Segmentation result evaluation method, device, electronic device and readable storage medium

Abstract

Embodiments of the present invention provide a method, apparatus, electronic device, and readable storage medium for evaluating segmentation results, relating to the technical field of data processing. The method includes: obtaining point cloud data of a target area, segmenting the point cloud data to obtain a point cloud segmentation result, determining boundary information of the target area based on the segmentation result of DOM data corresponding to the target area and the point cloud segmentation result, and then determining a segmentation evaluation parameter of the point cloud data based on the boundary information and the point cloud segmentation result, thereby realizing intelligent evaluation of the segmentation result.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and more particularly, to a method, apparatus, electronic device, and readable storage medium for evaluating segmentation results. Background Art

[0002] With the continuous development of digital devices, the amount of data that can be collected is also increasing continuously. For example, three-dimensional data that can represent spatial information plays a crucial role for intelligent operation devices. Exemplarily, the three-dimensional point cloud data is input into a segmentation model to obtain a segmentation result, and obstacles are detected based on the segmentation result, so as to effectively guide the flight strategy of operation devices such as drones. However, the accuracy of the segmentation result will directly affect the accuracy of obstacle recognition, and thus affect the reliability of the flight strategy. In the case of inaccurate segmentation results, it will misguide the flight route of the drone, and even cause a risk of collision. Therefore, it is important to evaluate the segmentation result. Summary of the Invention

[0003] One of the objectives of the present invention includes, for example, providing a method, apparatus, electronic device, and readable storage medium for evaluating segmentation results to automatically evaluate the segmentation results.

[0004] Embodiments of the present invention may be implemented as follows:

[0005] In a first aspect, an embodiment of the present invention provides a method for evaluating segmentation results, including:

[0006] Obtain the point cloud data of the target area, and segment the point cloud data to obtain a point cloud segmentation result;

[0007] Based on the segmentation result of the DOM data corresponding to the target area and the point cloud segmentation result, determine the boundary information of the target area;

[0008] Based on the boundary information and the point cloud segmentation result, determine the segmentation evaluation parameters of the point cloud data.

[0009] In an optional implementation, the step of determining the boundary information of the target area based on the segmentation result of the DOM data corresponding to the target area and the point cloud segmentation result includes:

[0010] Based on the segmentation result of each frame of DOM data corresponding to the target area, obtain the initial boundary information corresponding to the target area based on the boundary generation algorithm;

[0011] Obtain the first point cloud data located within each of the initial boundary information, and determine the category information of the first point cloud data based on the point cloud segmentation result;

[0012] Based on the category information and the quantity information of point clouds of different categories, select the initial boundary information that meets the preset conditions from the initial boundary information as the boundary information of the target area.

[0013] In an alternative embodiment, the category information includes a ground category and an obstacle category; the step of selecting the initial boundary information that meets the preset conditions from the initial boundary information based on the category information and the quantity information of point clouds of different categories as the boundary information of the target area includes:

[0014] Determine whether the initial boundary information meets the preset conditions based on the ratio of the quantity of point clouds of the ground category to the quantity of point clouds in the first point cloud data;

[0015] When the ratio is greater than the preset threshold, determine that the initial boundary information meets the preset conditions and use the initial boundary information as the boundary information of the target area.

[0016] In an alternative embodiment, the point cloud segmentation result includes the confidence levels of the categories corresponding to each point cloud in the point cloud data being the obstacle category and the ground category;

[0017] The step of determining the segmentation evaluation parameter of the point cloud data based on the boundary information and the point cloud segmentation result includes:

[0018] Determine the category-fuzzy points from the point clouds of the boundary information based on the confidence level of each point cloud;

[0019] Obtain the ratio of the quantity of the category-fuzzy points to the quantity of the point cloud data, and / or obtain the point cloud set inside each boundary information and the point cloud set outside each boundary information to determine the segmentation evaluation parameter of the point cloud data.

[0020] In an alternative embodiment, the category-fuzzy points include the point clouds of the fuzzy obstacle category, and the point clouds of the fuzzy obstacle category are the point clouds with confidence levels within a preset range among the point clouds corresponding to the obstacle category;

[0021] The step of obtaining the ratio of the quantity of the category-fuzzy points to the quantity of the point cloud data to determine the segmentation evaluation parameter of the point cloud data includes:

[0022] Obtain the ratio of the quantity of the point clouds of the fuzzy obstacle category to the quantity of the point clouds corresponding to the obstacle category in the point cloud data to determine the false detection rate of the obstacles.

[0023] In an alternative embodiment, the step of obtaining the point cloud set inside each boundary information and the point cloud set outside each boundary information to determine the segmentation evaluation parameter of the point cloud data includes:

[0024] Enlarge each piece of boundary information by a set multiple to obtain target information;

[0025] Based on the confidence level, for each piece of boundary information, determine the corresponding point cloud set g1 with the category of fuzzy ground from the point cloud set within the boundary information; determine the corresponding point cloud set g2 with the category of fuzzy ground from the area between the boundary information and the target information; determine the corresponding point cloud set g3 with the category of fuzzy ground from the area outside the target information;

[0026] Analyze and obtain the missed detection rate based on the point cloud set g1, the point cloud set g2, and the point cloud set g3;

[0027] Wherein, the point cloud of the fuzzy ground category is the point cloud with the confidence level within a preset range among the point clouds corresponding to the ground category.

[0028] In an optional implementation manner, the step of analyzing and obtaining the missed detection rate based on the point cloud set g1, the point cloud set g2, and the point cloud set g3 includes:

[0029] Determine the center point of each piece of boundary information;

[0030] For each piece of boundary information, based on the center point of the boundary information, as well as the corresponding point cloud set g1, the point cloud set g2, and the point cloud set g3, determine the missed detection rate of the boundary information;

[0031] Calculate the overall missed detection rate of all the boundary information according to the missed detection rate of each boundary information;

[0032] Analyze and obtain the effective missed detection rate according to the overall missed detection rate and the number of point clouds corresponding to the obstacle category in the point cloud data.

[0033] In an optional implementation manner, the confidence levels corresponding to the obstacle category and the ground category of each point cloud in the point cloud data are obtained based on binary classification segmentation, and the sum of the confidence level corresponding to the obstacle category and the confidence level corresponding to the ground category of each point cloud is 1; the category to which each point cloud belongs is the category with a confidence level greater than 0.5.

[0034] In an optional implementation manner, the step of obtaining the point cloud data of the target area includes:

[0035] Process the DOM data corresponding to the target area based on feature matching analysis and a three-dimensional reconstruction algorithm to obtain the point cloud data of the target area;

[0036] The segmentation result of the DOM data corresponding to the target area is obtained by performing semantic segmentation on the DOM data based on image segmentation technology.

[0037] In a second aspect, an embodiment of the present invention provides a segmentation result evaluation device, including:

[0038] A data acquisition module, configured to acquire point cloud data of a target area, and segment the point cloud data to obtain a point cloud segmentation result;

[0039] A data analysis module, configured to determine boundary information of the target area based on the segmentation result of the DOM data corresponding to the target area and the point cloud segmentation result; and determine a segmentation evaluation parameter of the point cloud data based on the boundary information and the point cloud segmentation result.

[0040] In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the program, it implements the segmentation result evaluation method according to any one of the foregoing embodiments.

[0041] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, where the computer-readable storage medium includes a computer program, and when the computer program runs, it controls an electronic device where the computer-readable storage medium is located to execute the segmentation result evaluation method according to any one of the foregoing embodiments.

[0042] The beneficial effects of the embodiments of the present invention include, for example: through the automated evaluation of the segmentation result, the manpower and time required for manual review are reduced, the implementation is relatively convenient, the efficiency is high, and the automated evaluation avoids the influence of personal subjective factors, thereby improving the evaluation reliability, and further ensuring the accuracy of obstacle recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 Shows a schematic diagram of an application scenario provided by an embodiment of the present invention.

[0045] Figure 2 Shows a schematic flowchart of a segmentation result evaluation method provided by an embodiment of the present invention.

[0046] Figure 3 Shows an example diagram of an image in a DOM data provided by an embodiment of the present invention.

[0047] Figure 4 Shows an example diagram of point cloud data provided by an embodiment of the present invention.

[0048] Figure 5 Shows Figure 3 An example diagram of the semantic segmentation result of the shown image.

[0049] Figure 6 Shows Figure 4 An example diagram of the segmentation result of the shown point cloud data.

[0050] Figure 7 Shows an example diagram of an overall segmentation result provided by an embodiment of the present invention.

[0051] Figure 8 Shows an example diagram of a plurality of initial boundary information provided by an embodiment of the present invention.

[0052] Figure 9 Shows an example diagram of each boundary information provided by an embodiment of the present invention.

[0053] Figure 10 Shows an example diagram of point cloud recognition with the corresponding category being the fuzzy obstacle category provided by an embodiment of the present invention.

[0054] Figure 11 Shows an example diagram of point cloud recognition with the corresponding category being the fuzzy ground category provided by an embodiment of the present invention.

[0055] Figure 12 Shows an example diagram of target information provided by an embodiment of the present invention.

[0056] Figure 13 Shows an exemplary structural block diagram of a segmentation result evaluation device provided by an embodiment of the present invention.

[0057] Icons: 100 - Electronic device; 110 - Memory; 120 - Processor; 130 - Communication module; 140 - Segmentation result evaluation device; 141 - Data acquisition module; 142 - Data analysis module. Detailed implementation manners

[0058] Nowadays, with the continuous development of digital devices, the amount of data that can be collected is also increasing continuously. For example, three - dimensional point cloud data can be collected through sensors, the three - dimensional point cloud data is input into a segmentation model to obtain a segmentation result, and obstacles are detected based on the segmentation result. However, when the accuracy of the segmentation result is poor, it will directly affect the accuracy of obstacle recognition.

[0059] After research, the current method mainly uses manual review to evaluate the results of obstacle recognition. Manual review requires a large amount of manpower and time, is inconvenient to implement, and is subject to many personal subjective factors, resulting in the need to improve the reliability of the evaluation and the low efficiency.

[0060] Based on the above research, the embodiments of the present invention provide a solution capable of automatically evaluating the segmentation results. Through automatic evaluation, the reliability and convenience of the segmentation result evaluation are improved, thereby ensuring the accuracy of obstacle recognition.

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention described and illustrated herein generally may be arranged and designed in a variety of different configurations.

[0062] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0063] It should be noted that the term "comprising", "including" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0064] It should be noted that like reference numerals and letters denote like items in the following figures, and thus, once an item is defined in one figure, it does not require further definition and explanation in subsequent figures.

[0065] It should be noted that, without conflict, the features in the embodiments of the present invention may be combined with each other.

[0066] Please refer to Figure 1, which is a block diagram of an electronic device 100 provided in this embodiment. The electronic device 100 in this embodiment can be a server, a processing device, a processing platform, etc. that can perform data interaction and processing. The electronic device 100 includes a memory 110, a processor 120, and a communication module 130. Each element of the memory 110, the processor 120, and the communication module 130 is electrically connected directly or indirectly to each other to achieve data transmission or interaction. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines.

[0067] Among them, the memory 110 is used to store programs or data. The memory 110 can be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0068] The processor 120 is used to read / write the data or programs stored in the memory 110 and perform corresponding functions.

[0069] The communication module 130 is used to establish a communication connection between the electronic device 100 and other communication terminals through the network and is used to send and receive data through the network.

[0070] It should be understood that Figure 1 The structure shown is only a schematic diagram of the structure of the electronic device 100. The electronic device 100 may further include more or fewer components than those shown in Figure 1 or have a configuration different from that shown in Figure 1 . Figure 1 Each component shown in can be implemented by hardware, software, or a combination thereof.

[0071] Please refer to Figure 2 , which is a flowchart of a segmentation result evaluation method provided in an embodiment of the present invention and can be executed by Figure 1 the electronic device 100, for example, can be executed by the processor 120 in the electronic device 100. The segmentation result evaluation method includes S110, S120, and S130.

[0072] S110. Obtain the point cloud data of the target area, and segment the point cloud data to obtain a point cloud segmentation result.

[0073] S120. Based on the segmentation result of the DOM data corresponding to the target area and the point cloud segmentation result, determine the boundary information of the target area.

[0074] S130. Based on the boundary information and the point cloud segmentation result, determine the segmentation evaluation parameters of the point cloud data.

[0075] Through the above process, the automatic evaluation of the segmentation result is realized. Compared with manual review, it reduces the required manpower and time, is more convenient and efficient. Moreover, the automatic evaluation avoids the influence of personal subjective factors, thus improving the evaluation reliability and further ensuring the accuracy of obstacle recognition.

[0076] In S110, the point cloud data of the target area can be obtained based on the DOM data corresponding to the target area.

[0077] Among them, the DOM data corresponding to the target area can be obtained by shooting with a camera device. For example, if it is necessary to identify obstacles for a certain target area, multiple images of the target area can be taken by using an unmanned aerial vehicle (UAV) shooting method to obtain the DOM data of the target area for the flight to be evaluated.

[0078] Please refer to Figure 3 , which is one of the images taken in the case where the target area is a farmland in this embodiment. It can be understood that in the case of a large target area, in order to achieve the complete recognition of the target area, multiple images are taken, and correspondingly, the DOM data of the target area for the flight to be evaluated is multiple frames.

[0079] Based on the DOM data corresponding to the target area, the method for obtaining the point cloud data may include: processing the DOM data corresponding to the target area based on feature matching analysis and 3D reconstruction algorithm to obtain the point cloud data of the target area. Please refer to Figure 4 , which is an example diagram of a kind of point cloud data provided in this embodiment ( Figure 4 in which is the point cloud data obtained in the case where the target area is a farmland and one of the images in the DOM data is as shown in Figure 3 ).

[0080] Based on the DOM data of the target area, the method for obtaining the segmentation result may include: performing semantic segmentation on the DOM data based on image segmentation technology to obtain the segmentation result of each image in the DOM data. Please refer to Figure 5 , which is the Figure 3 segmentation result of the image shown.

[0081] In the case where there are multiple images of the captured target area, semantic segmentation is performed on each image respectively, so that the segmentation result of each image (the segmentation result of each frame of DOM data) can be obtained.

[0082] The method for segmenting point cloud data to obtain the point cloud segmentation result may include: using a point cloud semantic segmentation model to perform segmentation processing on the point cloud data to obtain the point cloud segmentation result. Among them, the point cloud segmentation result includes the confidence levels of the categories corresponding to each point cloud as the obstacle category and the ground category. Based on the confidence levels, the category of each point cloud can be analyzed. Please refer to Figure 6 , which shows Figure 4 an example diagram of the point cloud segmentation result of the shown point cloud data.

[0083] In one implementation, the confidence levels of the categories corresponding to each point cloud in the point cloud data as the obstacle category and the ground category can be obtained based on binary classification segmentation. In the case where the category corresponding to each point cloud is obtained based on binary classification segmentation, during the segmentation process, the confidence level of the category corresponding to each said point cloud as the obstacle category and the confidence level of the corresponding category as the ground category will be generated, and the sum of the confidence levels of the two categories is 1. Based on this, the category to which each said point cloud belongs can be the category with a higher confidence level, that is, the category with a confidence level greater than 0.5.

[0084] For example, if based on binary classification segmentation, the confidence level of a certain point cloud corresponding to the obstacle category is 0.7, and the confidence level of the corresponding category as the ground category is 0.3, since the confidence level of this point cloud belonging to the obstacle category is higher, therefore, the obstacle category is taken as the category to which this point cloud belongs.

[0085] In S130, the method for determining the boundary information of the target area based on the segmentation result of the DOM data corresponding to the target area and the point cloud segmentation result can be set flexibly. For example, based on the segmentation result of each frame of DOM data corresponding to the target area, the initial boundary information corresponding to the target area can be obtained based on the boundary generation algorithm. The first point cloud data located within each said initial boundary information is acquired, and the category information of the first point cloud data is determined based on the point cloud segmentation result. Based on the category information and the quantity information of the point clouds of different categories, the initial boundary information that meets the preset conditions is selected from the initial boundary information to be used as the boundary information of the target area.

[0086] To achieve the complete recognition of the target area, the segmentation results of each frame of DOM data corresponding to the target area can be stitched together to obtain the overall segmentation result. Please refer to Figure 7 , for stitching Figure 5The overall segmentation result is obtained by splicing the segmentation results of other images in the DOM data corresponding to the shown segmentation result and the target area. Furthermore, analyze the overall segmentation result to obtain the boundary of the overall segmentation result.

[0087] For example, image stitching technology can be used to stitch the segmentation results of multiple images in the DOM data to obtain the overall segmentation result of the target area, and the boundary of the overall segmentation result can be obtained through the opencv boundary generation algorithm.

[0088] In one implementation, in the case of obtaining the boundary of the overall segmentation result, based on the conversion relationship between the tile coordinates and the longitude and latitude coordinates, the boundary can be converted into multiple initial boundary information. Please refer to Figure 8 , which is an example diagram of the multiple initial boundary information converted from the boundary. As Figure 8 shown, each initial boundary information forms an enclosed area respectively.

[0089] Among the converted initial boundary information, the enclosed area may mainly be the operation object or the interference object. For the differences in different scenarios, different operation objects, etc., the initial boundary information that obviously does not meet the conditions can be excluded from the initial boundary information to obtain the initial boundary information that meets the preset conditions, which is used as the boundary information of the target area, and then subsequent recognition and evaluation processing can be carried out on the selected boundary information.

[0090] Among them, according to the differences in scenarios, operation objects, etc., the preset conditions can be set flexibly. For example, the preset conditions can be determined according to the growth situation and planting characteristics of different stages of the operation object. Another example is that according to the environment where the operation object is located, such as mountains, hills, plains, etc., the different characteristics of the interference objects can be used to determine the preset conditions respectively.

[0091] Exemplarily, in some scenarios, the operation object should account for a relatively high proportion in the initial boundary information that meets the conditions. Therefore, it can be determined whether the initial boundary information meets the preset conditions based on the ratio of the number of point clouds belonging to the ground category in the first point cloud data and the number of point clouds in the first point cloud data. When the ratio is greater than the preset threshold, it is determined that the initial boundary information meets the preset conditions, and the initial boundary information is used as the boundary information of the target area.

[0092] By obtaining the number of point clouds enclosed by each initial boundary information, based on the confidence level, determine the category of each point cloud, and determine the occupancy rate of the point clouds corresponding to the ground category among the point clouds enclosed by the initial boundary information. Find all the initial boundary information with the occupancy rate meeting the preset conditions from the initial boundary information to form the boundary information of the target area.

[0093] Taking the target area as farmland as an example, the operation object is the crops in the farmland, and each initial boundary information is each farmland boundary. The boundary information of the target area is obtained through the following method: calculate the number of point clouds p enclosed by each farmland boundary, and through the method of point cloud category screening, obtain the number of point clouds p1 with the category of ground. Then, the occupancy rate rg of the point clouds with the category of ground is p1 / p. Since in actual farmland, the proportion of obstacles in the farmland is relatively small, it has been verified that when rg>0.7, the situation where data such as houses are misdetected as farmland boundaries can be avoided. Therefore, the preset condition can be determined as rg>0.7. For each farmland boundary, determine whether rg>0.7 is satisfied. If so, it is considered a legal boundary. If not, it is considered an illegal boundary. Furthermore, all farmland boundaries that satisfy rg>0.7 are used as the boundary information of the target area, denoted as boundarys.

[0094] The above selection of the occupancy rate is only for example, and the occupancy rate can be set flexibly. This embodiment does not limit this.

[0095] Please refer to Figure 9 , for an example diagram of the boundary information that satisfies the preset condition determined from the multiple initial boundary information shown in Figure 8 . Since in this example, Figure 8 all the initial boundary information shown is detected on the farmland, the determined boundary information is the same as the initial boundary information in Figure 8 . In the case of determining the confidence level and the boundary information that satisfies the preset condition, various methods can be used to obtain the segmentation evaluation parameter in S130. For example, based on the confidence level of each point cloud, the category ambiguous points can be determined from the point clouds of the boundary information. Obtain the quantity ratio of the category ambiguous points to the point cloud data to determine the segmentation evaluation parameter of the point cloud data. For another example, the point cloud set inside each boundary information and the point cloud set outside each boundary information can be obtained to determine the segmentation evaluation parameter of the point cloud data. For another example, the quantity ratio of the category ambiguous points to the point cloud data can be obtained, and the point cloud set inside each boundary information and the point cloud set outside each boundary information can be obtained to determine the segmentation evaluation parameter of the point cloud data. This embodiment does not limit this.

[0096] The segmentation evaluation parameter in S130 can be selected flexibly. For example, it can be the false detection rate, the missed detection rate, etc. Based on the corresponding relationship between each preset segmentation evaluation parameter and each evaluation result, the evaluation result of the analyzed segmentation evaluation parameter can be determined. For example, in the case where the segmentation evaluation parameter includes the false detection rate and the missed detection rate, the evaluation results corresponding to the false detection rate and the missed detection rate of the obstacles analyzed based on the segmentation result can be determined according to the pre-stored corresponding relationship between each false detection rate, missed detection rate and each evaluation result.

[0097] Since the false detection rate is directly related to the recognition accuracy of the category to which the point cloud belongs, and the recognition of the category to which the point cloud belongs is mainly achieved based on confidence, and when the confidence is within the "critical range", it is the concentrated area leading to misrecognition. Therefore, in one implementation, the false detection rate can be analyzed through the following steps: Determine the category-fuzzy points from the point cloud data of the boundary information based on the confidence of each point cloud, obtain the proportion of the number of the category-fuzzy points and the point cloud data, and determine the segmentation evaluation parameter of the point cloud data based on the proportion.

[0098] In the case where the false detection rate of obstacles needs to be analyzed, the category-fuzzy points may include the point clouds of the fuzzy obstacle category, where the point clouds of the fuzzy obstacle category are the point clouds with confidence within a preset range among the point clouds corresponding to the obstacle category. Correspondingly, by obtaining the proportion of the number of the point clouds of the fuzzy obstacle category to the number of the point clouds corresponding to the obstacle category in the point cloud data, the false detection rate of the obstacles can be determined.

[0099] Wherein, when the confidence of each point cloud corresponding to the obstacle category and the ground category is obtained based on binary classification segmentation, since the sum of the confidence of each point cloud is 1, therefore, the point cloud with a confidence of about 0.5 can be used as the category-fuzzy point. Thus, a certain value can be floated around 0.5 to obtain the "critical range".

[0100] It can be understood that when using the point cloud semantic segmentation model to segment the point cloud data to obtain the confidence of each point cloud corresponding to the obstacle category and the ground category, the confidence may also change due to the stability of the point cloud semantic segmentation model itself. Therefore, the "critical range" can also be determined by the influence of the stability of the point cloud semantic segmentation model on the obtained confidence.

[0101] In this embodiment, the preset range is the above-mentioned "critical range" that may lead to misrecognition. In order to realize the recognition of fuzzy obstacles, the preset range can be set flexibly. For example, it can be 0.5 - 0.55, 0.5 - 0.6, 0.5 - 0.65, etc.

[0102] Taking the preset range of 0.5 - 0.6 as an example, correspondingly, the point cloud with the confidence corresponding to the obstacle category between 0.5 and 0.6 is the point cloud of the fuzzy obstacle category. Please refer to Figure 10 for Figure 4The point cloud data shown, a point cloud recognition example diagram of the detected point cloud corresponding to the category of fuzzy obstacle. Similarly, the point cloud of the fuzzy ground category is the point cloud with a confidence level within a preset range among the point clouds corresponding to the ground category. When the preset range is 0.5 - 0.6, the point cloud with a confidence level between 0.5 - 0.6 corresponding to the ground category is the point cloud of the fuzzy ground category. Refer to Figure 11 , for Figure 4 The point cloud data shown, a point cloud recognition example diagram of the detected point cloud corresponding to the category of fuzzy ground.

[0103] Calculate the number of point clouds of the fuzzy obstacle category within each boundary information of the target area, and add up the calculated numbers of all point clouds of the fuzzy obstacle category to obtain the number b of point clouds of the fuzzy obstacle category within each boundary information in the entire target area. Count the number obs of all point clouds corresponding to the obstacle category in the point cloud data of the target area. Calculate the ratio of b to obs to obtain the false detection rate false Rate for measuring the obstacle false detection situation.

[0104] Since missed detection will directly affect the reliability of the operation, and in actual operations, the operating area of the operating device is often not limited to within each boundary information of the target area, but will exceed a certain range. To ensure the detection reliability, the point cloud set within each boundary information and the point cloud set outside each boundary information can be obtained to determine the segmentation evaluation parameters of the point cloud data and obtain the missed detection rate of the obstacle.

[0105] When determining the missed detection rate, each boundary information of the target area can be expanded by a set multiple to obtain the target information.

[0106] For example, the method of proportional expansion of the polygon contour can be used to expand each boundary information in boundarys. By setting the proportional parameter to determine the expansion scale, each boundary information in boundarys is expanded to n times its original size. Obtain the set boundarys1 of target information. Exemplarily, when the proportional parameter is set to 1.05 - 1.2, each boundary information in boundarys is correspondingly expanded to 1.05 - 1.2 times its original size. It can be understood that the proportional parameter can also be other values, such as 1.04 - 1.3, 1.1 - 1.25, etc., and this embodiment will not elaborate on them one by one.

[0107] Please refer to Figure 12 , which shows the Figure 9 Example diagram of the target information obtained after expanding each boundary information in the shown boundarys.

[0108] Based on the confidence level, for each piece of boundary information, a point cloud set g1 corresponding to the fuzzy ground category is determined from the point cloud set within the boundary information (the point cloud set g1 of the fuzzy ground category within the boundarys). A point cloud set g2 corresponding to the fuzzy ground category is determined from the area between the boundary information and the target information (the point cloud set g2 of the edge fuzzy ground category between boundary1 and boundarys). A point cloud set g3 corresponding to the fuzzy ground category is determined from the area outside the target information (the point cloud set g3 of the fuzzy ground category outside boundarys1). The missed detection rate is analyzed based on the point cloud set g1, the point cloud set g2, and the point cloud set g3.

[0109] Exemplarily, the center point of each piece of boundary information is determined. For each piece of boundary information, based on the center point of the boundary information, and the corresponding point cloud set g1, point cloud set g2, and point cloud set g3, the missed detection rate of the boundary information is determined. According to the missed detection rate of each piece of boundary information, the overall missed detection rate of all the boundary information is calculated, and then based on the overall missed detection rate and the number of point clouds corresponding to the obstacle category in the point cloud data, the effective missed detection rate is analyzed.

[0110] Among them, the calculation method of the centroid of a polygon can be used to obtain the center point corresponding to each piece of boundary information.

[0111] In the case of obtaining the center point, the mean distances from the corresponding point cloud set g1, point cloud set g2, and point cloud set g3 of each piece of boundary information to the center point can be calculated, and the mean distances are denoted as d1, d2, and d3 respectively. Then, the missed detection rate of each piece of boundary information is calculated through the following formula:

[0112] Missed detection rate mi = number of point clouds in g1 / d1 + number of point clouds in g2 / d2 + number of point clouds in g3 / d3

[0113] According to the calculated missed detection rate of each piece of boundary information, the overall missed detection rate of all the boundary information is calculated. For example, the missed detection rates of each piece of boundary information calculated are added together to obtain the overall missed detection rate m.

[0114] Then, based on the overall missed detection rate m and the number of point clouds obs corresponding to the obstacle category in the point cloud data, the effective missed detection rate is analyzed. For example, the effective missed detection rate missrate can be calculated through the following formula:

[0115] missrate = m / (m + obs)

[0116] In the case where the segmentation evaluation parameters such as the false detection rate and the missed detection rate are obtained through analysis, the corresponding relationships between the segmentation evaluation parameters and the evaluation results can be obtained in various ways. For example, they can be obtained based on user settings. Another example is that they can be determined based on big data collection and analysis. This embodiment does not limit this. The manifestation form of the corresponding relationship can be flexibly selected. For example, the values of the segmentation evaluation parameters can be segmented, and each value segment corresponds to an evaluation result. Another example is that grading can be performed based on each piece of recognition information.

[0117] Exemplarily, taking the segmentation evaluation parameters including the false detection rate and the missed detection rate as an example, the segmentation evaluation parameters of the obstacles can be scored in small batches, the relationship between the manual scoring result and the false detection rate and the missed detection rate can be compared, the missed detection rate and the false detection rate can be graded, and an obstacle recognition level scoring table can be generated. Based on the obstacle recognition level scoring table, in subsequent obstacle recognition evaluations, by calculating the false detection rate and the missed detection rate, and by comparing with the obstacle recognition level scoring table, the quality of obstacle recognition can be obtained.

[0118] Based on the above segmentation result evaluation method, reliable evaluation of obstacle recognition can be achieved based on the DOM data of the target area once and based on the segmentation result. It can be understood that, in order to further improve the evaluation reliability, for the same target area, DOM data can be obtained multiple times, evaluations can be performed respectively based on the DOM data each time, and the obstacle recognition situation can be determined based on the multiple evaluation results.

[0119] In order to execute the corresponding steps in the above embodiments and all possible ways, an implementation manner of a segmentation result evaluation device is given below. Please refer to Figure 13 , Figure 13 is a functional module diagram of a segmentation result evaluation device 140 provided by an embodiment of the present invention. The segmentation result evaluation device 140 can be applied to Figure 1 the electronic device 100 shown. It should be noted that for the segmentation result evaluation device 140 provided in this embodiment, its basic principle and the technical effects generated are the same as those in the above embodiments. For the sake of brief description, for the parts not mentioned in this embodiment, reference can be made to the corresponding content in the above embodiments. The segmentation result evaluation device 140 includes a data acquisition module 141 and a data analysis module 142.

[0120] Among them, the data acquisition module 141 is used to acquire the point cloud data of the target area and segment the point cloud data to obtain a point cloud segmentation result.

[0121] The data analysis module 142 is used to determine the boundary information of the target area based on the segmentation result of the DOM data corresponding to the target area and the point cloud segmentation result; and determine the segmentation evaluation parameters of the point cloud data based on the boundary information and the point cloud segmentation result.

[0122] On this basis, an embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a computer program, and when the computer program runs, it controls the electronic device where the computer-readable storage medium is located to execute the above-mentioned segmentation result evaluation method.

[0123] By adopting the above solution in the embodiment of the present invention, the automatic evaluation of the segmentation result is reliably realized, and then the reliable evaluation of obstacle recognition is realized, which provides guarantee for subsequent reasonable planning of the operation route and avoids the danger of collision of operation equipment such as drones. Compared with manual review, the manpower consumption is greatly reduced and the evaluation efficiency is improved.

[0124] In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are only illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0125] In addition, in each embodiment of the present invention, the various functional modules may be integrated together to form an independent part, or each module may exist alone, or two or more modules may be integrated to form an independent part.

[0126] When the above-mentioned function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0127] The foregoing are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating segmentation results, characterized in that, Including: Obtain the point cloud data of the target area, and segment the point cloud data to obtain a point cloud segmentation result; Based on the segmentation results of each frame of DOM data corresponding to the target area, obtain the initial boundary information corresponding to the target area based on a boundary generation algorithm; Obtain the first point cloud data located within each of the initial boundary information, and determine the category information of the first point cloud data based on the point cloud segmentation result; Based on the category information and the quantity information of point clouds of different categories, select the initial boundary information that meets the preset conditions from the initial boundary information as the boundary information of the target area; The point cloud segmentation result includes the confidence levels of the categories of obstacle category and ground category corresponding to each point cloud in the point cloud data. Determine the point clouds with ambiguous categories from the point clouds of the boundary information based on the confidence level of each point cloud; Obtain the quantity ratio of the point clouds with ambiguous categories to the point cloud data, and / or obtain the point cloud set within each boundary information and the point cloud set outside each boundary information to determine the segmentation evaluation parameters of the point cloud data.

2. The segmentation result evaluation method according to claim 1, wherein The category information includes the ground category and the obstacle category; the step of selecting the initial boundary information that meets the preset conditions from the initial boundary information based on the category information and the quantity information of point clouds of different categories as the boundary information of the target area includes: Determine whether the initial boundary information meets the preset conditions based on the ratio of the quantity of point clouds of the ground category to the quantity of point clouds in the first point cloud data; When the ratio is greater than the preset threshold, determine that the initial boundary information meets the preset conditions, and use the initial boundary information as the boundary information of the target area.

3. The segmentation result evaluation method according to claim 2, wherein The point clouds with ambiguous categories include the point clouds of the fuzzy obstacle category, and the point clouds of the fuzzy obstacle category are the point clouds with confidence levels within a preset range among the point clouds corresponding to the obstacle category; The step of obtaining the quantity ratio of the point clouds with ambiguous categories to the point cloud data to determine the segmentation evaluation parameters of the point cloud data includes: Obtain the ratio of the quantity of point clouds of the fuzzy obstacle category to the quantity of point clouds corresponding to the obstacle category in the point cloud data to determine the false detection rate of obstacles.

4. The segmentation result evaluation method according to claim 2, characterized in that, The step of obtaining the point cloud set within each boundary information and the point cloud set outside each boundary information to determine the segmentation evaluation parameters of the point cloud data includes: Expand each boundary information by a set multiple to obtain target information; Based on the confidence level, for each boundary information, determine the point cloud set g1 corresponding to the fuzzy ground category from the point cloud set within the boundary information; determine the point cloud set g2 corresponding to the fuzzy ground category from the area between the boundary information and the target information; determine the point cloud set g3 corresponding to the fuzzy ground category from the area outside the target information; Analyze the missed detection rate based on the point cloud sets g1, g2, and g3; Wherein, the point clouds of the fuzzy ground category are the point clouds with confidence levels within a preset range among the point clouds corresponding to the ground category.

5. The segmentation result evaluation method according to claim 4, wherein The steps of analyzing the missed detection rate based on the point cloud sets g1, g2, and g3 include: Determine the center point of each boundary information; For each boundary information, based on the center point of the boundary information, and the corresponding point cloud sets g1, g2, and g3, determine the missed detection rate of the boundary information; According to the missed detection rate of each boundary information, calculate the overall missed detection rate of all the boundary information; Based on the overall missed detection rate and the number of point clouds corresponding to the obstacle category in the point cloud data, analyze and obtain the effective missed detection rate.

6. The segmentation result evaluation method according to any one of claims 2 to 5, characterized in that The confidence levels of the category of each point cloud in the point cloud data being the obstacle category and the ground category are obtained based on binary classification segmentation, and the sum of the confidence level of each point cloud corresponding to the obstacle category and the confidence level of the corresponding ground category is 1; the category to which each point cloud belongs is the category with a confidence level greater than 0.

5.

7. The segmentation result evaluation method according to claim 1, wherein, The steps of obtaining the point cloud data of the target area include: Process the DOM data corresponding to the target area based on feature matching analysis and 3D reconstruction algorithm to obtain the point cloud data of the target area; The segmentation result of the DOM data corresponding to the target area is obtained by performing semantic segmentation on the DOM data based on image segmentation technology.

8. A segmentation result evaluation device, characterized in that It includes: A data acquisition module, configured to acquire the point cloud data of the target area, and segment the point cloud data to obtain a point cloud segmentation result; A data analysis module, configured to obtain the initial boundary information corresponding to the target area based on the boundary generation algorithm based on the segmentation result of each frame of DOM data corresponding to the target area; Obtain the first point cloud data located within each initial boundary information, and determine the category information of the first point cloud data based on the point cloud segmentation result; Based on the category information and the quantity information of point clouds of different categories, select the initial boundary information that meets the preset conditions from the initial boundary information as the boundary information of the target area; the point cloud segmentation result includes the confidence levels of the category of each point cloud in the point cloud data being the obstacle category and the ground category, and determine the category ambiguous points from the point clouds of the boundary information based on the confidence level of each point cloud; Obtain the proportion of the category ambiguous points in the point cloud data, and / or obtain the point cloud set within each boundary information and the point cloud set outside each boundary information to determine the segmentation evaluation parameters of the point cloud data.

9. An electronic device, characterized in that, It includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the program, it implements the segmentation result evaluation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a computer program, and when the computer program runs, it controls the electronic device where the computer-readable storage medium is located to execute the segmentation result evaluation method according to any one of claims 1 to 7.

Images