Target object identification method and electronic equipment

Abstract

The invention relates to a target object identification method and electronic equipment, and the method comprises the steps: inputting the data of a target object obtained in a current calculation period into a learning-based key target object identification model and a rule-based key target object identification model, obtaining a first recognition result and a second recognition result corresponding to the target object in the current calculation period; when the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet the preset probability requirement, determining a current waiting period threshold value corresponding to the current calculation period according to the first recognition result and the second recognition result; and generating a comprehensive identification result of the target object in the current calculation period according to a comparison result of the waiting period number and the current waiting period threshold value, the first identification result and the second identification result. According to the technical scheme, the robustness and the function safety level of the whole system can be effectively improved, and the target object can be recognized more accurately in the automatic driving process.

Description

Technical Field

[0001] The present disclosure relates to the field of autonomous driving technology, and in particular to a target object recognition method and electronic equipment. Background Art

[0002] After the multi-sensor data fusion system has processed all kinds of sensor data, the number of fused targets may still be large. Therefore, selecting key targets for the fused targets will help the autonomous driving system reduce algorithm complexity, improve computing efficiency, and further improve the timeliness, accuracy and safety of autonomous driving to ensure that the vehicle passes through the current area safely and effectively.

[0003] The target recognition algorithms under the autonomous driving system can be summarized into the following five categories: methods based on vehicle motion trajectory estimation, methods based on probability prediction calculation, grid methods, artificial potential field methods, and methods based on machine learning and deep learning. The first four methods are rule-based algorithms, also known as "expert systems". Among them, the "expert system" cannot take into account various event types or conditions, which may cause some boundary conditions to use default values, resulting in inaccurate or untimely target recognition. Methods based on machine learning and deep learning are difficult to learn data not included in the data set. In addition, for scene data not included in the training set, the actual response of the model is unknown, and its safety is questionable.

[0004] In view of this, the present invention is proposed. Summary of the invention

[0005] In order to solve the above technical problems or at least partially solve the above technical problems, the embodiments of the present disclosure provide a target object recognition method and electronic device, which can effectively improve the overall system robustness and functional safety level, improve the overall system performance, and make the recognition of target objects more accurate during autonomous driving.

[0006] In a first aspect, an embodiment of the present disclosure provides a method for identifying a target object, the method comprising:

[0007] Inputting the data of the target object acquired in the current calculation cycle into the key target object recognition model based on learning and the key target object recognition model based on rules respectively, to obtain a first recognition result and a second recognition result corresponding to the target object in the current calculation cycle, wherein the target object is an object in the surrounding environment of the autonomous driving vehicle, the first recognition result is a recognition result output by the key target object recognition model based on learning, and the second recognition result is a recognition result output by the key target object recognition model based on rules;

[0008] When the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet the preset probability requirement, perform the following operations:

[0009] Determine a current waiting period threshold corresponding to a current calculation period according to the first recognition result and the second recognition result;

[0010] A comprehensive recognition result of the target object in the current calculation cycle is generated according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result.

[0011] In a second aspect, an embodiment of the present disclosure further provides an electronic device, the electronic device comprising:

[0012] one or more processors;

[0013] A storage device for storing one or more programs;

[0014] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the methods described above.

[0015] In a third aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which implements any of the methods described above when executed by a processor.

[0016] An embodiment of the present disclosure provides a method for identifying a target object, an electronic device, and a storage medium, comprising inputting data of the target object acquired in the current computing cycle into a learning-based key target object recognition model and a rule-based key target object recognition model, respectively, to obtain a first recognition result and a second recognition result corresponding to the target object in the current computing cycle, wherein the target object is an object in the surrounding environment of the autonomous driving vehicle, the first recognition result is a recognition result output by the learning-based key target object recognition model, and the second recognition result is a recognition result output by the rule-based key target object recognition model; when the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet the preset probability requirement, executing: determining a current waiting cycle threshold corresponding to the current computing cycle according to the first recognition result and the second recognition result; and generating a comprehensive recognition result of the target object in the current computing cycle according to a comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result, and the second recognition result. The above-mentioned comprehensive consideration of the recognition results of the target object by the learning-based key target object recognition model and the rule-based key target object recognition model overcomes the deficiency of the existing technology that only one key target object recognition model is considered to recognize the target, and can effectively improve the overall system robustness and functional safety level, improve the overall performance of the system, and make the recognition of the target object during autonomous driving more accurate; when the recognition results of the target object by the learning-based key target object recognition model and the rule-based key target object recognition model are different, it is chosen to wait for a reasonable time period instead of arbitrarily selecting a branch result output, and compare and judge during the waiting process, and finally select the best target object recognition result in the current calculation cycle and output it, which further improves the accuracy of target object recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0018] Figure 1 is a flow chart of a method for identifying a target object in an embodiment of the present disclosure;

[0019] Figure 2 A schematic diagram of a target object corresponding to an autonomous driving vehicle in a disclosed embodiment;

[0020] Figure 3 A schematic diagram of a system architecture for executing a target object recognition method in an embodiment of the present disclosure;

[0021] Figure 4A schematic diagram of the deployment of a basic architecture corresponding to a target object recognition method in an embodiment of the present disclosure in a chip;

[0022] Figure 5 is a flow chart of another target object recognition method in an embodiment of the present disclosure;

[0023] Figure 6 is a schematic diagram of the structure of a target object recognition device in an embodiment of the present disclosure;

[0024] Figure 7 It is a structural schematic diagram of an electronic device in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0026] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0027] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0028] Figure 1 1 is a flow chart of a method for identifying a target object in an embodiment of the present disclosure. The method can be executed by a device for identifying a target object, which can be implemented in software and / or hardware, and can be configured in an electronic device. Figure 1 As shown, the method may specifically include the following steps:

[0029] S110. Input the data of the target object acquired in the current calculation cycle into the key target object recognition model based on learning and the key target object recognition model based on rules respectively, and obtain the first recognition result and the second recognition result corresponding to the target object in the current calculation cycle, wherein the target object is an object in the surrounding environment of the autonomous driving vehicle, the first recognition result is the recognition result output by the key target object recognition model based on learning, and the second recognition result is the recognition result output by the key target object recognition model based on rules.

[0030] Figure 2is a schematic diagram of a target object corresponding to an autonomous driving vehicle in a disclosed embodiment, such as Figure 2 As shown, the target objects in this embodiment may include the forward looking target (FLT), the rear looking target (RLT), the side looking target (SLT) on the adjacent lane, and the side looking target (OSLT) on the lanes other than the adjacent lanes of the autonomous driving vehicle, wherein the forward vehicle may further include FLT1, FLT2, FLT3, FLT4, FLT5, and FLT6, and the rear vehicle may further include RLT1, RLT2, and RLT3. It can be understood that Figure 2 This is for illustration only and is not intended to be limiting. Figure 2 In addition to the movable vehicle shown, the target object in this embodiment may also include two-wheeled vehicles, pedestrians, and various objects that can be detected by the sensor system in the surrounding environment of the autonomous driving vehicle. The data of the target object is a target object associated description parameter set, which includes multiple target object associated description parameters. The target object associated description parameter refers to the parameter information corresponding to the target object and can be used to describe the motion state of the target object and the motion state of the current autonomous driving vehicle. As an example but not limited, the target object associated description parameter may include: the distance description information of the target object from the current autonomous driving vehicle, the speed description information of the target object, the position relationship information between the target object and the lane line, and the speed and running state information of the current autonomous driving vehicle. Correspondingly, the data of the target object may specifically include multiple target object associated description parameters corresponding to the target object collected at multiple sampling time points, or the data of the target object may also be a description parameter time matrix composed of multiple parameter vectors after constructing parameter vectors according to the target object associated description parameters collected at different sampling time points. Preferably, obtaining the data of the target object can specifically include: obtaining multiple target object associated description parameters collected at the current sampling time point of the target object to form a real-time description parameter vector, and obtaining a historical description parameter vector collected at a preset number of consecutive historical sampling time points of the target object, and forming a description parameter time matrix corresponding to the target object based on the real-time description parameter vector and the multiple historical description parameter vectors, and the description parameter time matrix is ​​the data of the target object.

[0031] In this embodiment, the key target object recognition model based on learning is a machine learning and deep learning model for key target object recognition. Exemplarily, the key target object recognition model based on learning can specifically include a key target object recognition model based on LSTM (Long Short-Term Memory). Different key target object recognition models based on learning can be used to identify different types of targets. Exemplarily, if the first key target object recognition model based on learning can be used to identify the types of targets as FLT1 and FLT2, and the second key target object recognition model based on learning can be used to identify the types of targets as FLT3 and FLT4, then when the target type is FLT1, the first key target object recognition model based on learning can be used for recognition, and when the target type is FLT4, the second key target object recognition model based on learning can be used for recognition. The rule-based key target object recognition model is an expert system for key target object recognition. Exemplarily, the rule-based key target object recognition model can specifically include: a key target object recognition model based on vehicle motion trajectory estimation, a key target object recognition model based on probability prediction calculation, a key target object recognition model based on grid method, and a key target object recognition model based on artificial potential field, etc.

[0032] Among them, the key target refers to the target that will interfere with the driving trajectory of the autonomous driving vehicle during the autonomous driving process, that is, the target that needs to be avoided during the driving process of the autonomous driving vehicle. It can be understood that in the same calculation cycle, the input corresponding to the key target recognition model based on learning and the input corresponding to the key target recognition model based on rules are the same. The recognition result corresponding to the key target recognition model refers to the result of identifying whether the target is a key target. Preferably, the first recognition result may include the key target and the real-time probability corresponding to the key target, the non-key target and the real-time probability corresponding to the non-key target, wherein the sum of the real-time probability corresponding to the key target and the real-time probability corresponding to the non-key target is 1. When the real-time probability corresponding to the key target is greater than or equal to the third threshold, the first recognition result is considered to be that the target is a key target, when the real-time probability corresponding to the key target is less than the third threshold and the real-time probability corresponding to the non-key target is greater than or equal to the fourth threshold, the first recognition result is considered to be that the target is a non-key target, otherwise, the first recognition result is considered to be pending. The second recognition result may include that the target is a key target or that the target is a non-key target.

[0033] S120. When the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet the preset probability requirement, execute: determining a current waiting cycle threshold corresponding to a current calculation cycle according to the first recognition result and the second recognition result.

[0034] Specifically, the first recognition result and the second recognition result may be different and may include the following four situations: the first recognition result is that the target is a key target and the second recognition result is that the target is a non-key target, the first recognition result is that the target is a non-key target and the second recognition result is that the target is a key target, the first recognition result is that the target is pending and the second recognition result is that the target is a non-key target, and the first recognition result is that the target is pending and the second recognition result is that the target is a key target. Optionally, the preset probability requirement may specifically include: the target in the first recognition result is a key target, and the real-time probability that the target in the first recognition result is a key target is greater than or equal to a first threshold; or, the target in the first recognition result is a non-key target, and the real-time probability that the target in the first recognition result is a non-key target is greater than or equal to a second threshold. Among them, the first threshold is greater than the third threshold, and the second threshold is greater than the fourth threshold. Exemplarily, the first threshold is 85%, the second threshold is 85%, the third threshold is 80%, and the fourth threshold is 80%. Accordingly, the preset probability requirement may be that the first recognition result is that the target object is a key target object (i.e., the real-time probability corresponding to the key target object in the first recognition result is greater than or equal to 80%), and the real-time probability corresponding to the key target object in the first recognition result is greater than or equal to 85%), or that the first recognition result is that the target object is a non-key target object (i.e., the real-time probability corresponding to the non-key target object in the first recognition result is greater than or equal to 80%), and the real-time probability corresponding to the non-key target object in the first recognition result is greater than or equal to 85%.

[0035] It is understandable that when the recognition result of the target object by the key target object recognition model based on learning is different from the recognition result of the key target object recognition model based on rules, if one of the recognition results is arbitrarily selected for output, it is likely to be inaccurate. At this time, it is more accurate to select a reasonable waiting time period (the reasonable time period is the current waiting period threshold in this embodiment) and compare and judge the overall situation (including the first recognition result, the second recognition result and the related probability) during the waiting process, and then select the best comprehensive recognition result of the target object corresponding to the current calculation period. The above reasonable time period is the current waiting period threshold in this embodiment. In this embodiment, when the first recognition result is that the target is a key target, the current waiting cycle threshold is related to the real-time probability corresponding to the key target in the first recognition result; when the first recognition result is that the target is a non-key target, the current waiting cycle threshold is related to the real-time probability corresponding to the non-key target in the first recognition result; when the first recognition result is that the target is pending and the second recognition result is that the target is a key target, the current waiting cycle threshold is related to the correct probability of the rule-based key target recognition model identifying the target as a key target; when the first recognition result is that the target is pending and the second recognition result is that the target is a non-key target, the current waiting cycle threshold is related to the correct probability of the rule-based key target recognition model identifying the target as a non-key target.

[0036] S130. Generate a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result, and the second recognition result.

[0037] The number of waiting cycles is the number of consecutive waiting cycles from the first calculation cycle determined to be waiting to the calculation cycle before the current calculation cycle. Preferably, during the waiting process, the number of waiting cycles is automatically increased by 1 after each calculation cycle ends.

[0038] In this embodiment, the comparison result of the number of wait cycles and the current wait cycle threshold value may specifically include: the number of wait cycles is greater than or equal to the current wait cycle threshold value, or the number of wait cycles is less than the current wait cycle threshold value. Based on the comparison result of the number of wait cycles and the current wait cycle threshold value, the first recognition result and the second recognition result, the comprehensive recognition result of the target object in the current calculation cycle is generated, which may specifically include: when the number of wait cycles is greater than or equal to the current wait cycle threshold value, it is determined that no further waiting will be continued in the current calculation cycle. At this time, the first recognition result or the second recognition result can be selected as the comprehensive recognition result of the target object in the current calculation cycle, and the number of wait cycles is cleared; when the number of wait cycles is less than the current wait cycle threshold value, it is determined that further waiting is required in the current calculation cycle. At this time, the target object being a non-critical target object can be used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of wait cycles is increased by 1.

[0039] The technical solution of this embodiment can be executed by a target object recognition system. The following describes the system architecture for executing the target object recognition method in this embodiment in conjunction with the diagram. Figure 3 Schematic diagram of the system architecture for executing the target object recognition method in the embodiment of the present disclosure. Figure 3 As shown, the system architecture is mainly composed of three parts: a key target object recognition model based on learning, a key target object recognition model based on rules, and a comprehensive identifier based on rules, wherein the comprehensive identifier based on rules is used to perform: receiving the first recognition result and the second recognition result, and generating a comprehensive recognition result of the target object in the current calculation cycle based on the first recognition result and the second recognition result. The time series data on the left side of the figure is the data of the target object in this embodiment. The target object recognition method is executed by using this system architecture. When the calculation core of any calculation branch fails (i.e., when the key target object recognition model based on learning or the key target object recognition model based on rules fails), the calculation of the other branch can still run normally, and the comprehensive identifier based on rules can still obtain relevant calculation data and output the final result, which effectively improves the robustness of target object recognition. In addition, the advantages of the key target object recognition model based on learning can also be effectively utilized, and the target recognition performance of the branch can be quickly and effectively improved through a large amount of data training. It avoids the need to continuously increase strategies, subdivide scenes and other relatively time-consuming operations when optimizing the key target object recognition model based on rules, and can effectively improve the engineering application efficiency of the target object recognition method.

[0040] Figure 4 FIG. 1 is a schematic diagram of the deployment of the system architecture for executing the target object recognition method in the embodiment of the present disclosure in a chip. Figure 4As shown, taking the SoC (System on Chip) chip as an example, the key target object recognition model based on learning is deployed in a CPU (Central Processing Unit) core of the SoC, and the key target object recognition model based on rules and the comprehensive identifier based on rules are respectively deployed in two independent MCU (Microcontroller Unit) cores of the SoC. In the current calculation cycle, the key target object recognition model based on learning and the key target object recognition model based on rules simultaneously receive the data of the same target object, and perform independent calculations in different calculation cores. The first recognition result and the second recognition result obtained are both input into the comprehensive identifier based on rules for comparison and judgment, and finally generate the comprehensive recognition result of the target object in the current calculation cycle. Exemplarily, the duration corresponding to a single calculation cycle of the SoC chip is 50ms, so within 1s, for the same target, the system architecture can have 20 calculation cycles.

[0041] The method provided in this embodiment includes inputting the data of the target object obtained in the current calculation cycle into a learning-based key target object recognition model and a rule-based key target object recognition model, respectively, to obtain a first recognition result and a second recognition result corresponding to the target object in the current calculation cycle, wherein the target object is an object in the surrounding environment of the autonomous driving vehicle, the first recognition result is a recognition result output by the learning-based key target object recognition model, and the second recognition result is a recognition result output by the rule-based key target object recognition model; when the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet the preset probability requirement, executing: determining a current waiting cycle threshold corresponding to the current calculation cycle according to the first recognition result and the second recognition result; generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result. The above-mentioned comprehensive consideration of the recognition results of the target object by the learning-based key target object recognition model and the rule-based key target object recognition model overcomes the deficiency of the existing technology that only one key target object recognition model is considered to recognize the target, and can effectively improve the overall system robustness and functional safety level, improve the overall performance of the system, and make the recognition of the target object during autonomous driving more accurate; when the recognition results of the target object by the learning-based key target object recognition model and the rule-based key target object recognition model are different, it is chosen to wait for a reasonable time period instead of arbitrarily selecting a branch result output, and compare and judge during the waiting process, and finally select the best target object recognition result in the current calculation cycle and output it, which further improves the accuracy of target object recognition.

[0042] On the basis of the above embodiments, further, determining a current waiting period threshold corresponding to a current calculation period according to the first recognition result and the second recognition result includes:

[0043] Determining a first probability according to the first recognition result and the second recognition result;

[0044] The current waiting period threshold is determined using the following formula;

[0045]

[0046] Among them, N is the current waiting period threshold, P is the first probability, and C is a preset constant.

[0047] According to the different situations of the first recognition result and the second recognition result, the first probability may specifically include: the real-time probability corresponding to the key target in the first recognition result, the real-time probability corresponding to the non-key target in the first recognition result, the correct probability that the rule-based key target recognition model recognizes the target as the key target, and the correct probability that the rule-based key target recognition model recognizes the target as the non-key target. Preferably, the value of C can be 3.4×10 -6 The current waiting cycle threshold calculated by the above formula is within a reasonable range, and will not cause the problem of inaccurate comprehensive recognition result of the target object due to the current waiting cycle threshold being too small and the waiting time being insufficient, nor will it cause the problem of wasting system computing power due to the current waiting cycle threshold being too large and the waiting time being too long. While ensuring the accuracy of target recognition, the overall performance of the system is improved.

[0048] On the basis of the above embodiments, further, determining the first probability according to the first recognition result and the second recognition result includes:

[0049] If the target object in the first recognition result is a key target object and the target object in the second recognition result is a non-key target object, the first probability is determined to be a real-time probability that the target object in the first recognition result is a key target object.

[0050] In this embodiment, when the real-time probability corresponding to the key target in the first recognition result is greater than or equal to the third threshold, the first recognition result is considered to be the target as the key target. For example, the third threshold is 80%, and the real-time probability corresponding to the key target in the first recognition result is 81%. Since 81%>80%, the first recognition result is considered to be the target as the key target. At this time, the following formula can be used to determine the current waiting cycle number;

[0051]

[0052] Among them, P LS_1 is the real-time probability that the target object in the first recognition result is a key target object (ie, the real-time probability corresponding to the key target object in the aforementioned first recognition result).

[0053] By determining different first probabilities for different situations, the current waiting period threshold can be determined in a targeted manner, which is helpful for obtaining a more accurate comprehensive recognition result of the target object later.

[0054] On the basis of the above embodiments, further, determining the first probability according to the first recognition result and the second recognition result includes:

[0055] If the target object in the first recognition result is a non-critical target object and the target object in the second recognition result is a critical target object, the first probability is determined to be a real-time probability that the target object in the first recognition result is a non-critical target object.

[0056] In this embodiment, when the real-time probability corresponding to the key target in the first recognition result is less than the third threshold and the real-time probability corresponding to the non-key target is greater than or equal to the fourth threshold, illustratively, the third threshold is 80%, the fourth threshold is 80%, the real-time probability corresponding to the key target in the first recognition result is 16%, and the real-time probability corresponding to the non-key target is 84%. Since 16% < 80% (third threshold) and 84% > 80% (fourth threshold), the first recognition result is considered to be a non-key target. At this time, the following formula can be used to determine the current number of waiting cycles;

[0057]

[0058] P LS_0 is the real-time probability that the target object in the first recognition result is a non-critical target object (ie, the real-time probability corresponding to the non-critical target object in the aforementioned first recognition result).

[0059] By determining different first probabilities for different situations, the current waiting period threshold can be determined in a targeted manner, which is helpful for obtaining a more accurate comprehensive recognition result of the target object later.

[0060] On the basis of the above embodiments, further, determining the first probability according to the first recognition result and the second recognition result includes:

[0061] If the target object in the first recognition result is pending and the target object in the second recognition result is a non-critical target object, the first probability is determined to be the correct probability that the rule-based critical target recognition model recognizes the target object as a non-critical target.

[0062] When the real-time probability corresponding to the key target in the first recognition result is less than the third threshold and the real-time probability corresponding to the non-key target is less than the fourth threshold, the first recognition result is considered to be pending. For example, the third threshold is 80%, the fourth threshold is 80%, the real-time probability corresponding to the key target in the first recognition result is 45%, and the real-time probability corresponding to the non-key target is 55%. Since 45% < 80% (third threshold) and 55% < 80% (fourth threshold), the first recognition result is considered to be pending. At this time, if the target in the second recognition result is a non-key target, the current waiting cycle number can be determined using the following formula;

[0063]

[0064] P RU_NCT The correct probability of identifying a target as a non-key target by the rule-based key target recognition model.

[0065] By determining different first probabilities for different situations, the current waiting period threshold can be determined in a targeted manner, which is helpful for obtaining a more accurate comprehensive recognition result of the target object later.

[0066] On the basis of the above embodiments, further, determining the first probability according to the first recognition result and the second recognition result includes:

[0067] If the target object in the first recognition result is pending and the target object in the second recognition result is a key target object, the first probability is determined to be a correct probability that the rule-based key target recognition model recognizes the target object as a key target object.

[0068] When the real-time probability corresponding to the key target in the first recognition result is less than the third threshold and the real-time probability corresponding to the non-key target is less than the fourth threshold, the first recognition result is considered to be pending. At this time, if the target in the second recognition result is a key target, the current waiting cycle number can be determined using the following formula:

[0069]

[0070] P RU_CT The correct probability of identifying a target as a key target by the rule-based key target recognition model.

[0071] By determining different first probabilities for different situations, the current waiting period threshold can be determined in a targeted manner, which is helpful for obtaining a more accurate comprehensive recognition result of the target object later.

[0072] On the basis of the above embodiments, further, generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result includes:

[0073] When the target object in the first recognition result is a key target object and the target object in the second recognition result is a non-key target object, executing:

[0074] When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted;

[0075] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a first preset condition, the first recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0076] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the first preset condition, the second recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0077] The first preset condition is that within the number of waiting cycles, the real-time probability that the target object in the first recognition results of the number of waiting cycles is a key target object continues to increase.

[0078] For example, when the target object in the first recognition result is a critical target object and the target object in the second recognition result is a non-critical target object, in the current calculation cycle, if the number of wait cycles is 3 and the current waiting cycle threshold is 4, since 3<4 (the number of wait cycles is less than the current waiting cycle threshold), it can be determined that the comprehensive recognition result in the current calculation cycle is that the target object is a non-critical target, and the number of wait cycles is increased by 1, that is, the number of wait cycles becomes 4, and for the next calculation cycle, the number of wait cycles is 4.

[0079] In the current calculation cycle, if the number of waiting cycles is 4, the current waiting cycle threshold is 3, and within the 4 waiting calculation cycles, the real-time probability that the target object in the first recognition result is a key target is 80%, 81%, 82% and 83% respectively (still taking the third threshold of 80% as an example), since 4>3 (that is, the number of waiting cycles is greater than the current waiting cycle threshold) and 80%<81%<82%<83% (that is, within the 4 waiting calculation cycles, the real-time probability that the target object in the first recognition result is a key target continues to increase), the first recognition result corresponding to the current calculation cycle can be used as the comprehensive recognition result of the target object in the current calculation cycle, that is, the comprehensive recognition result of the target object in the current calculation cycle is that the target object is a key target, and the number of waiting cycles 4 is cleared.

[0080] In the current calculation cycle, if the number of waiting cycles is 4, the current waiting cycle threshold is 3, and within the 4 waiting calculation cycles, the real-time probability that the target object in the first recognition result is a key target is 81%, 80%, 81% and 82% respectively (still taking the third threshold of 80% as an example), although 4>3 (that is, the number of waiting cycles is greater than the current waiting cycle threshold), but within the 4 waiting calculation cycles, the condition that the real-time probability of the target object in the first recognition result being a key target continues to increase is not met. At this time, the second recognition result corresponding to the current calculation cycle can be used as the comprehensive recognition result of the target object in the current calculation cycle, that is, the comprehensive recognition result of the target object in the current calculation cycle is that the target object is a non-key target, and the number of waiting cycles 4 is cleared.

[0081] After determining the waiting time, this embodiment compares and determines within the current calculation cycle the number of waiting cycles and the current waiting cycle threshold, as well as whether the real-time probability that the target object in the first recognition result is a key target object continues to increase. Finally, the best target object recognition result of the current calculation cycle is selected and output, thereby further improving the accuracy of target object recognition.

[0082] On the basis of the above embodiments, further, generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result includes:

[0083] When the target object in the first recognition result is a non-critical target object and the target object in the second recognition result is a critical target object, executing:

[0084] When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted;

[0085] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a second preset condition, taking the first recognition result corresponding to the current calculation cycle as the comprehensive recognition result of the target object in the current calculation cycle, and clearing the number of waiting cycles;

[0086] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the second preset condition, the second recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0087] The second preset condition is that within the number of waiting cycles, the real-time probability that the target object in the first recognition results for the number of waiting cycles is a non-critical target object continues to increase.

[0088] For example, when the target object in the first recognition result is a non-critical target object and the target object in the second recognition result is a critical target object, in the current calculation cycle, if the number of wait cycles is 3 and the current waiting cycle threshold is 4, since 3<4 (the number of wait cycles is less than the current waiting cycle threshold), it can be determined that the comprehensive recognition result in the current calculation cycle is that the target object is a non-critical target, and the number of wait cycles is increased by 1, that is, the number of wait cycles becomes 4, and for the next calculation cycle, the number of wait cycles is 4.

[0089] In the current calculation cycle, if the number of wait cycles is 4, the current waiting cycle threshold is 3, and within the 4 waiting calculation cycles, the real-time probability that the target object in the first recognition result is a non-critical target is 80%, 81%, 82% and 83% respectively (still taking the fourth threshold of 80% as an example), since 4>3 (that is, the number of wait cycles is greater than the current waiting cycle threshold) and 80%<81%<82%<83% (that is, within the 4 waiting calculation cycles, the real-time probability that the target object in the first recognition result is a non-critical target continues to increase), the first recognition result corresponding to the current calculation cycle can be used as the comprehensive recognition result of the target in the current calculation cycle, that is, the comprehensive recognition result of the target in the current calculation cycle is that the target object is a non-critical target, and the number of wait cycles 4 is cleared.

[0090] In the current calculation cycle, if the number of waiting cycles is 4, the current waiting cycle threshold is 3, and within the 4 waiting calculation cycles, the real-time probability that the target object in the first recognition result is a non-critical target is 81%, 80%, 81% and 82% respectively (still taking the fourth threshold of 80% as an example), although 4>3 (that is, the number of waiting cycles is greater than the current waiting cycle threshold), but within the 4 waiting calculation cycles, the condition that the real-time probability of the target object in the first recognition result being a non-critical target continues to increase is not met. At this time, the second recognition result corresponding to the current calculation cycle can be used as the comprehensive recognition result of the target object in the current calculation cycle, that is, the comprehensive recognition result of the target object in the current calculation cycle is that the target object is a critical target, and the number of waiting cycles 4 is cleared.

[0091] After determining the waiting time, this embodiment compares and determines within the current calculation cycle the number of waiting cycles and the current waiting cycle threshold, as well as whether the real-time probability that the target object in the first recognition result is a non-critical target object continues to increase. Finally, the best target object recognition result of the current calculation cycle is selected and output, thereby further improving the accuracy of target object recognition.

[0092] On the basis of the above embodiments, further, generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result includes:

[0093] When the target object in the first recognition result is pending and the target object in the second recognition result is a key target object, or when the target object in the first recognition result is pending and the target object in the second recognition result is a non-key target object, execute:

[0094] When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted;

[0095] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a third preset condition, the target object is regarded as a key target object as a comprehensive identification result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0096] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the third preset condition, the target object is regarded as a non-critical target object as a comprehensive identification result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0097] Among them, the third preset condition is: within the number of waiting cycles, the cumulative number of times the target object is a key target in the second recognition results of the number of waiting cycles is greater than or equal to the cumulative number of times the target object is a non-key target in the second recognition results.

[0098] For example, when the target object in the first recognition result is pending and the target object in the second recognition result is a critical target object, or, when the target object in the first recognition result is pending and the target object in the second recognition result is a non-critical target object, in the current calculation cycle, if the number of wait cycles is 3 and the current waiting cycle threshold is 4, since 3<4 (the number of wait cycles is less than the current waiting cycle threshold), it can be determined that the comprehensive recognition result in the current calculation cycle is that the target object is a non-critical target, and the number of wait cycles is increased by 1, that is, the number of wait cycles becomes 4, and for the next calculation cycle, the number of wait cycles is 4.

[0099] In the current calculation cycle, if the number of wait cycles is 4, the current wait cycle threshold is 3, and within the 4 calculation cycles that have been waited, the cumulative number of times the target object in the second recognition result is a key target is 3 times, and the cumulative number of times the target object in the second recognition result is a non-key target is 1 time, since 4>3 (that is, the number of wait cycles is greater than the current wait cycle threshold) and 3>1 (that is, within the 4 calculation cycles that have been waited, the cumulative number of times the target object in the second recognition result is a key target is greater than the cumulative number of times the target object is a non-key target in the second recognition result), then the target object being a key target can be used as the comprehensive recognition result of the target in the current calculation cycle, and the number of wait cycles 4 can be cleared to zero.

[0100] In the current calculation cycle, if the number of wait cycles is 4, the current waiting cycle threshold is 3, and within the 4 waiting calculation cycles, the cumulative number of times the target object in the second recognition result is a critical target is 1, and the cumulative number of times the target object in the second recognition result is a non-critical target is 3. Although 4>3 (that is, the number of wait cycles is greater than the current waiting cycle threshold), within the 4 waiting calculation cycles, since 1<3, the cumulative number of times the target object in the second recognition result is a critical target is not greater than or equal to the cumulative number of times the target object is a non-critical target in the second recognition result is not satisfied. In this case, the target object being a non-critical target can be used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of wait cycles 4 can be cleared to zero.

[0101] After determining the waiting, this embodiment compares the number of waiting cycles and the current waiting cycle threshold, as well as the cumulative number of times the target object in the second recognition result is a key target object and the cumulative number of times the target object in the second recognition result is a non-key target object within the current calculation cycle, and finally selects the best target object recognition result in the current calculation cycle and outputs it, thereby further improving the accuracy of target object recognition.

[0102] On the basis of the above embodiments, further, the method also includes:

[0103] When the first recognition result and the second recognition result are the same, the first recognition result or the second recognition result is used as the comprehensive recognition result, and if the number of waited cycles exists, the number of waited cycles is cleared.

[0104] For example, if it is determined to continue waiting in the previous calculation cycle, and the number of waiting cycles is determined to be 4, and in the current calculation cycle, it is determined that the first recognition result and the second recognition result are the same, at this time, the first recognition result or the second recognition result can be used as the comprehensive recognition result, and the number of waiting cycles 4 can be cleared.

[0105] It can be understood that when the recognition result of the target object by the learning-based key target object recognition model and the recognition result of the target object by the rule-based key target object recognition model are the same, it can be determined that the recognition results of both are accurate. At this time, selecting any one recognition result output can ensure the accuracy of the comprehensive recognition result.

[0106] On the basis of the above embodiments, further, the method also includes:

[0107] When the first recognition result and the second recognition result are different and the real-time probability in the first recognition result meets the preset probability requirement, the first recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle. If the number of waiting cycles exists, the number of waiting cycles is cleared.

[0108] For example, the first threshold is 85%, the second threshold is 85%, the third threshold is 80%, and the fourth threshold is 80%. If in the current calculation cycle, the target object in the first recognition result is a key target object and the target object in the second recognition result is a non-key target object, and the real-time probability corresponding to the key target object in the first recognition result is 86%, since 86%>85% (satisfying the conditions in the preset probability requirements: the target object in the first recognition result is a key target, and the real-time probability of the target object in the first recognition result being a key target is greater than or equal to the first threshold), therefore, the first recognition result corresponding to the current calculation cycle can be used as the comprehensive recognition result of the target object in the current calculation cycle, that is, the comprehensive recognition result of the target object in the current calculation cycle is that the target object is a key target. If there is a waiting cycle number, the number of waiting cycles will be cleared.

[0109] If in the current calculation cycle, the target object in the first recognition result is a non-critical target object and the target object in the second recognition result is a critical target object, and the real-time probability corresponding to the non-critical target in the first recognition result is 86%, since 86%>85% (satisfying the conditions in the preset probability requirements: the target object in the first recognition result is a non-critical target, and the real-time probability of the target object in the first recognition result being a non-critical target is greater than or equal to the second threshold), therefore, the first recognition result corresponding to the current calculation cycle can be used as the comprehensive recognition result of the target object in the current calculation cycle, that is, the comprehensive recognition result of the target object in the current calculation cycle is that the target object is a non-critical target, and if there is a waiting cycle number, the number of waiting cycles will be cleared.

[0110] It should be noted that the steps of this embodiment can be performed before determining the current waiting cycle threshold, or can be performed after determining the size of the number of waiting cycles and the current waiting cycle threshold, and are not specifically limited here. This embodiment directly determines the comprehensive recognition result of the target object by determining that the real-time probability in the first recognition result meets the preset probability requirement, thereby simplifying the judgment process and improving the recognition efficiency of the target object.

[0111] Figure 5 is a flow chart of another target object recognition method in an embodiment of the present disclosure, such as Figure 5 As shown, P LS_1 is the real-time probability that the target object in the first recognition result is a key target object, T1 is the third threshold, P LS_0 is the real-time probability that the target object in the first recognition result is a non-key target object, T0 is the fourth threshold, Nc is the real-time probability P of the target object in the first recognition result being a key target object LS_1 The current waiting cycle threshold Nnc is calculated to be the real-time probability P of the target object being a non-critical target in the first recognition result. LS_0The current waiting cycle threshold is calculated, Mc is the correct probability P of identifying the target as a key target using the rule-based key target recognition model RU_CT The current waiting cycle threshold calculated, Mnc is the correct probability P of identifying the target as a non-critical target using the rule-based key target recognition model RU_NCT The current waiting period threshold is calculated, TD1 is the first threshold, and TD0 is the second threshold. Figure 5 In the above, Lst: key target refers to the first recognition result that the target is a key target, and Exp: non-key target refers to the second recognition result that the target is a non-key target. Cycle count satisfaction means that the number of waiting cycles is greater than or equal to the current waiting cycle threshold. It should be noted that in this embodiment, the judgment of the preset probability requirement can be performed before determining the current waiting cycle threshold, or after determining the size of the number of waiting cycles and the current waiting cycle threshold. Figure 5 The determination of the preset probability requirement is performed after determining the size of the number of waiting cycles and the current waiting cycle threshold.

[0112] Figure 6 FIG. 1 is a schematic diagram of a target object recognition device according to an embodiment of the present disclosure. Figure 6 As shown: the device includes: a recognition result acquisition module 610, a current waiting cycle threshold determination module 620 and a comprehensive recognition result determination module 630, wherein:

[0113] The recognition result acquisition module 610 is used to input the data of the target object acquired in the current calculation cycle into the key target object recognition model based on learning and the key target object recognition model based on rules, respectively, to obtain a first recognition result and a second recognition result corresponding to the target object in the current calculation cycle, wherein the target object is an object in the surrounding environment of the autonomous driving vehicle, the first recognition result is a recognition result output by the key target object recognition model based on learning, and the second recognition result is a recognition result output by the key target object recognition model based on rules;

[0114] The current waiting cycle threshold determination module 620 is configured to, when the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet the preset probability requirement, perform: determining the current waiting cycle threshold corresponding to the current calculation cycle according to the first recognition result and the second recognition result;

[0115] The comprehensive recognition result determination module 630 is used to generate a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result.

[0116] The target object recognition device provided in the embodiment of the present disclosure can execute the steps in the target object recognition method provided in the method embodiment of the present disclosure, and the execution steps and beneficial effects are no longer repeated here.

[0117] Figure 7 Schematic diagram of the structure of an electronic device in the embodiment of the present disclosure. Figure 7 , which shows a structural schematic diagram of an electronic device 500 suitable for implementing the embodiments of the present disclosure. Figure 7 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0118] like Figure 7 As shown, the electronic device 500 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes to implement the method of the embodiment described in the present disclosure according to the program stored in the read-only memory (ROM) 502 or the program loaded from the storage device 508 to the random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0119] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart, thereby implementing the positioning method as described above. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0120] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0121] The computer-readable medium may be included in the electronic device, or may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device executes the method of the present application.

[0122] Optionally, when the above one or more programs are executed by the electronic device, the electronic device may also execute other steps described in the above embodiments.

[0123] 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.

[0124] Solution 1: A method for identifying a target object, the method comprising:

[0125] Inputting the data of the target object acquired in the current calculation cycle into the key target object recognition model based on learning and the key target object recognition model based on rules respectively, to obtain a first recognition result and a second recognition result corresponding to the target object in the current calculation cycle, wherein the target object is an object in the surrounding environment of the autonomous driving vehicle, the first recognition result is a recognition result output by the key target object recognition model based on learning, and the second recognition result is a recognition result output by the key target object recognition model based on rules;

[0126] When the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet the preset probability requirement, executing:

[0127] Determine a current waiting period threshold corresponding to a current calculation period according to the first recognition result and the second recognition result;

[0128] A comprehensive recognition result of the target object in the current calculation cycle is generated according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result.

[0129] Solution 2, according to the method of Solution 1, determining a current waiting cycle threshold corresponding to a current calculation cycle according to the first recognition result and the second recognition result, including:

[0130] Determining a first probability according to the first recognition result and the second recognition result;

[0131] The current waiting period threshold is determined using the following formula;

[0132]

[0133] Among them, N is the current waiting period threshold, P is the first probability, and C is a preset constant.

[0134] Solution 3: According to the method of Solution 2, determining the first probability according to the first recognition result and the second recognition result includes:

[0135] If the target object in the first recognition result is a key target object and the target object in the second recognition result is a non-key target object, the first probability is determined to be a real-time probability that the target object in the first recognition result is a key target object.

[0136] Solution 4: According to the method of Solution 2, determining the first probability according to the first recognition result and the second recognition result includes:

[0137] If the target object in the first recognition result is a non-critical target object and the target object in the second recognition result is a critical target object, the first probability is determined to be a real-time probability that the target object in the first recognition result is a non-critical target object.

[0138] Solution 5. According to the method of Solution 2, determining a first probability according to the first recognition result and the second recognition result includes:

[0139] If the target object in the first recognition result is pending and the target object in the second recognition result is a non-critical target object, the first probability is determined to be the correct probability that the rule-based critical target recognition model recognizes the target object as a non-critical target.

[0140] Solution 6: According to the method of Solution 2, determining a first probability according to the first recognition result and the second recognition result includes:

[0141] If the target object in the first recognition result is pending and the target object in the second recognition result is a key target object, the first probability is determined to be a correct probability that the rule-based key target recognition model recognizes the target object as a key target object.

[0142] Solution 7, according to the method of Solution 1, generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result, including:

[0143] When the target object in the first recognition result is a key target object and the target object in the second recognition result is a non-key target object, executing:

[0144] When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted;

[0145] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a first preset condition, the first recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0146] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the first preset condition, the second recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0147] The first preset condition is that within the number of waiting cycles, the real-time probability that the target object in the first recognition results of the number of waiting cycles is a key target object continues to increase.

[0148] Solution 8. According to the method of Solution 1, generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result, includes:

[0149] When the target object in the first recognition result is a non-critical target object and the target object in the second recognition result is a critical target object, executing:

[0150] When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted;

[0151] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a second preset condition, taking the first recognition result corresponding to the current calculation cycle as the comprehensive recognition result of the target object in the current calculation cycle, and clearing the number of waiting cycles;

[0152] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the second preset condition, the second recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0153] The second preset condition is that within the number of waiting cycles, the real-time probability that the target object in the first recognition results for the number of waiting cycles is a non-critical target object continues to increase.

[0154] Solution 9. According to the method of Solution 1, generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result, includes:

[0155] When the target object in the first recognition result is pending and the target object in the second recognition result is a key target object, or when the target object in the first recognition result is pending and the target object in the second recognition result is a non-key target object, execute:

[0156] When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted;

[0157] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a third preset condition, the target object is regarded as a key target object as a comprehensive identification result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0158] When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the third preset condition, the target object is regarded as a non-critical target object as a comprehensive identification result of the target object in the current calculation cycle, and the number of waiting cycles is cleared;

[0159] Among them, the third preset condition is: within the number of waiting cycles, the cumulative number of times the target object is a key target in the second recognition results of the number of waiting cycles is greater than or equal to the cumulative number of times the target object is a non-key target in the second recognition results.

[0160] Solution 10. The method according to Solution 1, further comprising:

[0161] When the first recognition result and the second recognition result are the same, the first recognition result or the second recognition result is used as the comprehensive recognition result, and if the number of waited cycles exists, the number of waited cycles is cleared.

[0162] Solution 11. According to the method of Solution 1, the preset probability requirement includes:

[0163] The target object in the first recognition result is a key target object, and a real-time probability that the target object in the first recognition result is a key target object is greater than or equal to a first threshold;

[0164] Alternatively, the target object in the first recognition result is a non-critical target object, and the real-time probability that the target object in the first recognition result is a non-critical target object is greater than or equal to a second threshold.

[0165] Solution 12. The method according to Solution 1, further comprising:

[0166] When the first recognition result and the second recognition result are different and the real-time probability in the first recognition result meets the preset probability requirement, the first recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle. If the number of waiting cycles exists, the number of waiting cycles is cleared.

[0167] Solution 13: A device for identifying a target object, the device comprising:

[0168] A recognition result acquisition module, used to input the data of the target object acquired in the current calculation cycle into the key target object recognition model based on learning and the key target object recognition model based on rules, respectively, to obtain a first recognition result and a second recognition result corresponding to the target object in the current calculation cycle, wherein the target object is an object in the surrounding environment of the autonomous driving vehicle, the first recognition result is a recognition result output by the key target object recognition model based on learning, and the second recognition result is a recognition result output by the key target object recognition model based on rules;

[0169] a current waiting cycle threshold determination module, configured to, when the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet a preset probability requirement, perform: determining a current waiting cycle threshold corresponding to a current calculation cycle according to the first recognition result and the second recognition result;

[0170] The comprehensive recognition result determination module is used to generate a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result.

[0171] Solution 14. An electronic device, comprising:

[0172] one or more processors;

[0173] A storage device for storing one or more programs;

[0174] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of schemes 1-12.

[0175] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

Claims

1. A method for identifying a target object, characterized in that: The method comprises: Inputting the data of the target object acquired in the current calculation cycle into the key target object recognition model based on learning and the key target object recognition model based on rules respectively, to obtain a first recognition result and a second recognition result corresponding to the target object in the current calculation cycle, wherein the target object is an object in the surrounding environment of the autonomous driving vehicle, the first recognition result is a recognition result output by the key target object recognition model based on learning, and the second recognition result is a recognition result output by the key target object recognition model based on rules; When the first recognition result and the second recognition result are different and the real-time probability in the first recognition result does not meet the preset probability requirement, perform the following operations: Determine a current waiting period threshold corresponding to a current calculation period according to the first recognition result and the second recognition result; A comprehensive recognition result of the target object in the current calculation cycle is generated according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result.

2. The method according to claim 1, characterized in that Determining a current waiting period threshold corresponding to a current calculation period according to the first recognition result and the second recognition result includes: Determining a first probability according to the first recognition result and the second recognition result; The current waiting period threshold is determined using the following formula; Among them, N is the current waiting period threshold, P is the first probability, and C is a preset constant.

3. The method according to claim 2, characterized in that The determining a first probability according to the first recognition result and the second recognition result includes: If the target object in the first recognition result is a key target object and the target object in the second recognition result is a non-key target object, determining the first probability as a real-time probability that the target object in the first recognition result is a key target object; The determining a first probability according to the first recognition result and the second recognition result includes: If the target object in the first recognition result is a non-critical target object and the target object in the second recognition result is a critical target object, determining the first probability as a real-time probability that the target object in the first recognition result is a non-critical target object; The determining a first probability according to the first recognition result and the second recognition result includes: If the target object in the first recognition result is pending and the target object in the second recognition result is a non-key target object, determining the first probability as a correct probability that the rule-based key target recognition model recognizes the target object as a non-key target; The determining a first probability according to the first recognition result and the second recognition result includes: If the target object in the first recognition result is pending and the target object in the second recognition result is a key target object, the first probability is determined to be a correct probability that the rule-based key target recognition model recognizes the target object as a key target object.

4. The method according to claim 1, characterized in that: The step of generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result comprises: When the target object in the first recognition result is a key target object and the target object in the second recognition result is a non-key target object, perform the following operations: When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted; When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a first preset condition, the first recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared; When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the first preset condition, the second recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared; The first preset condition is that within the number of waiting cycles, the real-time probability that the target object in the first recognition results of the number of waiting cycles is a key target object continues to increase.

5. The method according to claim 1, characterized in that The step of generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result comprises: When the target object in the first recognition result is a non-critical target object and the target object in the second recognition result is a critical target object, perform the following operations: When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted; When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a second preset condition, taking the first recognition result corresponding to the current calculation cycle as the comprehensive recognition result of the target object in the current calculation cycle, and clearing the number of waiting cycles; When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the second preset condition, the second recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle, and the number of waiting cycles is cleared; The second preset condition is that within the number of waiting cycles, the real-time probability that the target object in the first recognition results for the number of waiting cycles is a non-critical target object continues to increase.

6. The method according to claim 1, characterized in that The step of generating a comprehensive recognition result of the target object in the current calculation cycle according to the comparison result of the number of waiting cycles and the current waiting cycle threshold, the first recognition result and the second recognition result comprises: When the target object in the first recognition result is pending and the target object in the second recognition result is a key target object, or when the target object in the first recognition result is pending and the target object in the second recognition result is a non-key target object, perform the following operations: When the number of wait cycles is less than the current wait cycle threshold, determining that the comprehensive identification result in the current calculation cycle is that the target object is a non-critical target object, and the number of wait cycles continues to be counted; When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and satisfies a third preset condition, the target object is regarded as a key target object as a comprehensive identification result of the target object in the current calculation cycle, and the number of waiting cycles is cleared; When the number of waiting cycles is greater than or equal to the current waiting cycle threshold and does not meet the third preset condition, the target object is regarded as a non-critical target object as a comprehensive identification result of the target object in the current calculation cycle, and the number of waiting cycles is cleared; Among them, the third preset condition is: within the number of waiting cycles, the cumulative number of times the target object is a key target in the second recognition results of the number of waiting cycles is greater than or equal to the cumulative number of times the target object is a non-key target in the second recognition results.

7. The method according to claim 1, characterized in that The method further includes: When the first recognition result and the second recognition result are the same, the first recognition result or the second recognition result is used as the comprehensive recognition result, and if the number of waited cycles exists, the number of waited cycles is cleared.

8. The method according to claim 1, characterized in that: The preset probability requirements include: The target object in the first recognition result is a key target object, and a real-time probability that the target object in the first recognition result is a key target object is greater than or equal to a first threshold; Alternatively, the target object in the first recognition result is a non-critical target object, and a real-time probability that the target object in the first recognition result is a non-critical target object is greater than or equal to a second threshold.

9. The method according to claim 1, characterized in that: The method further includes: When the first recognition result and the second recognition result are different and the real-time probability in the first recognition result meets the preset probability requirement, the first recognition result corresponding to the current calculation cycle is used as the comprehensive recognition result of the target object in the current calculation cycle. If the number of waiting cycles exists, the number of waiting cycles is cleared.

10. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 9.

Images