Scene processing method, device, equipment, storage medium and product

In the massive scene test of autonomous driving vehicles, the sampling probability is optimized based on the degree of scene importance and target compression coefficient, and the problem of decreasing sample diversity and missing test risks caused by traditional uniform sampling is solved, and more efficient test coverage and Processing efficiency.

CN112583414BActive Publication Date: 2024-12-03BEIJING BAIDU NETCOM SCI & TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011446055.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-11
Publication Date
2024-12-03
Estimated Expiration
2040-12-11

AI Technical Summary

Technical Problem

In the massive scenario test of autonomous vehicles, traditional uniform sampling methods lead to a decrease in sample diversity, especially low-frequency scenarios being sparsely sampled, resulting in insufficient test coverage and risk of missed testing.

Method used

Based on the target compression coefficient of the scene set and the degree of importance of each scene, the sampling probability of each scene is determined, and the sampling process is optimized to improve diversity and comprehensiveness through event mining and information entropy analysis.

Benefits of technology

It realizes effective compression of massive scenarios, improves sampling diversity and comprehensiveness, reduces the risk of missed testing in low-frequency scenarios, and improves testing efficiency and processing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112583414B_ABST
    Figure CN112583414B_ABST
Patent Text Reader

Abstract

The present disclosure discloses a scene processing method, device, equipment, storage medium and product, which are related to the fields of autonomous driving and intelligent transportation. The specific implementation scheme is: based on the target compression coefficient of the scene set and the importance value of each scene in the scene set, the probability of each scene being sampled is determined; based on the probability of each scene being sampled, the scene set is sampled to achieve compression of the scene set. The sampling diversity is improved, and a balance between processing efficiency, sampling diversity and comprehensiveness is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, and in particular to the fields of autonomous driving and intelligent transportation. Background Art

[0002] Unmanned vehicles, also known as autonomous vehicles, are autonomous driving controls based on big data and can cope with various scenarios. In order to control accurately, it is necessary to test various scenarios covered by the vehicle. From the perspective of vehicle test coverage, the more scenarios tested, the more comprehensive the coverage. However, as the number of scenarios increases, the cycle and resource requirements for a full test will increase significantly in the face of massive scenarios. Therefore, from the perspective of practical use, resource event consumption can be reduced as much as possible while meeting a certain coverage rate. In traditional testing schemes, a certain proportion of random and uniform sampling is generally used for testing massive scenarios to maintain consistency with the original sample distribution. Although this scheme can achieve the compression effect of massive scenarios, the obvious disadvantage of this scheme is that while keeping the overall distribution of samples unchanged, uniform sampling leads to a decrease in sample diversity, especially low-frequency scenarios (i.e., uncommon scenarios) with a low proportion will be lost due to sparse sampling, which will cause insufficient test coverage and thus cause the risk of missed tests. Summary of the invention

[0003] The present disclosure provides a scene processing method, apparatus, device, storage medium and product.

[0004] According to one aspect of the present disclosure, a scene processing method is provided, comprising:

[0005] Determining the probability of each scene being sampled based on a target compression factor of the scene set and an importance value of each scene in the scene set;

[0006] The scene set is sampled based on the probability of each of the scenes being sampled.

[0007] According to another aspect of the present disclosure, there is provided a scene processing device, comprising:

[0008] A first determination module is used to determine the probability of each scene being sampled based on a target compression coefficient of a scene set and an importance value of each scene in the scene set;

[0009] The scene sampling module is used to sample the scene set based on the probability of each scene being sampled.

[0010] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0011] at least one processor; and

[0012] a memory communicatively connected to the at least one processor; wherein,

[0013] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method in any embodiment of the present disclosure.

[0014] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided. The computer instructions are used to enable a computer to execute the method in any embodiment of the present disclosure.

[0015] According to another aspect of the present disclosure, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method according to any embodiment of the present disclosure is implemented.

[0016] The technology disclosed in the present invention improves sampling diversity and achieves a balance among processing efficiency, sampling diversity and comprehensiveness.

[0017] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0019] Figure 1 is a schematic diagram of a scene processing method according to a first embodiment of the present disclosure;

[0020] Figure 2 is a schematic diagram of a scene processing method according to a second embodiment of the present disclosure;

[0021] Figure 3 is an event frequency timing diagram according to the second embodiment of the present disclosure;

[0022] Figure 4 is an information entropy timing diagram according to the second embodiment of the present disclosure;

[0023] Figure 5 is a schematic diagram of a scene processing method according to a third embodiment of the present disclosure;

[0024] Figure 6 is a schematic diagram of a scene processing method according to a fourth embodiment of the present disclosure;

[0025] Figure 7 is a schematic diagram of a scene processing method according to a fifth embodiment of the present disclosure;

[0026] Figure 8is a schematic diagram of a scene processing method according to a sixth embodiment of the present disclosure;

[0027] Fig. 9 is a schematic diagram of the relationship between the entropy threshold and the test difference according to the sixth embodiment of the present disclosure;

[0028] Fig.10 is a system framework diagram of a scene processing method according to a seventh embodiment of the present disclosure;

[0029] Fig.11 is a schematic diagram of a scene processing method according to a seventh embodiment of the present disclosure;

[0030] Fig.12 is a schematic diagram of the structure of a scene processing device according to an embodiment of the present disclosure Figure 1 ;

[0031] Fig.13 is a schematic diagram of the structure of a scene processing device according to an embodiment of the present disclosure Figure 2 ;

[0032] Fig.14 is a schematic diagram of the structure of a scene processing device according to an embodiment of the present disclosure Figure 3 ;

[0033] Fig.15 is a schematic diagram of the structure of a scene processing device according to an embodiment of the present disclosure Figure 4 ;

[0034] Fig.16 It is a block diagram of an electronic device used to implement the scene processing method of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0035] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0036] The present disclosure provides a scene processing method. Figure 1 As shown, the method includes:

[0037] Step S101: Based on the target compression coefficient of the scene set and the importance value of each scene in the scene set, the probability of each scene being sampled is determined.

[0038] Step S102: based on the probability of each scene being sampled, the scene set is sampled to achieve compression of the scene set.

[0039] The above scene set may include various scenes that autonomous driving can deal with. Here, as the number of scenes increases, a large number of scenes can be formed. At this time, it is necessary to sample the scene set to obtain a part of the scenes to achieve compression of the scene set.

[0040] In this example, the compression factor refers to the ratio of the number of scenes sampled in the scene set to the total number of scenes. For example, if there are 100,000 scenes in the scene set and 30,000 scenes are sampled from the scene set, the compression factor is 0.3. The target compression factor is the compression factor that actually needs to be achieved.

[0041] The importance value of a scene is a quantitative value of the importance of the scene. In practical applications, it can be quantified based on certain important factors in the scene according to actual needs. The higher the importance value of a scene, the more important the scene is. The more important the scene is, the greater the probability that the scene is sampled.

[0042] This scheme can determine the probability of a scene being sampled based on the importance value of the scene and the target compression coefficient. Sampling based on this can enable various important scenes (for example, uncommon but important scenes) to be sampled, thereby improving the diversity of sampled scenes, providing more complete sampling coverage, and further improving the comprehensiveness of sampling. In this way, it not only achieves compression of massive scenes and ensures the efficiency of subsequent processing, but also takes into account the diversity and comprehensiveness of sampling, thereby achieving a balance among processing efficiency, diversity and comprehensiveness.

[0043] In one embodiment, before determining the probability of each scene being sampled based on the target compression coefficient of the scene set and the importance value of each scene in the scene set, the scene processing method also includes: counting the number of events occurring at each moment of the scene; and determining the importance value of the scene based on the number of events occurring at each moment.

[0044] Events are a reference to some important information in a scene. In the scene of autonomous driving, events can be a motor vehicle cutting in, a pedestrian crossing, or passing an intersection. Events in practical applications include but are not limited to the above events. In addition, events do not need to be strictly orthogonal, as long as they can remain different. For example, encountering a pedestrian when turning left and turning left can exist at the same time and can be considered two different events.

[0045] In practical applications, events can be mined for scenes based on data mining techniques such as machine learning. After event mining, mining information such as the start and end time of each event can be obtained, so as to count the number of events occurring at each moment in the scene based on the mining information.

[0046] For a scene, the number of events at each moment reflects the frequency of events at each moment of the scene, and can reflect whether a scene is a scene with high event incidence or a scene with low event incidence. Therefore, the frequency of events is a very important factor of the scene. In this embodiment, by counting the number of events at each moment of the scene, the importance value of the scene is determined based on the statistical results. This idea of ​​sampling event mining has a more in-depth characterization of the scene, making the subsequent scene selection more targeted, and solving the problem of how to focus on important information of the scene. In this way, with events as the main goal, various important scenes are sampled from the perspective of event frequency, thereby achieving an improvement in sampling diversity.

[0047] In one embodiment, based on the number of events occurring at each moment, the importance value of the scene is determined, such as Figure 2 As shown, the specific implementation methods may include:

[0048] Step S201: Calculate the information entropy corresponding to each moment based on the number of events occurring at each moment.

[0049] In this step, first, the number of events occurring at each moment can be normalized and processed into a value of 0 to 1 as the frequency of event occurrence at each moment. There are at least two ways to perform normalization: one is to perform individual normalization, that is, perform normalization in a single scene, determine the maximum value of the number of events occurring in the single scene as the first maximum value, and divide the number of events occurring at each moment in the single scene by the first maximum value to obtain a value of 0 to 1. The second is to perform global normalization, that is, perform normalization in the entire scene set, determine the maximum value of the number of events occurring in the entire scene set as the second maximum value, and then divide the number of events occurring at each moment in each scene by the second maximum value to obtain a value of 0 to 1. The processing results for each scene can be found in. Figure 3 In the event frequency timing diagram shown, the horizontal axis is the time t and the vertical axis is the event occurrence frequency P.

[0050] Afterwards, based on the frequency of events at each moment, the information entropy corresponding to each moment is obtained.

[0051] Information entropy is used to measure the amount of information. Information entropy with high information content is low. The formula for calculating information entropy is as follows:

[0052] e=-P*ln(P)-(1-P)*ln(1-P) (1)

[0053] Among them, e represents information entropy and P is the frequency of event occurrence.

[0054] like Figure 4As shown in the figure, the dotted curve represents the event frequency time series diagram, the solid curve represents the information entropy time series diagram, the horizontal axis is the time t, and the vertical axis is the information entropy e. It can be seen from the figure that the information entropy corresponding to the high frequency moment (that is, the moment when the number of events is high) and the low frequency moment (that is, the moment when the number of events is low) is low.

[0055] Step S202: Count the moments when the information entropy is lower than or equal to the first threshold, and obtain the length of the small entropy period.

[0056] The first threshold value can be set in advance according to actual needs. Figure 4 As shown, the dotted straight line represents the first threshold, that is, the entropy threshold θ. Based on the first threshold, filtering statistics are performed, and the moments when the information entropy is lower than or equal to the first threshold are counted to obtain the length of the small entropy period, and the high-frequency period and the low-frequency period in the scene are accurately selected.

[0057] Step S203: Calculate the proportion of the small entropy period length in the total period length of the scene as the importance value of the scene.

[0058] The high-frequency period is a period of high-incidence events. The events that occur in this period are high-value events, and their proportion in the compressed scene should be higher. Although the low-frequency period is a period of low-incidence events, it is more likely to have some uncommon events, which is conducive to improving diversity and comprehensiveness. Therefore, the events that occur in this period are also high-value events, and more should be retained in the compressed scene. In this embodiment, the importance of the scene is measured based on information entropy, and the importance value is obtained by the proportion of the selected small entropy period, so that both the high-frequency period and the low-frequency period can be used as part of the contribution to the importance of the scene. Especially for the scene with a large contribution in the low-frequency period, its importance value is high and it is more likely to be sampled. Even the scene that is not hit by any event can be retained with the highest probability, avoiding the possibility of loss due to incomplete event coverage, and avoiding the loss of uncommon scenes. In this way, through the reasonable retention of high-frequency period and low-frequency period, the data of high-frequency period and low-frequency period simultaneously affect the sampling, and the balance of diversity and comprehensiveness is fully considered. In addition, the scheme for importance analysis based on information entropy can obtain the importance value of the scene based on only the first threshold, which is more efficient.

[0059] In one embodiment, based on the number of events occurring at each moment, the importance value of the scene is determined, such as Figure 5 As shown, the specific implementation methods may include:

[0060] Step S501: Count the moments when the number of events occurring is lower than or equal to a second threshold, and obtain the time period length of low-frequency events.

[0061] Step S502: Count the moments when the number of events occurring is greater than or equal to a third threshold, and obtain the time period length of high-frequency events.

[0062] Step S503: Calculate the proportion of the sum of the time period length of the low-frequency event and the time period length of the high-frequency event in the total time period length of the scene as the importance value of the scene.

[0063] The third threshold is greater than the second threshold. The second threshold and the third threshold can be set according to actual needs.

[0064] In this way, the importance of the scene can be measured directly based on the number of events, and high-frequency periods and low-frequency periods can be selected. The importance value can be obtained based on the proportion of the high-frequency periods and low-frequency periods. The high-frequency periods and low-frequency periods can also be made part of the contribution to the importance of the scene, thereby improving the diversity and comprehensiveness of sampling.

[0065] In one embodiment, based on the target compression factor of the scene set and the importance value of each scene in the scene set, the probability of each scene being sampled is determined, such as Figure 6 As shown, the specific implementation methods may include:

[0066] Step S601: Based on the importance value of the scene, determine the initial probability of the scene being sampled.

[0067] Step S602: based on the initial probability of the scene being sampled, obtain an initial compression coefficient corresponding to the scene.

[0068] Step S603: Calculate the ratio of the target compression coefficient corresponding to the scene to the initial compression coefficient corresponding to the scene.

[0069] Step S604: taking the product of the initial probability of the scene being sampled and the ratio as the probability of the scene being sampled, so as to obtain the probability of each scene being sampled.

[0070] The scene set is sampled based on the initial probability of each scene being sampled. The initial compression coefficient of the scene set that is achieved may not be equal to the target compression coefficient, that is, it does not meet the compression target of the scene set. By converting the initial probability of the scene being sampled through the ratio of the target compression coefficient to the initial compression coefficient, the final probability of the scene being sampled is obtained, thereby accurately achieving the compression target of the scene set.

[0071] In one embodiment, based on the importance value of the scene, the initial probability of the scene being sampled is determined, which may specifically include: using the importance value of the scene as the initial probability of the scene being sampled. Specifically, the importance value of the scene may be processed into a value of 0 to 1, and the importance value of the scene may be directly used as the initial probability of the scene being sampled. The higher the importance value, the higher the initial probability of the scene being sampled. In this way, the processing is simple and fast, and the processing efficiency is improved.

[0072] In one implementation, based on the initial probability of the scene being sampled, an initial compression coefficient corresponding to the scene is obtained, such as Figure 7 As shown, the specific implementation methods may include:

[0073] Step S701: obtaining an initial sampled expected value corresponding to the scene based on the initial probability of the scene being sampled.

[0074] Specifically, the sampled initial expected value corresponding to the scene is:

[0075] E=1*P0+0*(1-P0)=P0 (2)

[0076] Where E represents the initial expected value of the scene being sampled, and P0 represents the initial probability of the scene being sampled. Based on this, the initial expected value of the scene being sampled is equal to the initial probability of the scene being sampled.

[0077] Step S702: Count the initial expected values ​​corresponding to all scenes to obtain the expected number of scenes sampled in the scene set.

[0078] Specifically, the initial expected values ​​corresponding to all scenes are summed to obtain the expected number of scenes sampled in the scene set.

[0079] Step S703: Calculate the ratio of the expected number to the total number of scenes in the scene set to obtain an initial compression coefficient.

[0080] In this way, based on the initial expected value of the samples that can be achieved based on the initial probability of each scene being sampled, the comprehensive expected number of samples is obtained. The ratio of this expected number to the total number of scenes in the scene set is the comprehensive initial compression coefficient, and the calculation method of this initial compression coefficient is simple and accurate.

[0081] In one embodiment, the scene processing method may further include: testing the sampled target scene to obtain a first test result. In the scenario of autonomous driving, the sampled target scene may be tested for autonomous driving. During the test, a simulation test may be performed based on a simulation environment of a simulation test engine, or based on a real environment. Since the diversity and comprehensiveness of the sampled target scenes are improved, a balance is achieved in test efficiency, diversity, and comprehensiveness during the test.

[0082] In one embodiment, if Figure 8 As shown, the scene processing method may further include:

[0083] Step S801: Test all the scenes in the scene set to obtain a second test result.

[0084] Step S802: Determine the difference information between the first test result and the second test result.

[0085] Specifically, in the autonomous driving test, the difference information between the first test result and the second test result can be determined based on the test results of the simulation test engine. The test results are determined according to the actual measurement system, and can provide a variety of indicators such as the allowable failure amount, safety indicators, and comfort indicators for quantitative analysis. The goal of the analysis is to determine the test difference (i.e., difference information) between the first test result and the second test result, and the goal of the compression is to obtain consistent results as much as possible so that the test difference is small. For example, the difference between the first test result and the second test result and the percentage of the second test result can be used as the test difference.

[0086] Step S803: Based on the difference information, adjust the importance value for the target compression coefficient.

[0087] When the target compression coefficient is constant, the importance values ​​of the scenes are different, and accordingly, the sampling results are also different, and the final test results may also be different. The more scenes are sampled, the smaller the test variability corresponding to the test results, but the test efficiency will also decrease accordingly. In order to take into account both the test efficiency and the test variability, it is necessary to adjust the importance value of the target compression coefficient to achieve a balance between test efficiency and test variability, which is to achieve closed-loop verification of the compression effect.

[0088] In this way, by adding a compression effect verification loop, the test effects before and after compression are differentiated and analyzed, and the importance value of the target compression coefficient is flexibly adjusted, so as to better perform targeted compression control according to the characteristics of the actual scene set and achieve scene compression tuning.

[0089] The above closed-loop verification process is not necessary, and the effect comparison can also be achieved through manual qualitative analysis and other methods to directly adjust the importance value of the target compression factor.

[0090] When the importance of a scene is measured based on information entropy, the importance value can be adjusted by adjusting the size of the entropy threshold. When the entropy threshold changes, the importance value of the scene changes, the initial probability of the scene being sampled, and the ratio of the target compression coefficient to the initial compression coefficient will change accordingly, causing the probability of the scene being sampled to change. Based on this, the sampling results of the scene set will also change, and the final test results will also change, thereby adjusting the size of the test difference. Fig. 9 As shown, the relationship between the test difference obtained by statistics and the entropy threshold θ is illustrated. As the entropy threshold increases, the test difference gradually decreases. In implementation, the importance value for the target compression factor can be adjusted based on the relationship between the test difference and the entropy threshold θ. In this way, rapid adjustment can be achieved to meet actual needs.

[0091] The following uses a specific application scenario as an example to illustrate the scenario processing method provided by the embodiment of the present disclosure in more detail.

[0092] In this embodiment, the scene set to be tested is a large number of scenes to be tested for autonomous driving. Fig.10 The closed-loop feedback system architecture is implemented as shown, and the closed-loop feedback process implemented based on the closed-loop feedback system is described in detail below.

[0093] like Fig.10 As shown in the figure, the entire scene processing process can be divided into the following three parts:

[0094] Part 1: The massive scene data mining process mainly focuses on events.

[0095] Since the mining of massive scene data is based on events, a list of core events can be set in advance, such as being cut in by a motor vehicle, pedestrians crossing, and passing through an intersection. After that, the data miner mines the events and obtains the event mining results. After event mining, the start and end time of each event can be output for the subsequent information entropy analysis process.

[0096] Part 2: Index analysis and data compression process, mainly uses information entropy as the analysis target to compress data.

[0097] The focus of this process is to obtain the sampling index (i.e., the importance value of the scene) required for the compression of massive scenes from the event mining results, and to achieve compression based on certain compression parameters (i.e., the initial probability of the scene being sampled and the ratio of the target compression coefficient to the initial compression coefficient). This process is divided into two sub-processes: index solution and data compression.

[0098] Assume that m events are mined for n autonomous driving scenes (ads). Based on the event mining results and the core event list, the analyzer performs information entropy analysis to obtain sampling indicators, i.e., indicator solution. Fig.11 As shown in Figure 2, the indicator solution subprocess includes:

[0099] Step 1: Time series statistics of event frequency.

[0100] First, perform event frequency time series statistics for each scene, that is, count the number of events occurring at each moment, and then perform normalization processing (either individual normalization or global normalization) to process the values ​​​​into 0-1. The final result can be seen in Figure 3 The frequency timing diagram is shown.

[0101] Step 2: Solve the information entropy.

[0102] Based on the frequency time series diagram, the information entropy is solved using formula (1). The final result can be found in Figure 4 The information entropy timing diagram is shown.

[0103] Step 3: Information entropy filtering statistics.

[0104] This step is to perform threshold filtering statistics on information entropy, select an entropy threshold θ, and count the length of the time period in which the information entropy is lower than the entropy threshold in the time series to obtain the proportion of the small entropy period in the total length of the scene period, that is, the small entropy ratio. The reason for choosing small entropy is to ensure that events in high-frequency and low-frequency periods can be regarded as part of the importance contribution.

[0105] Step 4: Sampling indicators.

[0106] The obtained small entropy ratio is used as the sampling index of the scene.

[0107] The data compression sub-process includes:

[0108] Step 1. After obtaining the sampling index of the scene, directly use the sampling index as the initial probability of the scene being sampled, and then combine it with the compression coefficient (that is, the ratio of the target compression coefficient corresponding to the scene to the initial compression coefficient corresponding to the scene) to obtain the comprehensive sampling probability (that is, the probability of the scene being sampled), that is: comprehensive sampling probability = sampling index * compression coefficient.

[0109] Step 2: The data compressor randomly samples each scene according to the comprehensive sampling probability, that is, generates a 0-1 random number according to the comprehensive sampling probability. If the random number is 1, the scene is selected, otherwise the scene is abandoned.

[0110] After sampling, the final compressed scene is obtained.

[0111] Part 3: The closed-loop verification process of compression effect is mainly to achieve the effect of closed-loop verification in order to select appropriate compression parameters.

[0112] The compression parameters may include the above sampling index and compression coefficient. The sampling index and compression coefficient are related to the selection of the entropy threshold. Different entropy thresholds result in different sampling indexes and compression coefficients of the obtained scenes.

[0113] The goal of this process is to verify the effect after compression, and to optimize the selection of compression parameters and entropy thresholds through verification, which plays a closed-loop adjustment process. This process mainly uses the original data (i.e., massive scene data) and the compressed data as test sources, respectively, for testing, and then analyzes the differences in the test results. The test process requires the support of a test engine, and a simulation test engine can be used here. The test engine obtains test result A after testing massive scene data, and obtains test result B after testing compressed data. The test results are determined according to the actual measurement system. Here you can view various indicators such as the number of operating failures, safety indicators, and comfort indicators for quantitative analysis. The result difference analyzer analyzes test result A and test result B to obtain the result difference (i.e., test difference). Based on this, the compression parameter optimizer optimizes the compression parameters to reduce the difference. By adjusting different entropy thresholds and compression parameters, you can get the following: Figure 5 The relationship between the test difference and the entropy threshold θ is shown in FIG. In actual use, a more appropriate entropy threshold can be found according to the relationship between the test difference and the entropy threshold θ, and then a suitable compression parameter can be selected.

[0114] The solution of this embodiment improves the efficiency of mass testing. In addition, event mining is performed based on data mining technology, and based on the event mining results, the information entropy of the scene is analyzed, which retains the diversity information of the scene, thereby improving the comprehensiveness of sampling, and reducing the risk of low-frequency scenes being missed during testing. In addition, effective compression of mass scenes is achieved through certain closed-loop analysis and parameter control to achieve a balance between coverage (i.e. comprehensiveness) and efficiency, and ultimately achieves a balance between test efficiency, diversity and comprehensiveness, and the test results are more valuable.

[0115] The present disclosure also provides a scene processing device, such as Fig.12 As shown, including:

[0116] A first determination module 1201 is used to determine the probability of each scene being sampled based on the target compression coefficient of the scene set and the importance value of each scene in the scene set;

[0117] The scene sampling module 1202 is used to sample the scene set based on the probability of each scene being sampled, so as to achieve compression of the scene set.

[0118] In one embodiment, if Fig.13 As shown, the second determining module 1203 is also included, which is used to:

[0119] Count the number of events that occur at each moment in the scene;

[0120] Based on the number of events occurring at each moment, the importance value of the scene is determined.

[0121] In one implementation, the second determining module 1203 is specifically configured to:

[0122] Calculate the information entropy corresponding to each moment based on the number of events occurring at each moment;

[0123] Counting the moments when the information entropy is lower than or equal to the first threshold, and obtaining the length of the small entropy period;

[0124] The proportion of the small entropy period length in the total period length of the scene is calculated as the importance value of the scene.

[0125] In one implementation, the second determining module 1203 is specifically configured to:

[0126] Counting the moments when the number of events is lower than or equal to the second threshold, and obtaining the length of the period of low-frequency events;

[0127] Counting the moments when the number of events occurring is greater than or equal to the third threshold, and obtaining the length of the period of high-frequency events;

[0128] The proportion of the sum of the time period length of low-frequency events and the time period length of high-frequency events in the total time period length of the scene is calculated as the importance value of the scene.

[0129] In one implementation, the first determining module 1201 is specifically configured to:

[0130] Based on the importance value of the scene, determine the initial probability of the scene being sampled;

[0131] Based on the initial probability of the scene being sampled, the initial compression coefficient corresponding to the scene is obtained;

[0132] Calculate the ratio of the target compression coefficient corresponding to the scene to the initial compression coefficient corresponding to the scene;

[0133] The product of the initial probability of the scene being sampled and the ratio is taken as the probability of the scene being sampled to obtain the probability of each scene being sampled.

[0134] In one implementation, the first determining module 1201 is specifically configured to:

[0135] The importance value of the scene is used as the initial probability of the scene being sampled.

[0136] In one implementation, the first determining module 1201 is specifically configured to:

[0137] Based on the initial probability of the scene being sampled, the initial expected value of the scene being sampled is obtained;

[0138] Count the initial expected values ​​corresponding to all scenes to obtain the expected number of scenes sampled in the scene set;

[0139] The ratio of the expected number to the total number of scenes in the scene set is calculated to obtain the initial compression factor.

[0140] In one embodiment, if Fig.14 As shown, it also includes a first test module 1204, which is used to:

[0141] The sampled target scene is tested to obtain a first test result.

[0142] In one embodiment, if Fig.15 As shown, it also includes:

[0143] The second test module 1205 is used to test all the scenes in the scene set to obtain a second test result;

[0144] A difference determination module 1206, configured to determine difference information between the first test result and the second test result;

[0145] The difference adjustment module 1207 is used to adjust the importance value for the target compression coefficient based on the difference information.

[0146] The functions of each module in each device of the embodiments of the present disclosure can be found in the corresponding description in the above-mentioned scene processing method, and will not be repeated here.

[0147] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0148] Fig.16 A schematic block diagram of an example electronic device 1600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0149] like Fig.16 As shown, the device 1600 includes a computing unit 1601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1602 or a computer program loaded from a storage unit 1608 into a random access memory (RAM) 1603. In the RAM 1603, various programs and data required for the operation of the device 1600 can also be stored. The computing unit 1601, the ROM 1602, and the RAM 1603 are connected to each other via a bus 1604. An input / output (I / O) interface 1605 is also connected to the bus 1604.

[0150] A number of components in the device 1600 are connected to the I / O interface 1605, including: an input unit 1606, such as a keyboard, a mouse, etc.; an output unit 1607, such as various types of displays, speakers, etc.; a storage unit 1608, such as a disk, an optical disk, etc.; and a communication unit 1609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1609 allows the device 1600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0151] The computing unit 1601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 160 performs the various methods and processes described above, such as the scene processing method. For example, in some embodiments, the scene processing method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 1608. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1600 via ROM 1602 and / or communication unit 1609. When the computer program is loaded into RAM 1603 and executed by the computing unit 1601, one or more steps of the scene processing method described above may be performed. Alternatively, in other embodiments, the computing unit 1601 may be configured to perform the scene processing method in any other appropriate manner (e.g., by means of firmware).

[0152] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0153] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0154] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0156] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0157] A computer system may include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship to each other.

[0158] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0159] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A scene processing method, comprising: Determining the probability of each scene being sampled based on a target compression coefficient of a scene set and an importance value of each scene in the scene set, wherein the scene set includes a plurality of autonomous driving scenes, and the target compression coefficient is used to represent a ratio of the number of sampled scenes in the scene set to the total number of the scenes that needs to be achieved when sampling the scene set; Sampling the scene set based on the probability of each scene being sampled; Performing an autonomous driving test on the sampled target scene to obtain a first test result; Among them, the probability of each scene being sampled is determined based on the target compression coefficient of the scene set and the importance value of each scene in the scene set, including: determining the initial probability of the scene being sampled based on the importance value of the scene; obtaining the initial compression coefficient corresponding to the scene set based on the initial probability of the scene being sampled; calculating the ratio of the target compression coefficient corresponding to the scene set to the initial compression coefficient corresponding to the scene set; and taking the product of the initial probability of the scene being sampled and the ratio as the probability of the scene being sampled to obtain the probability of each scene being sampled.

2. The method according to claim 1, before determining the probability of each scene being sampled based on the target compression coefficient of the scene set and the importance value of each scene in the scene set, further comprising: Count the number of events occurring at each moment of the scenario; Based on the number of events occurring at each moment, the importance value of the scene is determined.

3. The method according to claim 2, wherein: The step of determining the importance value of the scene based on the number of events occurring at each moment includes: Calculate the information entropy corresponding to each moment based on the number of events occurring at each moment; Counting the moments when the information entropy is lower than or equal to the first threshold, and obtaining the length of the small entropy period; The proportion of the small entropy period length in the total period length of the scene is calculated as the importance value of the scene.

4. The method according to claim 2, wherein: The step of determining the importance value of the scene based on the number of events occurring at each moment includes: Counting the moments when the number of occurrences of the event is lower than or equal to a second threshold, and obtaining the length of a period of low-frequency events; Counting the moments when the number of occurrences of the event is greater than or equal to a third threshold, to obtain the length of a period of high-frequency events; The proportion of the sum of the time period length of the low-frequency event and the time period length of the high-frequency event in the total time period length of the scene is calculated as the importance value of the scene.

5. The method according to claim 1, wherein: The determining, based on the importance value of the scene, an initial probability of the scene being sampled includes: The importance value of the scene is used as the initial probability of the scene being sampled.

6. The method according to claim 1, wherein: The obtaining an initial compression coefficient corresponding to the scene based on the initial probability of the scene being sampled includes: Obtaining an initial expected value of the scene based on the initial probability of the scene being sampled; Counting the initial expected values ​​corresponding to all the scenes to obtain the expected number of the scenes sampled in the scene set; The ratio of the expected number to the total number of the scenes in the scene set is calculated to obtain the initial compression coefficient.

7. The method according to claim 1, further comprising: Testing all the scenarios in the scenario set to obtain a second test result; Determining difference information between the first test result and the second test result; Based on the difference information, the importance level value for the target compression factor is adjusted.

8. A scene processing device, comprising: A first determination module is used to determine the probability of each scene being sampled based on a target compression coefficient of a scene set and an importance value of each scene in the scene set, wherein the scene set includes a plurality of autonomous driving scenes, and the target compression coefficient is used to represent a ratio of the number of sampled scenes in the scene set to the total number of the scenes that needs to be achieved when sampling the scene set; A scene sampling module, used for sampling the scene set based on the probability of each scene being sampled; The device also includes: a first test module, configured to perform an automatic driving test on the sampled target scene to obtain a first test result; Among them, the first determination module is specifically used to: determine the initial probability of the scene being sampled based on the importance value of the scene; obtain the initial compression coefficient corresponding to the scene set based on the initial probability of the scene being sampled; calculate the ratio of the target compression coefficient corresponding to the scene set to the initial compression coefficient corresponding to the scene set; use the product of the initial probability of the scene being sampled and the ratio as the probability of the scene being sampled, so as to obtain the probability of each of the scenes being sampled.

9. The apparatus according to claim 8, further comprising a second determining module, configured to: Count the number of events occurring at each moment of the scenario; Based on the number of events occurring at each moment, the importance value of the scene is determined.

10. The device according to claim 9, wherein: The second determining module is specifically used for: Calculate the information entropy corresponding to each moment based on the number of events occurring at each moment; Counting the moments when the information entropy is lower than or equal to the first threshold, and obtaining the length of the small entropy period; The proportion of the small entropy period length in the total period length of the scene is calculated as the importance value of the scene.

11. The device according to claim 9, wherein: The second determining module is specifically used for: Counting the moments when the number of occurrences of the event is lower than or equal to a second threshold, and obtaining the length of a period of low-frequency events; Counting the moments when the number of occurrences of the event is greater than or equal to a third threshold, to obtain the length of a period of high-frequency events; The proportion of the sum of the time period length of the low-frequency event and the time period length of the high-frequency event in the total time period length of the scene is calculated as the importance value of the scene.

12. The device according to claim 8, wherein: The first determining module is specifically used for: The importance value of the scene is used as the initial probability of the scene being sampled.

13. The device according to claim 8, wherein: The first determining module is specifically used for: Obtaining an initial expected value of the scene based on the initial probability of the scene being sampled; Counting the initial expected values ​​corresponding to all the scenes to obtain the expected number of the scenes sampled in the scene set; The ratio of the expected number to the total number of the scenes in the scene set is calculated to obtain the initial compression coefficient.

14. The apparatus according to claim 8, further comprising: A second testing module, used for testing all the scenarios in the scenario set to obtain a second testing result; A difference determination module, used to determine difference information between the first test result and the second test result; A difference adjustment module is used to adjust the importance value for the target compression coefficient based on the difference information.

15. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

16. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to make a computer execute the method according to any one of claims 1 to 7.

17. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.