the width is not less than the preset guardrail width threshold, the search direction is adjusted (from 0 to 1), the jump row number (denoted as switch_row, i.e., the row where the currently traversed horizontal point cloud block is located) is recorded, and it is determined whether the jump row number is not equal to the preset value and the jump column number is not equal to the preset value (i.e., the initial values of the jump row number and the jump column number can be set as -1). If yes, it means that the traversal in both horizontal and vertical directions is completed. If no, the traversal needs to be continued, and the step of determining whether the search direction is 0 is returned to, so as to switch to vertical traversal. The process of vertical traversal can refer to the process of horizontal traversal, which will not be described here.

[0082] Exemplary, Figure 5 A schematic diagram of an obstacle target in an embodiment of the present disclosure is shown in FIG. 2. Figure 5 For example, assuming that the preset guardrail width threshold is 4, the number of horizontal nodes row_cnt of the obstacle target is 3, the number of vertical nodes col_cnt is 8, and the jump row number switch_row and the jump column number switch_col take the preset value -1. First, the traversal starts from the first horizontal point cloud block (curr_cow=3) of the obstacle target, and when it reaches the third horizontal point cloud block (curr_cow=5), the width is not less than the preset guardrail width threshold. At this time, the number of guardrail nodes pos_cnt=2, and switch_row=5 (i.e., the third horizontal point cloud block is located at the fifth row) is recorded. It is determined whether switch_row and switch_col are not equal to -1. If yes, the traversal is stopped. If no, the vertical traversal is jumped to.

[0083] Further, starting from the first longitudinal point cloud block (curr_col=3) of the obstacle target, traversing until the eighth longitudinal point cloud block (curr_col=10), the width of each is less than the preset guardrail width threshold, at this time the number of guardrail nodes pos_cnt=8, and the switch_col=10 is recorded. It is determined whether pos_cnt is equal to col_cnt. If yes, the traversal is ended. Or it is determined whether switch_row and switch_col are not equal to -1. If yes, the traversal is stopped.

[0084] In S120, it is determined whether each obstacle target meets a length continuity condition based on the final node statistical information of each obstacle target. For the obstacle target that does not meet the length continuity condition, it is determined whether the obstacle target meets a height continuity condition based on the final node statistical information of the obstacle target.

[0085] The length continuity condition can be used to determine whether the obstacle target is continuous in the guardrail direction thereof, and the height continuity condition is used to determine whether the obstacle target is continuous in height.

[0086] In a specific embodiment, the determination of whether each obstacle target meets the length continuity condition based on the final node statistical information of each obstacle target includes: for each obstacle target, determining a distance between two most distant guardrail nodes in the guardrail node list in the final node statistical information of the obstacle target, and determining the distance as a maximum length of the obstacle target; and if the maximum length of the obstacle target is greater than a preset length threshold, determining that the obstacle target meets the length continuity condition.

[0087] Specifically, the global coordinates and the maximum height of each guardrail node in the guardrail node list can be recorded, and the distance between the two most distant guardrail nodes can be obtained through the guardrail node list, and the distance is taken as the maximum length of the obstacle target.

[0088] Further, if the maximum length is greater than the preset length threshold, it indicates that the obstacle target is continuous in the guardrail direction thereof, and meets the length continuity condition. Through this embodiment, the length continuity of each obstacle target is accurately determined, and the accuracy of the identified candidate guardrail target is ensured.

[0089] In the embodiments of the present disclosure, for the obstacle target that meets the length continuity condition, it can be determined as a candidate guardrail target, and it is not necessary to determine whether it meets the height continuity condition. For the obstacle target that does not meet the length continuity condition, in order to avoid missing detection of the guardrail due to occlusion or other factors, considering that the guardrail has the characteristic of height continuity in addition to length continuity, it can be further determined whether it meets the height continuity condition. If it meets the height continuity condition, it can also be determined as a candidate guardrail target.

[0090] In a specific embodiment, for the obstacle target not satisfying the length continuity condition, it is determined whether the obstacle target satisfies a height continuity condition based on final node statistics information of the obstacle target, including:

[0091] For the obstacle target not satisfying the length continuity condition, it is determined whether the number of guardrail nodes in the final node statistics information of the obstacle target is greater than a preset first number and whether the number of boundary nodes is less than a preset second number.

[0092] If yes, the maximum height of the guardrail nodes and the maximum height difference between the guardrail nodes are determined based on the guardrail node list in the final node statistics information of the obstacle target, it is determined whether the maximum height is less than a preset maximum height threshold and whether the maximum height difference is less than a preset maximum height difference threshold, and if yes, it is determined that the obstacle target satisfies the height continuity condition.

[0093] For the obstacle target not satisfying the length continuity condition, it can be determined whether the number of guardrail nodes of the obstacle target is greater than a preset first number and whether the number of boundary nodes is less than a preset second number, and if yes, it indicates that the obstacle target can be a guardrail. Further, it can be determined whether it satisfies the height continuity condition in combination with the maximum height of the guardrail nodes and the maximum height difference between the guardrail nodes.

[0094] Specifically, if the maximum height of the guardrail nodes is less than a preset maximum height threshold and the maximum height difference between the guardrail nodes is less than a preset maximum height difference threshold, it can be determined that the obstacle target is continuous in height and satisfies the height continuity condition, i.e., has height continuity. Through this embodiment, accurate determination of the height continuity of each obstacle target is realized, and the accuracy of the identified candidate guardrail target is ensured.

[0095] It should be noted that if the maximum height of the guardrail nodes of the obstacle target is not less than a preset maximum height threshold or the maximum height difference is not less than a preset maximum height difference threshold, it can continue to determine whether it satisfies the height continuity condition in combination with the number of continuous nodes.

[0096] Optionally, the method provided by the embodiments of the present disclosure further includes: if the maximum height is not less than a preset maximum height threshold or the maximum height difference is not less than a preset maximum height difference threshold, determining the height difference between adjacent guardrail nodes based on the guardrail node list in the final node statistics information of the obstacle target; determining the number of continuous nodes according to the height difference between all adjacent guardrail nodes, and if the number of continuous nodes is greater than a preset continuous number, determining that the obstacle target satisfies the height continuity condition; wherein the preset continuous number is the product of the number of guardrail nodes in the final node statistics information and a height consistency proportionality coefficient.

[0097] That is, the height difference between any two adjacent guardrail nodes can be continuously determined, and for each height difference, if the height difference is less than a set height difference threshold, the two guardrail nodes corresponding to the height difference can be determined as continuous nodes, and the number of continuous nodes is updated. Further, it is determined whether the number of continuous nodes is greater than a preset continuous number, and if so, it is determined that the obstacle target satisfies the height continuity condition.

[0098] It should be noted that the preset continuous number can be determined according to the product of the number of guardrail nodes and the height consistency proportion coefficient, wherein the height consistency proportion coefficient can have a value range of [0, 1]. For example, the height consistency proportion coefficient can take a set value, or the height consistency proportion coefficient can be determined according to the maximum length of the obstacle target, for example, the height consistency proportion coefficient can be obtained according to the coefficients corresponding to each length described in the calibration table.

[0099] In the above manner, the number of continuous nodes can be further combined to determine whether the obstacle target satisfies the height continuity condition, to check the continuity between adjacent nodes, and further ensure the detection accuracy of the candidate guardrail target.

[0100] S130, determining the obstacle target satisfying the length continuity condition and the obstacle target satisfying the height continuity condition as the candidate guardrail target of the current frame.

[0101] Specifically, for the obstacle target satisfying the length continuity condition, it can be determined as the candidate guardrail target of the current frame, and for the obstacle target satisfying the height continuity condition, it can be determined as the candidate guardrail target of the current frame.

[0102] S140, determining the official guardrail target of the current frame based on each candidate guardrail target of the current frame and each official guardrail target of the historical frame, and determining the guardrail detection result of the current frame according to each official guardrail target of the current frame.

[0103] Specifically, each candidate guardrail target of the current frame can be matched with each official guardrail target of the historical frame, and each official guardrail target of the current frame can be determined according to the matching result, and then the guardrail detection result of the current frame can be obtained by curve fitting on each official guardrail target of the current frame.

[0104] In a specific embodiment, determining the official guardrail target of the current frame based on each candidate guardrail target of the current frame and each official guardrail target of the historical frame includes the following steps:

[0105] Step 31, for each candidate guardrail target of the current frame, projecting the candidate guardrail target into the road network map to obtain the position of the candidate guardrail target on the road network map;

[0106] Step 32, position matching is performed on each candidate guardrail target of the current frame and each official guardrail target of the historical frame, and each candidate guardrail target of the current frame that is successfully position matched is determined as each official guardrail target of the current frame;

[0107] Step 33, for each candidate guardrail target of the current frame that fails in the position matching, each candidate guardrail target located on the same straight line is merged to obtain a merged guardrail target, and each merged guardrail target with a length greater than a preset threshold is determined as each official guardrail target of the current frame.

[0108] Specifically, in step 31, each candidate guardrail target can be projected into a road network map first, and a relative offset distance of each candidate guardrail target relative to a road network line is calculated to obtain a position of each candidate guardrail target in the map, and each candidate guardrail target is initialized according to the position, that is, each candidate guardrail target is transformed from a vehicle coordinate system to a map coordinate system.

[0109] Further, each candidate guardrail target of the current frame is position matched with each official guardrail target of the historical frame (such as the previous k frames), and each candidate guardrail target that is successfully matched can be determined as an official guardrail target of the current frame.

[0110] For each candidate guardrail target that fails in the matching, each candidate guardrail target located on the same straight line is merged to obtain a merged guardrail target, and each merged guardrail target with a length greater than a preset threshold is determined as each official guardrail target of the current frame.

[0111] In this way, the positions of the official guardrail targets of the historical frame are combined, and each candidate guardrail target located on the same straight line that is possibly connected is merged to realize the screening of the official guardrail targets and ensure the accuracy of the official guardrail targets.

[0112] It is considered that each official guardrail target of the historical frame can have an official guardrail target that has not been successfully matched for a long time, for example, due to vehicle driving out of the area where the guardrail is located. In order to further ensure the accuracy of the guardrail detection in the current frame, in an optional implementation, the method provided by the embodiment of the disclosure further includes: for each candidate guardrail target or each merged guardrail target that is determined as an official guardrail target, updating a corresponding state identifier.

[0113] Correspondingly, before the position matching is performed on each candidate guardrail target of the current frame and each official guardrail target of the historical frame, it further includes: in each official guardrail target of the historical frame, each official guardrail target with a frame number of the state identifier that is not updated exceeding a set frame number is removed.

[0114] That is, for each candidate guardrail target or each merged guardrail target that is obtained after the merging and that is successfully matched, the corresponding state identifier can be updated. The state identifier is used to describe whether the guardrail target is successfully matched or merged.

[0115] Further, before position matching each candidate guardrail target of the current frame with each official guardrail target of the historical frame, the official guardrail target whose state identifier is not updated for more than a set number of frames can be removed from each official guardrail target of the historical frame, so as to avoid matching each candidate guardrail target of the current frame with the guardrail that has left the field of view of the vehicle, and further ensure the identification accuracy of the official guardrail target of the current frame.

[0116] After obtaining each official guardrail target of the current frame, the least square method can be used to fit each official guardrail target of the current frame, to obtain a fitted guardrail curve, and the guardrail curve is taken as the guardrail detection result of the current frame.

[0117] In the embodiments of the present disclosure, by using multiple judgment methods such as the length continuity condition and the height continuity condition, the guardrail target can be stably and effectively detected, and the real-time operation requirement of the automatic driving embedded platform can be met. In addition, a guardrail maximum width screening algorithm based on a two-dimensional plane grid map is proposed, the potential guardrail target in each direction is searched by window traversal, and it is ensured that the guardrail target can be effectively detected and is not easy to be missed. In addition, a guardrail screening algorithm combining the height and length continuity of the guardrail target is proposed, which does not depend on a single threshold, and the guardrail target judgment result is more accurate and more robust.

[0118] In addition, by using the map matching method, the detected candidate guardrail target is projected onto the map by coordinate system transformation on the basis of the existing map, the position of the guardrail is further verified, and the accuracy of the guardrail detection result is ensured.

[0119] The guardrail detection method provided by the embodiments of the present disclosure comprises the following steps: acquiring a current frame of point clouds, determining at least one obstacle target in the current frame of point clouds, determining, for each obstacle target, final node statistical information of the obstacle target through a corresponding point cloud block of the obstacle target, judging whether the obstacle target meets a length continuity condition according to the final node statistical information, judging whether the obstacle target that does not meet the length continuity condition meets a height continuity condition according to the final node statistical information, determining the obstacle target that meets the length continuity condition and the obstacle target that meets the height continuity condition as candidate guardrail targets of the current frame, and determining each official guardrail target of the current frame in combination with each official guardrail target of a historical frame, thereby determining a guardrail detection result of the current frame. The guardrail detection method based on the laser radar point cloud is realized. In the detection process of the guardrail target, the length continuity and the height continuity of the guardrail are considered, and the result of the detection of the historical frame is combined for guardrail identification, so as to avoid the missed detection of the guardrail as much as possible, improve the robustness of the guardrail detection, solve the problem of the missed detection of the guardrail caused by the occlusion or other factors of the guardrail point cloud, realize the fast, effective and stable detection of the guardrail target, and do not increase the calculation load, are not affected by the use scene and the sensor configuration, and can be run in real time under the condition that the calculation resources of the vehicle-mounted platform are limited.

[0120] In addition, the method provided by the embodiments of the present disclosure can accurately detect the guardrail target through multiple feature verifications. Compared with the data-driven deep learning method, the detection speed is faster, the calculation resource occupation amount is smaller, and no labeled data needs to be collected, thereby saving the cost. The robustness of the guardrail detection is greatly improved based on the existing map data.

[0121] Figure 6 FIG. 1 is a structural schematic diagram of a guardrail detection device according to an embodiment of the present disclosure. As shown in the figure, the device comprises a node statistical module 610, a continuity judgment module 620, a candidate determination module 630 and an official determination target 640, wherein: Figure 6

[0122] The node statistical module 610 is configured to acquire a current frame of point clouds, and determine at least one obstacle target in the current frame of point clouds. For each obstacle target, the final node statistical information of the obstacle target is determined based on the point cloud block corresponding to the obstacle target.

[0123] The continuity judgment module 620 is configured to judge whether each obstacle target meets a length continuity condition based on the final node statistical information of each obstacle target. For the obstacle target that does not meet the length continuity condition, whether the obstacle target meets a height continuity condition is judged based on the final node statistical information of the obstacle target.

[0124] ​The candidate determination module 630 is used to determine the obstacle targets that satisfy the length continuity condition and the obstacle targets that satisfy the height continuity condition as candidate guardrail targets for the current frame;

[0125] The formal target 640 is used to determine each formal guardrail target in the current frame based on each candidate guardrail target in the current frame and each formal guardrail target in the historical frames, and to determine the guardrail detection result of the current frame based on each formal guardrail target in the current frame.

[0126] The guardrail detection device provided in this embodiment can execute the steps in the guardrail detection method provided in this embodiment, and has the execution steps and beneficial effects, which will not be repeated here.

[0127] Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of this disclosure. See below for details. Figure 7 It shows a schematic diagram of a structure suitable for implementing the electronic device 500 in the embodiments of this disclosure. Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0128] like Figure 7 As shown, the electronic device 500 may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 501, which can perform various appropriate actions and processes to implement the methods of the embodiments described herein, based on a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device 500. The processing device 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0129] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts, thereby implementing the guardrail detection method as described above. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by the processing device 501, it performs the functions defined in the methods of embodiments of this disclosure.

[0130] It is noted that the computer-readable medium described above can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, a computer-readable storage medium can be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium can include a computer-readable storage medium in baseband or propagated as a carrier wave in a propagated data signal, which can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable signal medium can be any computer-readable medium that is not a computer-readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0131] The computer-readable medium described above can be included within the electronic device described above; or can exist separately from the electronic device and be not assembled into the electronic device. The computer-readable medium described above carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the steps of any of the embodiments described above.

[0132] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable storage media can include, without limitation, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0133] Scheme 1, a guardrail detection method, the method comprising:

[0134] obtaining a current frame point cloud, and determining at least one obstacle target in the current frame point cloud, for each of the obstacle targets, determining final node statistical information of the obstacle target based on a point cloud block corresponding to the obstacle target;

[0135] determining whether each of the obstacle targets meets a length continuity condition based on the final node statistical information of each of the obstacle targets, for an obstacle target that does not meet the length continuity condition, determining whether the obstacle target meets a height continuity condition based on the final node statistical information of the obstacle target;

[0136] determining an obstacle target that meets the length continuity condition and an obstacle target that meets the height continuity condition as a candidate guardrail target of the current frame;

[0137] determining each official guardrail target of the current frame based on each of the candidate guardrail targets of the current frame and each official guardrail target of a historical frame, and determining a guardrail detection result of the current frame according to each of the official guardrail targets of the current frame.

[0138] Scheme 2, the method according to scheme 1, the determining at least one obstacle target in the current frame point cloud comprises:

[0139] screening the current frame point cloud to obtain point clouds in the current frame point cloud that are located in a region of interest;

[0140] based on a preset vehicle body size parameter, a minimum height threshold, and a maximum height threshold, removing vehicle body point clouds, ground point clouds, and suspended point clouds in the point clouds located in the region of interest to obtain obstacle point clouds;

[0141] rasterizing the region of interest to obtain a corresponding raster map, and determining a state of each of the rasters according to corresponding obstacle points of each of the rasters in the obstacle point cloud, wherein the state of each of the rasters is an occupied state or a non-occupied state;

[0142] determining at least one obstacle target according to the state of each of the rasters.

[0143] Scheme 3, the method according to Scheme 2,

[0144] for each of the obstacle targets, determining final node statistical information of the obstacle target based on the point cloud block corresponding to the obstacle target, including:

[0145] for each of the obstacle targets, determining each of the lateral point cloud blocks corresponding to the obstacle target in a lateral coordinate direction based on the rasters corresponding to the obstacle target, wherein the lateral coordinate direction is a direction perpendicular to the vehicle head, and each of the lateral point cloud blocks is each row in the point cloud block of the obstacle target;

[0146] starting from a first of the lateral point cloud blocks, determining whether there are a lateral guardrail node and a lateral boundary node according to the width of the lateral point cloud block and the guardrail detection region boundary until the width of the lateral point cloud block is not less than a preset guardrail width threshold, to obtain a lateral guardrail node list, a lateral guardrail node quantity, and a lateral boundary node quantity;

[0147] determining each of the longitudinal point cloud blocks corresponding to the obstacle target in a longitudinal coordinate direction based on the rasters corresponding to the obstacle target, wherein the longitudinal coordinate direction is the vehicle head direction;

[0148] starting from a first of the longitudinal point cloud blocks, determining whether there are a longitudinal guardrail node and a longitudinal boundary node according to the width of the longitudinal point cloud block and the guardrail detection region boundary until the width of the longitudinal point cloud block is not less than a preset guardrail width threshold or all of the longitudinal point cloud blocks are determined, to obtain a longitudinal guardrail node list, a longitudinal guardrail node quantity, and a longitudinal boundary node quantity;

[0149] if the lateral guardrail node quantity is greater than the longitudinal guardrail node quantity, determining the final node statistical information of the obstacle target according to the lateral guardrail node list, the lateral guardrail node quantity, and the lateral boundary node quantity, otherwise, determining the final node statistical information of the obstacle target according to the longitudinal guardrail node list, the longitudinal guardrail node quantity, and the longitudinal boundary node quantity.

[0150] Option 4, according to the method of option 3, starting from the first lateral point cloud block in each of the lateral point cloud blocks, determining whether there is a lateral guardrail node and a lateral boundary node according to the width of the lateral point cloud block and the guardrail detection region boundary, until the width of the lateral point cloud block is not less than the preset guardrail width threshold, obtaining the lateral guardrail node list, the number of lateral guardrail nodes and the number of lateral boundary nodes, comprising:

[0151] Taking the first lateral point cloud block in the abscissa direction as the current point cloud block, traversing each grid in the current point cloud block along the abscissa direction to obtain the width of the current point cloud block, and determining whether the width of the current point cloud block is less than the preset guardrail width threshold;

[0152] If yes, the center point of the current point cloud block is determined as a lateral guardrail node, the number of lateral guardrail nodes is updated, and the lateral guardrail node is determined as a lateral boundary node in the case that the lateral guardrail node is located on the guardrail detection region boundary, and the number of lateral boundary nodes is updated;

[0153] The next lateral point cloud block of the current point cloud block is taken as a new current point cloud block, and the step of traversing each grid in the current point cloud block along the abscissa direction is returned to be executed until the width of the lateral point cloud block is not less than the preset guardrail width threshold, obtaining the lateral guardrail node list, the number of lateral guardrail nodes and the number of lateral boundary nodes.

[0154] Option 5, according to the method of option 3, the method further comprises:

[0155] If the width of all lateral point cloud blocks is less than the preset guardrail width threshold, the final node statistical information of the obstacle target is determined according to the lateral guardrail node list, the number of lateral guardrail nodes and the number of lateral boundary nodes.

[0156] Option 6, according to the method of option 3, after obtaining the lateral guardrail node list, the number of lateral guardrail nodes and the number of lateral boundary nodes, further comprising:

[0157] Determine the current traversal row number;

[0158] Correspondingly, after obtaining the longitudinal guardrail node list, the number of longitudinal guardrail nodes and the number of longitudinal boundary nodes, further comprising:

[0159] Determine the current traversal column number;

[0160] Determine whether the current traversal row number and the current traversal column number are equal to the preset value, if not, execute the step of determining the final node statistical information of the obstacle target.

[0161] In scheme 7, the method according to scheme 3, determining whether each obstacle object satisfies the length continuity condition based on the final node statistical information of each obstacle object comprises:

[0162] For each obstacle object, determining the distance between the two guardrail nodes farthest away based on the guardrail node list in the final node statistical information of the obstacle object, and determining the distance as the maximum length of the obstacle object;

[0163] If the maximum length of the obstacle object is greater than a preset length threshold, it is determined that the obstacle object satisfies the length continuity condition.

[0164] In scheme 8, the method according to scheme 3, for the obstacle object that does not satisfy the length continuity condition, determining whether the obstacle object satisfies the height continuity condition based on the final node statistical information of the obstacle object comprises:

[0165] For the obstacle object that does not satisfy the length continuity condition, determining whether the number of guardrail nodes in the final node statistical information of the obstacle object is greater than a preset first number and the number of boundary nodes is less than a preset second number;

[0166] If yes, determining the maximum height of the guardrail nodes and the maximum height difference between the guardrail nodes based on the guardrail node list in the final node statistical information of the obstacle object, determining whether the maximum height is less than a preset maximum height threshold and the maximum height difference is less than a preset maximum height difference threshold, and if yes, determining that the obstacle object satisfies the height continuity condition.

[0167] In scheme 9, the method according to scheme 8, the method further comprises:

[0168] If the maximum height is not less than the preset maximum height threshold or the maximum height difference is not less than the preset maximum height difference threshold, determining the height difference between adjacent guardrail nodes based on the guardrail node list in the final node statistical information of the obstacle object;

[0169] Determining the number of continuous nodes according to the height differences between all adjacent guardrail nodes, and if the number of continuous nodes is greater than a preset continuous number, it is determined that the obstacle object satisfies the height continuity condition.

[0170] The preset continuous number is the product of the number of guardrail nodes in the final node statistical information and a height consistency proportionality coefficient.

[0171] In scheme 10, the method according to scheme 1, determining each official guardrail object of the current frame based on each candidate guardrail object of the current frame and each official guardrail object of the historical frame comprises:

[0172] For each of the candidate guardrail targets of the current frame, the candidate guardrail target is projected into a road network map to obtain a position of the candidate guardrail target on the road network map;

[0173] The position matching is performed on each of the candidate guardrail targets of the current frame and each of the official guardrail targets of the historical frame, and each of the candidate guardrail targets of the current frame that is successfully position matched is determined as an official guardrail target of the current frame;

[0174] For each of the candidate guardrail targets of the current frame that fails in the position matching, the candidate guardrail targets located on the same straight line are merged to obtain a merged guardrail target, and the merged guardrail target with a length greater than a preset threshold is determined as an official guardrail target of the current frame.

[0175] Scheme 11, the method according to scheme 10, the method further comprises:

[0176] For the candidate guardrail target or the merged guardrail target determined as the official guardrail target, the corresponding state identifier is updated;

[0177] Correspondingly, before the position matching is performed on each of the candidate guardrail targets of the current frame and each of the official guardrail targets of the historical frame, the method further comprises:

[0178] Among the official guardrail targets of the historical frame, the official guardrail target with a frame number of the state identifier not being updated exceeding a set frame number is removed.

[0179] Scheme 12, a guardrail detection device, comprising:

[0180] A node statistics module is configured to acquire a current frame point cloud, and determine at least one obstacle target in the current frame point cloud, and for each of the obstacle targets, determine final node statistics information of the obstacle target based on a point cloud block corresponding to the obstacle target;

[0181] A continuous judgment module is configured to judge whether each of the obstacle targets meets a length continuity condition based on the final node statistics information of each of the obstacle targets, and for the obstacle target that does not meet the length continuity condition, judge whether the obstacle target meets a height continuity condition based on the final node statistics information of the obstacle target;

[0182] A candidate determination module is configured to determine the obstacle target meeting the length continuity condition and the obstacle target meeting the height continuity condition as a candidate guardrail target of a current frame;

[0183] Formally determine the target, for determining each formal guardrail target of the current frame based on each candidate guardrail target of the current frame and each formal guardrail target of the historical frame, and determining the guardrail detection result of the current frame according to each formal guardrail target of the current frame.

[0184] Scheme 13, an electronic device, comprising:

[0185] one or more processors;

[0186] a memory device for storing one or more programs;

[0187] When the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of schemes 1-11.

[0188] Scheme 14, a computer readable storage medium, having stored thereon a computer program, which is executed by a processor to implement the method of any one of schemes 1-11.

[0189] The above description is merely preferred embodiments of the present disclosure and a description of the principles of the technology applied. It should be understood by those skilled in the art that the disclosed scope of the present disclosure is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combinations of the above technical features or their equivalent features without departing from the disclosed concept. For example, the above features are replaced with the technical features disclosed in the present disclosure (but not limited to) having similar functions to form technical solutions.

Claims

1. A method for detecting guardrails, characterized in that, The method includes: The current frame point cloud is acquired, and at least one obstacle target in the current frame point cloud is identified. For each obstacle target, based on the point cloud block corresponding to the obstacle target, the final node statistics of the obstacle target are determined. The final node statistics include a list of guardrail nodes, the number of guardrail nodes, and the number of boundary nodes. Based on the final node statistics of each obstacle target, it is determined whether each obstacle target satisfies the length continuity condition. For obstacle targets that do not satisfy the length continuity condition, it is determined whether the obstacle target satisfies the height continuity condition based on the final node statistics of the obstacle target. Obstacle targets that satisfy the length continuity condition and obstacle targets that satisfy the height continuity condition are identified as candidate guardrail targets for the current frame; For each candidate guardrail target in the current frame, the candidate guardrail target is projected onto the road network map to obtain the position of the candidate guardrail target on the road network map; position matching is performed on each candidate guardrail target in the current frame and each official guardrail target in the historical frames, and the candidate guardrail targets in the current frame with successful position matching are determined as the official guardrail targets in the current frame; for each candidate guardrail target in the current frame with unsuccessful position matching, the candidate guardrail targets located on the same straight line are merged to obtain merged guardrail targets, and the merged guardrail targets with a length greater than a preset threshold are determined as the official guardrail targets in the current frame, and the guardrail detection result of the current frame is determined based on the official guardrail targets in the current frame.

2. The method according to claim 1, characterized in that, Determining at least one obstacle target in the current frame point cloud includes: The current frame point cloud is filtered to obtain the point cloud located within the region of interest in the current frame point cloud; Based on preset vehicle body size parameters, minimum height threshold and maximum height threshold, the vehicle body point cloud, ground point cloud and suspended point cloud in the point cloud located in the region of interest are removed to obtain the obstacle point cloud; The region of interest is divided into grid maps to obtain corresponding grid maps. The state of each grid is determined according to the obstacle points corresponding to each grid in the obstacle point cloud. The state of each grid is either occupied or unoccupied. At least one obstacle target is determined based on the state of each grid.

3. The method according to claim 2, characterized in that, For each obstacle target, based on the point cloud block corresponding to the obstacle target, the final node statistics information of the obstacle target is determined, including: For each obstacle target, based on each grid corresponding to the obstacle target, determine each horizontal point cloud block corresponding to the obstacle target in the horizontal coordinate direction, wherein the horizontal coordinate direction is the direction perpendicular to the front of the vehicle, and the horizontal point cloud block is each row in the point cloud block of the obstacle target; Starting from the first horizontal point cloud block in each of the horizontal point cloud blocks, determine whether there are horizontal guardrail nodes and horizontal boundary nodes based on the width of the horizontal point cloud block and the boundary of the guardrail detection area, until the width of the horizontal point cloud block is not less than the preset guardrail width threshold, and obtain the list of horizontal guardrail nodes, the number of horizontal guardrail nodes and the number of horizontal boundary nodes. Based on each grid corresponding to the obstacle target, determine each longitudinal point cloud block corresponding to the obstacle target in the longitudinal coordinate direction, wherein the longitudinal coordinate direction is the direction of the vehicle's front; Starting from the first vertical point cloud block in each of the vertical point cloud blocks, determine whether there are vertical guardrail nodes and vertical boundary nodes based on the width of the vertical point cloud block and the boundary of the guardrail detection area, until the width of the vertical point cloud block is not less than the preset guardrail width threshold or all vertical point cloud blocks are determined, and obtain the list of vertical guardrail nodes, the number of vertical guardrail nodes and the number of vertical boundary nodes. If the number of horizontal guardrail nodes is greater than the number of vertical guardrail nodes, the final node statistics of the obstacle target are determined based on the list of horizontal guardrail nodes, the number of horizontal guardrail nodes, and the number of horizontal boundary nodes; otherwise, the final node statistics of the obstacle target are determined based on the list of vertical guardrail nodes, the number of vertical guardrail nodes, and the number of vertical boundary nodes.

4. The method according to claim 3, characterized in that, Starting from the first horizontal point cloud block in each of the aforementioned horizontal point cloud blocks, the presence of horizontal guardrail nodes and horizontal boundary nodes is determined based on the width of the horizontal point cloud block and the boundary of the guardrail detection area, until the width of the horizontal point cloud block is not less than a preset guardrail width threshold, thereby obtaining a list of horizontal guardrail nodes, a number of horizontal guardrail nodes, and a number of horizontal boundary nodes, including: Take the first horizontal point cloud block in the horizontal coordinate direction as the current point cloud block, traverse each grid in the current point cloud block along the horizontal coordinate direction to obtain the width of the current point cloud block, and determine whether the width of the current point cloud block is less than the preset guardrail width threshold. If so, the center point of the current point cloud block is determined as a horizontal guardrail node, the number of horizontal guardrail nodes is updated, and if the horizontal guardrail node is located on the boundary of the guardrail detection area, the horizontal guardrail node is determined as a horizontal boundary node, and the number of horizontal boundary nodes is updated. The next horizontal point cloud block of the current point cloud block is taken as the new current point cloud block, and the step of traversing each grid in the current point cloud block along the horizontal coordinate direction is returned until the width of the horizontal point cloud block is not less than the preset guardrail width threshold, so as to obtain the horizontal guardrail node list, the number of horizontal guardrail nodes and the number of horizontal boundary nodes.

5. The method according to claim 3, characterized in that, The method further includes: If the width of all horizontal point cloud blocks is less than the preset guardrail width threshold, then the final node statistics of the obstacle target are determined based on the horizontal guardrail node list, the number of horizontal guardrail nodes, and the number of horizontal boundary nodes.

6. The method according to claim 3, characterized in that, After obtaining the list of horizontal guardrail nodes, the number of horizontal guardrail nodes, and the number of horizontal boundary nodes, the process also includes: Determine the current row number being traversed; Accordingly, after obtaining the list of longitudinal guardrail nodes, the number of longitudinal guardrail nodes, and the number of longitudinal boundary nodes, the following is also included: Determine the current column number to be traversed; Determine whether the current number of rows and the current number of columns traversed are equal to preset values. If not, proceed to the step of determining the final node statistics of the obstacle target.

7. The method according to claim 3, characterized in that, The step of determining whether each obstacle target satisfies the length continuity condition based on the final node statistical information of each obstacle target includes: For each obstacle target, based on the list of guardrail nodes in the final node statistics information of the obstacle target, the distance between the two farthest guardrail nodes is determined, and the distance is determined as the maximum length of the obstacle target; If the maximum length of the obstacle target is greater than a preset length threshold, then the obstacle target is determined to satisfy the length continuity condition.

8. The method according to claim 3, characterized in that, For obstacle targets that do not meet the length continuity condition, determine whether the obstacle target meets the height continuity condition based on the final node statistics of the obstacle target, including: For obstacle targets that do not meet the length continuity condition, determine whether the number of guardrail nodes in the final node statistics of the obstacle target is greater than a preset first number, and whether the number of boundary nodes is less than a preset second number. If so, based on the guardrail node list in the final node statistics information of the obstacle target, determine the maximum height of the guardrail node and the maximum height difference between the guardrail nodes, determine whether the maximum height is less than a preset maximum height threshold, and whether the maximum height difference is less than a preset maximum height difference threshold. If so, determine that the obstacle target meets the height continuity condition.

9. The method according to claim 8, characterized in that, The method further includes: If the maximum height is not less than the preset maximum height threshold, or the maximum height difference is not less than the preset maximum height difference threshold, then the height difference between adjacent guardrail nodes is determined based on the guardrail node list in the final node statistics information of the obstacle target. The number of consecutive nodes is determined based on the height difference between all adjacent guardrail nodes. If the number of consecutive nodes is greater than the preset number of consecutive nodes, then the obstacle target is determined to meet the height continuity condition. The preset continuous quantity is the product of the number of guardrail nodes and the height consistency ratio coefficient in the final node statistics.

10. The method according to claim 1, characterized in that, The method further includes: For candidate guardrail targets or merged guardrail targets that have been identified as official guardrail targets, update the corresponding status identifiers. Accordingly, before performing position matching on each candidate guardrail target in the current frame and each official guardrail target in historical frames, the method further includes: In the historical frames of each official guardrail target, those official guardrail targets whose status flag has not been updated for more than a set number of frames are removed.

11. A guardrail detection device, characterized in that, include: The node statistics module is used to acquire the current frame point cloud and identify at least one obstacle target in the current frame point cloud. For each obstacle target, based on the point cloud block corresponding to the obstacle target, the final node statistics information of the obstacle target is determined. The final node statistics information includes a list of guardrail nodes, the number of guardrail nodes, and the number of boundary nodes. The continuity judgment module is used to determine whether each obstacle target satisfies the length continuity condition based on the final node statistical information of each obstacle target, and for obstacle targets that do not satisfy the length continuity condition, to determine whether the obstacle target satisfies the height continuity condition based on the final node statistical information of the obstacle target. The candidate determination module is used to determine the obstacle targets that satisfy the length continuity condition and the obstacle targets that satisfy the height continuity condition as candidate guardrail targets for the current frame; The formal target determination module is used to project each of the candidate guardrail targets in the current frame onto the road network map to obtain the position of the candidate guardrail target on the road network map; perform position matching on each of the candidate guardrail targets in the current frame and each of the formal guardrail targets in historical frames, and determine each of the candidate guardrail targets in the current frame that have successfully matched the positions as the formal guardrail targets in the current frame. For each candidate guardrail target in the current frame where the location matching fails, the candidate guardrail targets located on the same straight line are merged to obtain merged guardrail targets. The merged guardrail targets with a length greater than a preset threshold are determined as the official guardrail targets of the current frame. The guardrail detection result of the current frame is determined based on the official guardrail targets of the current frame.

12. An electronic device, characterized in that, The electronic device includes: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-10.

Images