Obstacle detection method, device, equipment and storage medium based on laser radar

Abstract

The present invention provides an obstacle detection method, device, equipment and storage medium based on laser radar, which uses laser radar to scan the target block at the target distance of the self-vehicle to obtain radar point cloud data, inputs the radar point cloud data into the deep learning model, outputs the model detection result of the obstacle in the target block, and obtains the error detection result and the correct detection result of each radar point position in the radar point cloud data in the model detection result. According to the error detection result and the correct detection result, the confidence of the deep learning model corresponding to the target block can be calculated, and then the confidence of different deep learning models corresponding to different target blocks can be automatically and efficiently counted, solving the problem of low efficiency of manual statistics. In this way, during driving, according to the confidence of the deep learning model corresponding to the target block obtained in advance, the target deep learning model with confidence reaching the set conditions is selected for obstacle detection, and the obstacle detection result has high accuracy.

Description

Technical Field

[0001] The present invention relates to the field of driving technology, and in particular to a laser radar-based obstacle detection method, device, equipment and storage medium. Background Art

[0002] Automobiles are becoming increasingly intelligent, and the number of cars equipped with driverless technology is gradually increasing. In driverless technology, accurately sensing objects around the car body is an important prerequisite for safe driving. Summary of the invention

[0003] The present invention provides a laser radar-based obstacle detection method, device, equipment and storage medium, which are used to solve the defect of low accuracy in obstacle detection around a vehicle during driving in the prior art and improve the accuracy of obstacle detection around the vehicle.

[0004] An embodiment of the present invention provides a laser radar-based obstacle detection method, the method comprising:

[0005] Use the laser radar to scan the target area at a distance from the vehicle to obtain radar point cloud data;

[0006] Input radar point cloud data into the deep learning model, output the model detection results of obstacles in the target block, and obtain the error detection results and correct detection results of the positions of each radar point in the radar point cloud data in the model detection results;

[0007] The confidence of the deep learning model corresponding to the target block is calculated based on the false detection results and the correct detection results.

[0008] According to an embodiment of the present invention, a laser radar-based obstacle detection method is provided, which uses a laser radar to scan a target block at a distance from a self-vehicle target to obtain radar point cloud data, including:

[0009] For each target block in multiple target blocks with different target distances or the same target distance from the vehicle, use the lidar to scan the target block at the target distance from the vehicle to obtain radar point cloud data until the confidence of the deep learning model corresponding to each target block is obtained.

[0010] According to an embodiment of the present invention, a laser radar-based obstacle detection method is provided, the method further comprising:

[0011] When the difference between the confidences corresponding to adjacent blocks in multiple target blocks does not exceed the target value, the adjacent blocks are merged to obtain a merged target block, and the error detection results and correct detection results of the adjacent blocks are merged to calculate the confidence of the merged target block corresponding to the deep learning model.

[0012] According to an embodiment of the present invention, a laser radar-based obstacle detection method is provided, the method further comprising:

[0013] Draw a confidence distribution map based on the confidence of multiple target blocks corresponding to the deep learning model.

[0014] According to an embodiment of the present invention, a laser radar-based obstacle detection method is provided, which calculates the confidence of a target block corresponding to a deep learning model according to an erroneous detection result and a correct detection result, including:

[0015] The larger quartile of the confidence of the wrong detection result is obtained, the smaller quartile of the confidence of the correct detection result is obtained, and the average of the larger quartile and the smaller quartile is calculated as the confidence of the target block corresponding to the deep learning model.

[0016] According to an embodiment of the present invention, a laser radar-based obstacle detection method is provided, the method further comprising:

[0017] When the confidences of different deep learning models corresponding to the target blocks are obtained, a target deep learning model whose confidence meets the set conditions is selected from the different deep learning models, and a corresponding relationship between the target deep learning model and the target block is established.

[0018] The embodiment of the present invention further provides a laser radar-based obstacle detection method, the method comprising:

[0019] Use the laser radar to scan the target block in the driving environment that is far from the vehicle target, and obtain the radar point cloud data of the target block;

[0020] The radar point cloud data of the target block is input into the deep learning model, and the obstacle detection result is output. When the confidence level of the pre-acquired deep learning model corresponding to the target block meets the set conditions, the obstacle detection result is adopted.

[0021] According to an obstacle detection method based on laser radar provided by an embodiment of the present invention, before inputting radar point cloud data of a target block into a deep learning model and outputting an obstacle detection result, the method further includes:

[0022] Based on the confidence of the deep learning model corresponding to the target block obtained in advance, a target deep learning model whose confidence meets the set conditions is selected from multiple candidate deep learning models;

[0023] Input the radar point cloud data of the target block into the deep learning model, output the obstacle detection result, and adopt the obstacle detection result when the confidence of the target block corresponding to the pre-acquired deep learning model meets the set conditions, including:

[0024] Input the radar point cloud data of the target area into the target deep learning model, and output and accept the obstacle detection results.

[0025] According to an embodiment of the present invention, a laser radar-based obstacle detection method is provided, in which radar point cloud data of a target block is input into a deep learning model, an obstacle detection result is output, and when the confidence of the pre-acquired deep learning model corresponding to the target block meets a set condition, the obstacle detection result is adopted, including:

[0026] Input the radar point cloud data of the target area into different deep learning models respectively, and output the corresponding multiple obstacle detection results;

[0027] Based on the confidence of the target block corresponding to the deep learning model obtained in advance, the weight coefficients of the multiple obstacle detection results corresponding to different deep learning models are determined, and the weight coefficients of the multiple obstacle detection results are used to perform weighted calculation on the multiple obstacle detection results to obtain and adopt the weighted calculation result.

[0028] An embodiment of the present invention further provides an obstacle detection device based on laser radar, the device comprising:

[0029] The first scanning module uses a laser radar to scan a target area at a distance from the vehicle to obtain radar point cloud data;

[0030] The obstacle detection module inputs the radar point cloud data into the deep learning model, outputs the model detection results of the obstacles in the target block, and obtains the error detection results and correct detection results of the positions of each radar point in the radar point cloud data in the model detection results;

[0031] The confidence calculation module calculates the confidence of the deep learning model corresponding to the target block based on the error detection results and the correct detection results.

[0032] An embodiment of the present invention further provides an obstacle detection device based on laser radar, the device comprising:

[0033] The second scanning module uses a laser radar to scan a target block in the driving environment that is far from the vehicle target, and obtains radar point cloud data of the target block;

[0034] The acceptance module inputs the radar point cloud data of the target block into the deep learning model, outputs the obstacle detection result, and accepts the obstacle detection result when the confidence of the pre-acquired deep learning model corresponding to the target block meets the set conditions.

[0035] An embodiment of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of any of the above-mentioned laser radar-based obstacle detection methods are implemented.

[0036] An embodiment of the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of any of the above-mentioned laser radar-based obstacle detection methods are implemented.

[0037] An embodiment of the present invention further provides a computer program product, including a computer program, which implements the steps of any of the above-mentioned laser radar-based obstacle detection methods when executed by a processor.

[0038] The laser radar-based obstacle detection method, device, equipment and storage medium provided by the embodiment of the present invention use a laser radar to scan the target block at the target distance of the self-vehicle to obtain radar point cloud data, input the radar point cloud data into the deep learning model, output the model detection result of the obstacle in the target block, and obtain the error detection result and correct detection result of each radar point position in the radar point cloud data in the model detection result. In this case, the confidence of the deep learning model corresponding to the target block can be calculated based on the error detection result and the correct detection result. Using the technical solution of the embodiment of the present invention, the confidence of the deep learning model for the target block at each target distance around the self-vehicle can be automatically and efficiently counted, and the confidence of different deep learning models corresponding to different target blocks can be realized, solving the problem of low efficiency of manual statistics.

[0039] Furthermore, during the actual driving process, according to the confidence of the target block corresponding to the pre-acquired deep learning model, a target deep learning model with a confidence that meets the set conditions is selected to perform obstacle detection on the radar point cloud data obtained by the lidar scanning of the target block to obtain an obstacle detection result with high confidence and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 This is one of the flow charts of the obstacle detection method based on laser radar provided in an embodiment of the present invention;

[0042] Figure 2This is a second flow chart of the obstacle detection method based on laser radar provided in an embodiment of the present invention;

[0043] Figure 3 This is a third flow chart of the obstacle detection method based on laser radar provided in an embodiment of the present invention;

[0044] Figure 4 This is a fourth flow chart of the obstacle detection method based on laser radar provided in an embodiment of the present invention;

[0045] Figure 5-Figure 7 is a schematic diagram of a laser radar-based obstacle detection method provided by an embodiment of the present invention;

[0046] Figure 8 This is a fifth flow chart of the obstacle detection method based on laser radar provided in an embodiment of the present invention;

[0047] Fig. 9 This is the sixth flow chart of the obstacle detection method based on laser radar provided in an embodiment of the present invention;

[0048] Fig.10 FIG7 is a flowchart of a laser radar-based obstacle detection method according to an embodiment of the present invention;

[0049] Fig.11 It is one of the structural schematic diagrams of the obstacle detection device based on laser radar provided in an embodiment of the present invention;

[0050] Fig.12 This is a second structural schematic diagram of the obstacle detection device based on laser radar provided in an embodiment of the present invention;

[0051] Fig.13 This is a third structural schematic diagram of the obstacle detection device based on laser radar provided in an embodiment of the present invention;

[0052] Fig.14 This is a fourth structural schematic diagram of the obstacle detection device based on laser radar provided in an embodiment of the present invention;

[0053] Fig.15 This is a fifth structural diagram of the obstacle detection device based on laser radar provided in an embodiment of the present invention;

[0054] Fig.16 This is the sixth structural schematic diagram of the obstacle detection device based on laser radar provided in an embodiment of the present invention;

[0055] Fig.17 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] In practice, it is found that when performing lidar obstacle detection, the deep learning models used are not the same, and there are differences in the obstacle detection results of each deep learning model.

[0058] It is further found that the performance of each deep learning model varies with distances. Some models have high detection accuracy for distant targets, while some models have high detection accuracy for close targets.

[0059] The embodiments of the present invention provide a laser radar-based obstacle detection method, device, equipment and storage medium, which uses a laser radar to scan a target block at a distance from a self-vehicle target to obtain radar point cloud data, inputs the radar point cloud data into a deep learning model, outputs a model detection result of obstacles in the target block, and obtains error detection results and correct detection results of each radar point position in the radar point cloud data in the model detection result.

[0060] Among them, the false detection result is the result of the wrong detection of the radar point position, which includes the radar point that does not belong to the obstacle being mistakenly detected as belonging to the obstacle.

[0061] In this case, the confidence of the deep learning model corresponding to the target block can be calculated based on the false detection results and the correct detection results.

[0062] By using the technical solution of the embodiment of the present invention, it is possible to automatically and efficiently count the confidence of the target blocks at various target distances around the vehicle using the deep learning model, and it is possible to realize the confidence of different target blocks corresponding to different deep learning models, thereby solving the problem of low efficiency of manual statistics.

[0063] Furthermore, during the actual driving process, according to the confidence of the target block corresponding to the pre-acquired deep learning model, a target deep learning model with a confidence that meets the set conditions is selected to perform obstacle detection on the radar point cloud data obtained by the lidar scanning of the target block to obtain an obstacle detection result with high confidence and accuracy.

[0064] Combine the following Figure 1-Figure 4 A flowchart describing a laser radar-based obstacle detection method according to an embodiment of the present invention is provided. The method is performed by an obstacle detection system, or a driving computer or a vehicle-mounted system equipped with the obstacle detection system.

[0065] refer to Figure 1 The laser radar-based obstacle detection method provided by the embodiment of the present invention comprises the following steps:

[0066] Step 110: Use a laser radar to scan a target area at a distance from the vehicle to obtain radar point cloud data;

[0067] Step 120: inputting radar point cloud data into the deep learning model, outputting the model detection result of obstacles in the target block, and obtaining the error detection result and the correct detection result of each radar point position in the radar point cloud data in the model detection result;

[0068] Step 130: Calculate the confidence of the deep learning model corresponding to the target block according to the error detection result and the correct detection result.

[0069] By using the technical solution of the embodiment of the present invention, it is possible to automatically and efficiently count the confidence of the deep learning models for each target block around the vehicle, thereby solving the problem of low efficiency of manual statistics.

[0070] In an embodiment of the present invention, the area around the vehicle can be divided into multiple blocks, wherein the distances from each block to the vehicle are sometimes the same and sometimes different, thereby enabling automatic detection of the confidence of the deep learning model for each block.

[0071] Optionally, a laser radar is used to scan a target area at a target distance from the vehicle to obtain radar point cloud data, including:

[0072] For each target block in multiple target blocks with different target distances or the same target distance from the vehicle, use the lidar to scan the target block at the target distance from the vehicle to obtain radar point cloud data until the confidence of the deep learning model corresponding to each target block is obtained.

[0073] In an optional embodiment of the present invention, reference Figure 2 , the obstacle detection method based on laser radar includes the following steps:

[0074] Step 210: Use a laser radar to scan a target area at a target distance from the vehicle to obtain radar point cloud data;

[0075] Step 220: inputting radar point cloud data into the deep learning model, outputting the model detection result of obstacles in the target block, and obtaining the error detection result and the correct detection result of each radar point position in the radar point cloud data in the model detection result;

[0076] Step 230: Calculate the confidence of the deep learning model corresponding to the target block according to the error detection result and the correct detection result;

[0077] For each target block in multiple target blocks with different target distances or the same target distance from the vehicle, use the lidar to scan the target block with the target distance from the vehicle to obtain the radar point cloud data until the confidence of the deep learning model corresponding to each target block is obtained;

[0078] Step 240: When the difference between the confidences corresponding to adjacent blocks in multiple target blocks does not exceed the target value, the adjacent blocks are merged to obtain a merged target block, and the error detection results and correct detection results of the adjacent blocks are merged to calculate the confidence of the merged target block corresponding to the deep learning model.

[0079] Using the method of this embodiment, adjacent blocks with similar confidences are merged according to the breadth-first traversal algorithm, and the confidences of the merged blocks are recalculated, which can expand the range of the blocks and improve the detection accuracy of larger-sized targets.

[0080] Optionally, the target value is preset, and the difference is an absolute value of the difference, such as 0.05 or other values, which are not specifically limited here.

[0081] In an optional embodiment of the present invention, the confidence of the deep learning model corresponding to the target block is calculated according to the error detection result and the correct detection result, including:

[0082] The larger quartile of the confidence of the wrong detection result is obtained, the smaller quartile of the confidence of the correct detection result is obtained, and the average of the larger quartile and the smaller quartile is calculated as the confidence of the target block corresponding to the deep learning model.

[0083] Among them, quartile (English: Quartile) is a type of quantile in statistics, that is, all values ​​are arranged from small to large and divided into four equal parts, and the values ​​at the three dividing points are quartiles.

[0084] Among them, the smaller quartile is also called the first quartile (Q1), which is equal to the 25% number after all the values ​​in the sample are arranged from small to large, and the larger quartile is also called the third quartile (Q3), also known as the larger quartile, which is equal to the 75% number after all the values ​​in the sample are arranged from small to large, wherein the sample corresponds to the confidence of each radar point in the correct detection result of this embodiment.

[0085] In an optional embodiment of the present invention, when the confidence of the deep learning model corresponding to multiple target blocks is obtained, a confidence distribution map is drawn according to the confidence of the deep learning model corresponding to the multiple target blocks.

[0086] In this embodiment, the confidence distribution map is highly visualized and hierarchical, and can be applied to the actual lidar obstacle detection process. The confidence of the target block corresponding to each deep learning model is found through the confidence distribution map, so that the target deep learning model with relatively high confidence can be selected for each block.

[0087] Reference Figure 3 In the optional embodiment shown, the obstacle detection method based on laser radar includes the following steps:

[0088] Step 310: Use a laser radar to scan a target area at a target distance from the vehicle to obtain radar point cloud data;

[0089] Step 320: inputting radar point cloud data into the deep learning model, outputting the model detection result of obstacles in the target block, and obtaining the error detection result and the correct detection result of each radar point position in the radar point cloud data in the model detection result;

[0090] Step 330: Calculate the confidence of the deep learning model corresponding to the target block according to the error detection result and the correct detection result;

[0091] Step 340: When the confidences of different deep learning models corresponding to the target block are obtained, a target deep learning model whose confidence meets the set conditions is selected from the different deep learning models, and a corresponding relationship between the target deep learning model and the target block is established.

[0092] In this way, a target deep learning model with high confidence can be pre-built for the specific target block, and then the obstacle detection results of the target deep learning model can be directly used during the application process, and the obstacle detection results have high credibility.

[0093] In another embodiment of the present invention, the weight coefficient can also be determined according to the confidence of the target blocks corresponding to multiple deep learning models. In actual application, each obstacle detection result is weightedly calculated according to the weight coefficient to obtain a weighted result, which is used as the final obstacle detection result.

[0094] refer to Figure 4 In an optional embodiment of the present invention, the obstacle detection method based on laser radar includes the following steps:

[0095] Step 410: Obtaining a model detection result of obstacles in the front block using a deep learning model;

[0096] Step 420: determine whether the detection position of each radar point in the model detection result is the same as the actual position of the actual obstacle, and obtain a correct detection result and an incorrect detection result;

[0097] Step 430: Count the smaller quartiles of confidence in all correct detection results;

[0098] Step 440: Count the larger quartiles of confidence in all the error detection results;

[0099] Step 450: Take the mean of the two quartiles as the target confidence of the deep learning model for the current block.

[0100] Combine the following Figure 5-Figure 7 The principle of confidence calculation in the embodiment of the present invention is described.

[0101] refer to Figure 5 , cut the vehicle detection range into multiple blocks;

[0102] refer to Figure 6 , count the correct detection results and wrong detection results in the model detection results in each block, and calculate the target confidence of each block. The specific values ​​are shown in the figure, which is only an example;

[0103] refer to Figure 7 , according to the breadth-first traversal algorithm, adjacent blocks with similar target confidences are merged, and the confidences of the merged blocks are recalculated. Finally, an easy-to-use confidence distribution diagram with a relatively rough division and significant separation of confidences between blocks is obtained, such as 0.11, 0.205 and 0.3, which are all the confidences of the merged blocks.

[0104] Figure 8 The flowchart of the obstacle detection method based on laser radar provided in the embodiment of the present invention may include the following steps:

[0105] Step 810: Use a laser radar to scan a target block in the driving environment that is away from the vehicle target, and obtain radar point cloud data of the target block;

[0106] Step 820: Input the radar point cloud data of the target block into the deep learning model, output the obstacle detection result, and adopt the obstacle detection result when the confidence of the target block corresponding to the pre-acquired deep learning model meets the set conditions.

[0107] In addition, in the case where the confidence level of the corresponding target block of the corresponding deep learning model does not meet the set conditions, the obstacle detection result is not accepted, and a decision opposite to the obstacle detection result is made.

[0108] Using the laser radar-based obstacle detection method of the embodiment of the present invention, it is possible to determine whether to accept the obstacle detection result according to the confidence of the deep learning model for the target block, wherein the confidence of the deep learning model corresponding to the target block is calculated using Figure 1-Figure 7The method of any embodiment is obtained, so that the obstacle detection result of this embodiment has high accuracy, so that driving decisions can be made according to the obstacle detection result.

[0109] In an optional embodiment of the present invention, reference Fig. 9 , the obstacle detection method based on laser radar may include the following steps:

[0110] Step 910: Use a laser radar to scan a target block in the driving environment that is away from the vehicle target, and obtain radar point cloud data of the target block;

[0111] Step 920: Based on the pre-obtained confidence of the deep learning model corresponding to the target block, a target deep learning model whose confidence meets the set conditions is selected from multiple candidate deep learning models;

[0112] Step 930: Input the radar point cloud data of the target block into the target deep learning model, and output and accept the obstacle detection result.

[0113] In this embodiment, a target deep learning model can be selected in advance based on the pre-obtained confidence level, and the obstacle detection result is deemed to have a high credibility, and then the obstacle detection result is adopted.

[0114] Optionally, the above setting condition may refer to the maximum confidence level among the confidence levels or any confidence level that is not less than a confidence level threshold.

[0115] In an optional embodiment of the present invention, reference Fig.10 , the obstacle detection method based on laser radar may include the following steps:

[0116] Step 1010: Use a laser radar to scan a target block in the driving environment that is away from the vehicle target, and obtain radar point cloud data of the target block;

[0117] Step 1020: inputting radar point cloud data of the target area into different deep learning models respectively, and outputting corresponding multiple obstacle detection results;

[0118] Step 1030: Determine the weight coefficients of multiple obstacle detection results corresponding to different deep learning models based on the confidence of the target block corresponding to the deep learning model obtained in advance, and use the weight coefficients of the multiple obstacle detection results to perform weighted calculation on the multiple obstacle detection results to obtain and adopt the weighted calculation result.

[0119] In this case, the weighted calculation result is used as the final obstacle detection result.

[0120] The following is a description of an obstacle detection device based on a laser radar provided in an embodiment of the present invention. The obstacle detection device based on a laser radar described below and the obstacle detection method based on a laser radar described above can refer to each other.

[0121] Reference Fig.11 , the obstacle detection device based on laser radar provided in an embodiment of the present invention may include:

[0122] The first scanning module 1110 uses a laser radar to scan a target block at a distance from the vehicle to obtain radar point cloud data;

[0123] The obstacle detection module 1120 inputs the radar point cloud data into the deep learning model, outputs the model detection result of the obstacles in the target block, and obtains the error detection result and the correct detection result of each radar point position in the radar point cloud data in the model detection result;

[0124] The confidence calculation module 1130 calculates the confidence of the deep learning model corresponding to the target block based on the error detection result and the correct detection result.

[0125] Using the laser radar-based obstacle detection device provided by the embodiment of the present invention, the laser radar is used to scan the target block at the target distance of the self-vehicle to obtain radar point cloud data, the radar point cloud data is input into the deep learning model, the model detection result of the obstacle in the target block is output, and the error detection result and the correct detection result of each radar point position in the radar point cloud data in the model detection result are obtained. In this case, the confidence of the deep learning model corresponding to the target block can be calculated based on the error detection result and the correct detection result.

[0126] By using the technical solution of the embodiment of the present invention, it is possible to automatically and efficiently count the confidence of the deep learning model for the target blocks at each target distance around the vehicle, and to realize the confidence of different deep learning models corresponding to different target blocks, thereby solving the problem of low efficiency of manual statistics. Furthermore, in the actual driving process, according to the confidence of the deep learning model corresponding to the target block obtained in advance, a target deep learning model with a confidence that meets the set conditions is selected to perform obstacle detection on the radar point cloud data obtained by scanning the target block by the laser radar, and obtain an obstacle detection result with high confidence and accuracy.

[0127] Optionally, the first scanning module 1110 is specifically used for:

[0128] For each target block in multiple target blocks with different target distances or the same target distance from the vehicle, use the lidar to scan the target block at the target distance from the vehicle to obtain radar point cloud data until the confidence of the deep learning model corresponding to each target block is obtained.

[0129] Optionally, the confidence calculation module 1130 is specifically used for:

[0130] The larger quartile of the confidence of the wrong detection result is obtained, the smaller quartile of the confidence of the correct detection result is obtained, and the average of the larger quartile and the smaller quartile is calculated as the confidence of the target block corresponding to the deep learning model.

[0131] Optionally, with Fig.11 compared to, Fig.12 The laser radar-based obstacle detection device also includes:

[0132] The merging calculation module 1210 merges the adjacent blocks in the multiple target blocks to obtain a merged target block, and merges the error detection results and correct detection results of the adjacent blocks to calculate the confidence of the merged target block corresponding to the deep learning model, when the difference between the confidences corresponding to the adjacent blocks in the multiple target blocks does not exceed the target value.

[0133] Optionally, with Fig.12 compared to, Fig.13 The laser radar-based obstacle detection device also includes:

[0134] The drawing module 1310 draws a confidence distribution map according to the confidence of the deep learning model corresponding to the multiple target blocks.

[0135] Optionally, with Fig.11 compared to, Fig.14 The laser radar-based obstacle detection device also includes:

[0136] The correspondence establishing module 1410, when obtaining the confidences of different deep learning models corresponding to the target block, selects a target deep learning model whose confidence meets the set conditions from the different deep learning models, and establishes the correspondence between the target deep learning model and the target block.

[0137] Reference Fig.15 , the obstacle detection device based on laser radar provided in an embodiment of the present invention may include:

[0138] The second scanning module 1510 uses a laser radar to scan a target block in the driving environment that is away from the vehicle target, and obtains radar point cloud data of the target block;

[0139] The acceptance module 1520 inputs the radar point cloud data of the target block into the deep learning model, outputs the obstacle detection result, and accepts the obstacle detection result when the confidence of the pre-acquired deep learning model corresponding to the target block meets the set conditions.

[0140] Using the laser radar-based obstacle detection method of the embodiment of the present invention, it is possible to determine whether to accept the obstacle detection result according to the confidence of the deep learning model for the target block, wherein the confidence of the deep learning model corresponding to the target block is calculated using Figure 1-Figure 7 The method of any embodiment is obtained, so that the obstacle detection result of this embodiment has high accuracy, so that driving decisions can be made according to the obstacle detection result.

[0141] Optionally, the acceptance module 1520 is specifically used for:

[0142] Input the radar point cloud data of the target area into different deep learning models respectively, and output the corresponding multiple obstacle detection results;

[0143] Based on the confidence of the target block corresponding to the deep learning model obtained in advance, the weight coefficients of the multiple obstacle detection results corresponding to different deep learning models are determined, and the weight coefficients of the multiple obstacle detection results are used to perform weighted calculation on the multiple obstacle detection results to obtain and adopt the weighted calculation result.

[0144] Optionally, with Fig.15 compared to, Fig.16 The laser radar-based obstacle detection device also includes:

[0145] A selection module 1610 selects a target deep learning model whose confidence meets a set condition from among multiple candidate deep learning models based on the confidence of the deep learning model corresponding to the target block obtained in advance;

[0146] The acceptance module 1620 is specifically used to input the radar point cloud data of the target block into the target deep learning model, and output and accept the obstacle detection result.

[0147] Fig.17 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.17 As shown, the electronic device may include: a processor 1710, a communication interface 1720, a memory 1730 and a communication bus 1740, wherein the processor 1710, the communication interface 1720 and the memory 1730 communicate with each other through the communication bus 1740. The processor 1710 may call the logic instructions in the memory 1730 to execute the obstacle detection method based on the laser radar, and the method includes:

[0148] Use the laser radar to scan the target area at a distance from the vehicle to obtain radar point cloud data;

[0149] Input radar point cloud data into the deep learning model, output the model detection results of obstacles in the target block, and obtain the error detection results and correct detection results of the positions of each radar point in the radar point cloud data in the model detection results;

[0150] Calculate the confidence of the deep learning model corresponding to the target block based on the error detection results and the correct detection results;

[0151] Alternatively, the method further comprises:

[0152] Use the laser radar to scan the target area at a distance from the vehicle to obtain radar point cloud data;

[0153] Input radar point cloud data into the deep learning model, output the model detection results of obstacles in the target block, and obtain the error detection results and correct detection results of the positions of each radar point in the radar point cloud data in the model detection results;

[0154] The confidence of the deep learning model corresponding to the target block is calculated based on the false detection results and the correct detection results.

[0155] In addition, the logic instructions in the above-mentioned memory 1730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0156] On the other hand, the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the laser radar-based obstacle detection method provided by the above methods, and the method includes:

[0157] Use the laser radar to scan the target area at a distance from the vehicle to obtain radar point cloud data;

[0158] Input radar point cloud data into the deep learning model, output the model detection results of obstacles in the target block, and obtain the error detection results and correct detection results of the positions of each radar point in the radar point cloud data in the model detection results;

[0159] Calculate the confidence of the deep learning model corresponding to the target block based on the error detection results and the correct detection results;

[0160] Alternatively, the method further comprises:

[0161] Use the laser radar to scan the target area at a distance from the vehicle to obtain radar point cloud data;

[0162] Input radar point cloud data into the deep learning model, output the model detection results of obstacles in the target block, and obtain the error detection results and correct detection results of the positions of each radar point in the radar point cloud data in the model detection results;

[0163] The confidence of the deep learning model corresponding to the target block is calculated based on the false detection results and the correct detection results.

[0164] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the laser radar-based obstacle detection method provided by the above methods, the method comprising:

[0165] Use the laser radar to scan the target area at a distance from the vehicle to obtain radar point cloud data;

[0166] Input radar point cloud data into the deep learning model, output the model detection results of obstacles in the target block, and obtain the error detection results and correct detection results of the positions of each radar point in the radar point cloud data in the model detection results;

[0167] Calculate the confidence of the deep learning model corresponding to the target block based on the error detection results and the correct detection results;

[0168] Alternatively, the method further comprises:

[0169] Use the laser radar to scan the target area at a distance from the vehicle to obtain radar point cloud data;

[0170] Input radar point cloud data into the deep learning model, output the model detection results of obstacles in the target block, and obtain the error detection results and correct detection results of the positions of each radar point in the radar point cloud data in the model detection results;

[0171] The confidence of the deep learning model corresponding to the target block is calculated based on the false detection results and the correct detection results.

[0172] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0173] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A laser radar-based obstacle detection method, characterized in that: include: Use the laser radar to scan the target area at a distance from the vehicle to obtain radar point cloud data; Input the radar point cloud data into the deep learning model, output the model detection result of the obstacles in the target block, and obtain the error detection result and the correct detection result of each radar point position in the radar point cloud data in the model detection result; The confidence of the deep learning model corresponding to the target block is calculated according to the error detection result and the correct detection result. The confidence calculation steps are as follows: Divide the vehicle detection range into multiple blocks; Obtain the correct detection results and the wrong detection results in the model detection results in each block, and calculate the target confidence of each block; According to the breadth-first traversal algorithm, adjacent blocks with similar target confidence are merged, and the confidence of the merged blocks is recalculated; Also includes: When the confidences of different deep learning models corresponding to the target block are obtained, a target deep learning model whose confidence meets the set conditions is selected from the different deep learning models, and a corresponding relationship between the target deep learning model and the target block is established.

2. The obstacle detection method based on laser radar according to claim 1, characterized in that: The laser radar is used to scan the target block at a distance from the vehicle to obtain radar point cloud data, including: For each target block in multiple target blocks with different target distances or the same target distance from the vehicle, a laser radar is used to scan the target block at the target distance from the vehicle to obtain radar point cloud data until the confidence of the deep learning model corresponding to each target block is obtained.

3. The obstacle detection method based on laser radar according to claim 2, characterized in that: Also includes: When the difference between the confidences corresponding to adjacent blocks among the multiple target blocks does not exceed the target value, the adjacent blocks are merged to obtain a merged target block, and the error detection results and correct detection results of the adjacent blocks are merged to calculate the confidence of the deep learning model corresponding to the merged target block.

4. The obstacle detection method based on laser radar according to claim 3, characterized in that: Also includes: A confidence distribution map is drawn according to the confidence of the deep learning model corresponding to the multiple target blocks.

5. The obstacle detection method based on laser radar according to claim 1, characterized in that: The calculating the confidence of the deep learning model corresponding to the target block according to the error detection result and the correct detection result includes: Obtain the larger quartile of the confidence of the false detection result, obtain the smaller quartile of the confidence of the correct detection result, and calculate the average of the larger quartile and the smaller quartile as the confidence of the deep learning model corresponding to the target block.

6. A laser radar-based obstacle detection method, characterized in that: include: Use a laser radar to scan a target block in the driving environment that is away from the vehicle target, and obtain radar point cloud data of the target block; The radar point cloud data of the target block is input into the deep learning model, and the obstacle detection result is output. When the confidence of the deep learning model corresponding to the target block obtained in advance meets the set conditions, the obstacle detection result is adopted. The confidence calculation steps are as follows: Divide the vehicle detection range into multiple blocks; Obtain the correct detection results and the wrong detection results in the model detection results in each block, and calculate the target confidence of each block; According to the breadth-first traversal algorithm, adjacent blocks with similar target confidence are merged, and the confidence of the merged blocks is recalculated; Among them, when the confidence levels of different deep learning models corresponding to the target block are obtained, a target deep learning model whose confidence level meets the set conditions is selected from the different deep learning models, and a corresponding relationship between the target deep learning model and the target block is established.

7. The obstacle detection method based on laser radar according to claim 6, characterized in that: Before inputting the radar point cloud data of the target block into the deep learning model and outputting the obstacle detection result, the method further includes: Based on the confidence of the pre-obtained deep learning model corresponding to the target block, a target deep learning model whose confidence meets the set conditions is selected from multiple candidate deep learning models; The inputting of the radar point cloud data of the target block into the deep learning model, outputting the obstacle detection result, and adopting the obstacle detection result when the confidence of the deep learning model corresponding to the target block obtained in advance meets the set condition, includes: The radar point cloud data of the target block is input into the target deep learning model, and the obstacle detection result is output and accepted.

8. The obstacle detection method based on laser radar according to claim 6, characterized in that: The inputting of the radar point cloud data of the target block into the deep learning model, outputting the obstacle detection result, and adopting the obstacle detection result when the confidence of the deep learning model corresponding to the target block obtained in advance meets the set condition, includes: Inputting the radar point cloud data of the target block into different deep learning models respectively, and outputting corresponding multiple obstacle detection results; Based on the pre-obtained confidence of the deep learning model corresponding to the target block, the weight coefficients of the multiple obstacle detection results respectively corresponding to the different deep learning models are determined, and the weight coefficients of the multiple obstacle detection results are used to perform weighted calculation on the multiple obstacle detection results to obtain and adopt the weighted calculation result.

9. An obstacle detection device based on laser radar, characterized in that: include: The first scanning module uses a laser radar to scan a target block at a distance from the vehicle to obtain radar point cloud data; An obstacle detection module, which inputs the radar point cloud data into a deep learning model, outputs a model detection result of obstacles in the target block, and obtains an error detection result and a correct detection result of each radar point position in the radar point cloud data in the model detection result; The confidence calculation module calculates the confidence of the deep learning model corresponding to the target block according to the error detection result and the correct detection result. The confidence calculation steps are as follows: Divide the vehicle detection range into multiple blocks; Obtain the correct detection results and the wrong detection results in the model detection results in each block, and calculate the target confidence of each block; According to the breadth-first traversal algorithm, adjacent blocks with similar target confidence are merged, and the confidence of the merged blocks is recalculated; Also includes: When the confidences of different deep learning models corresponding to the target block are obtained, a target deep learning model whose confidence meets the set conditions is selected from the different deep learning models, and a corresponding relationship between the target deep learning model and the target block is established.

10. An obstacle detection device based on laser radar, characterized in that: include: The second scanning module uses a laser radar to scan a target block in the driving environment that is away from the vehicle target, and obtains radar point cloud data of the target block; The confidence module inputs the radar point cloud data of the target block into the deep learning model, outputs the obstacle detection result, and when the confidence of the deep learning model corresponding to the target block obtained in advance meets the set conditions, the obstacle detection result is accepted. The confidence calculation steps are as follows: Divide the vehicle detection range into multiple blocks; Obtain the correct detection results and the wrong detection results in the model detection results in each block, and calculate the target confidence of each block; According to the breadth-first traversal algorithm, adjacent blocks with similar target confidence are merged, and the confidence of the merged blocks is recalculated; Among them, when the confidence levels of different deep learning models corresponding to the target block are obtained, a target deep learning model whose confidence level meets the set conditions is selected from the different deep learning models, and a corresponding relationship between the target deep learning model and the target block is established.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the laser radar-based obstacle detection method as described in any one of claims 1 to 8 are implemented.

12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the laser radar-based obstacle detection method according to any one of claims 1 to 8 are implemented.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the laser radar-based obstacle detection method according to any one of claims 1 to 8 are implemented.

Images