Environmental adaptive method, device, medium and roadside perception and computing system

Abstract

The present invention discloses an environmental adaptation method, including: traversing the computing requirements of roadside computing equipment, deploying multiple groups of algorithms for the same computing requirement; configuring algorithm strategies for multiple groups of algorithms corresponding to different computing requirements according to the environmental factors of the equipment; after the configuration is completed, obtaining the requirements to be calculated, running the multiple groups of algorithms corresponding to the requirements to be calculated, and obtaining the algorithm results and confidences corresponding to each group of algorithms; obtaining the computing results of the requirements to be calculated according to the algorithm results and confidences of the multiple groups of algorithms. The present invention configures an algorithm combination for each computing requirement, deploys multiple groups of different algorithms with the same computing objectives to cope with different equipment deployment environments or weather environments, and gives priority to algorithms that better match the actual environment, effectively solving the problems of poor equipment applicability, easy degradation of accuracy, and inability to take into account specific scenarios when the same set of algorithms is used in the roadside perception and computing system in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of vehicle networking and artificial intelligence technology, and in particular to an environment adaptation method, device, medium and roadside perception and computing system. Background Art

[0002] In the application of Internet of Vehicles and intelligent transportation, single sensing devices, such as lidar, millimeter-wave radar, cameras, etc., all have their own shortcomings and are difficult to apply to all scenarios.

[0003] The existing roadside perception and computing system intelligently fuses the data obtained from different perception devices to achieve complementary advantages. Specifically, first, the reference points are calibrated within the perception range of the sensor, the coordinate system conversion relationship between multiple sensors is found, and the spatial synchronization between different sensors is achieved; then, a unique clock source is set to provide the same reference time for each device, and each device calibrates the clock time of its own system according to the reference time to achieve time synchronization between different devices; then, the data from different sensors that have achieved time and space synchronization are fused; finally, machine vision, point cloud segmentation, target detection, classification and tracking algorithms are used to extract the target or detect traffic events to form structured data. There are differences in the algorithms used by different devices, but generally the same device uses the same set of algorithms in different deployment environments, different lighting conditions, different visibility, and different meteorological conditions. The equipment has poor applicability. Once the deployment location, lighting conditions, and climate conditions change, the accuracy rate will deteriorate sharply, and it is even more impossible to take into account specific scenarios. Summary of the invention

[0004] The embodiments of the present invention provide an environment adaptation method, device, medium and roadside perception and computing system to solve the problems of poor equipment applicability, easy degradation of accuracy and inability to take into account specific scenarios when the same set of algorithms are used in the roadside perception and computing system in the prior art.

[0005] An environment adaptation method, the method comprising:

[0006] Traverse the computing requirements of roadside computing equipment and deploy multiple sets of algorithms for the same computing requirement;

[0007] Configure algorithm strategies for multiple groups of algorithms corresponding to different computing requirements based on the environmental factors of the device;

[0008] After the configuration is completed, the requirements to be calculated are obtained, and multiple groups of algorithms corresponding to the requirements to be calculated are run to obtain the algorithm results and confidence levels corresponding to each group of algorithms;

[0009] The calculation results of the requirements to be calculated are obtained according to the algorithm results and confidence levels of the multiple groups of algorithms.

[0010] Optionally, the environmental factors include equipment deployment factors and weather factors;

[0011] The algorithm strategies for configuring multiple groups of algorithms corresponding to different computing requirements according to the environmental factors of the device include:

[0012] For each computing requirement, obtaining multiple groups of algorithms corresponding to the computing requirement;

[0013] Traverse each set of algorithms and configure the algorithm strategy corresponding to the algorithm according to the equipment deployment factor and weather factor;

[0014] Among them, the algorithm strategy corresponding to each algorithm includes an algorithm parameter set and an algorithm weight.

[0015] Optionally, before configuring algorithm strategies for multiple groups of algorithms corresponding to different computing requirements according to environmental factors of the device, the method includes:

[0016] Setting threshold intervals for environmental factors of the equipment respectively;

[0017] When training multiple groups of algorithms corresponding to the same computing requirement, the environmental factors are combined according to different threshold intervals to obtain at least one environmental factor combination set;

[0018] Classify the preset training samples according to the environmental factor combination set to obtain a training sample subset corresponding to each environmental factor combination set;

[0019] For each combination of environmental factors, the corresponding training sample subset is used to train and optimize the parameters of each group of algorithms.

[0020] Optionally, after completing the training and parameter tuning of the algorithm, the method includes:

[0021] For each combination of environmental factors, calculate the average accuracy of the algorithm results of each group of algorithms;

[0022] Obtaining an algorithm weight of the algorithm according to the average accuracy;

[0023] Among them, the higher the average accuracy, the greater the algorithm weight.

[0024] Optionally, after calculating the average accuracy of the algorithm results of each group of algorithms for each combination set of environmental factors, the method further includes:

[0025] Traverse each group of algorithms and compare the average accuracy of the algorithm results with the preset accuracy threshold;

[0026] If the average accuracy is less than the preset accuracy threshold, acquiring perception information collected by the roadside perception device from the actual environment;

[0027] The algorithm is adaptively trained using the perception information.

[0028] Optionally, the equipment deployment factor includes equipment installation height and equipment pitch angle;

[0029] The weather factors include quantitative indicators of light, quantitative indicators of visibility, and meteorological types.

[0030] Optionally, when the demand to be calculated is a target classification, obtaining the calculation result of the demand to be calculated according to the algorithm results and confidence levels of the multiple groups of algorithms includes:

[0031] Calculate the election factor of each group of algorithms, where the election factor is the product of the confidence level corresponding to the algorithm and the algorithm weight;

[0032] If the target types of the outputs of multiple groups of algorithms corresponding to the requirements to be calculated are all different, the algorithm corresponding to the election factor with the largest value is obtained as the target algorithm, the target type obtained by election is the target type output by the target algorithm, and the confidence obtained by election is the confidence corresponding to the target algorithm;

[0033] If the target types output by multiple groups of algorithms corresponding to the calculation requirements are not exactly the same, the election factors of multiple algorithms that output the same target type are added together to obtain the election factor and value, and the election factors and values ​​of different target types are compared. The target type corresponding to the election factor and value with the largest numerical value is obtained as the elected target type, and the confidence of the election is the ratio of the election factor and value to the sum of the algorithm weights of the algorithms corresponding to the target type.

[0034] An environment self-adapting device, comprising:

[0035] A deployment module is used to traverse the computing requirements of roadside computing equipment and deploy multiple sets of algorithms for the same computing requirement;

[0036] A configuration module is used to configure algorithm strategies for multiple groups of algorithms corresponding to different computing requirements according to the environmental factors of the device;

[0037] The calculation module is used to obtain the requirements to be calculated after the configuration is completed, run multiple groups of algorithms corresponding to the requirements to be calculated, and obtain the algorithm results and confidence levels corresponding to each group of algorithms;

[0038] The acquisition module is used to obtain the calculation results of the requirements to be calculated according to the algorithm results and confidence levels of the multiple groups of algorithms.

[0039] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the environment adaptation method as described above is implemented.

[0040] A roadside perception and computing system comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned environment adaptation method when executing the computer program.

[0041] The embodiment of the present invention deploys multiple sets of algorithms for the same computing requirement by traversing the computing requirements of the roadside computing equipment; configures algorithm strategies for the multiple sets of algorithms corresponding to different computing requirements according to the environmental factors of the equipment; after the configuration is completed, obtains the requirements to be calculated, runs the multiple sets of algorithms corresponding to the requirements to be calculated, obtains the algorithm results and confidence; obtains the computing results of the requirements to be calculated according to the algorithm results and confidence of the multiple sets of algorithms. The present invention configures an algorithm combination for each computing requirement, deploys multiple sets of different algorithms with the same computing objectives to cope with different equipment deployment environments or weather environments, and gives priority to algorithms that better match the actual environment, effectively solving the problems of poor equipment applicability, easy degradation of accuracy, and inability to take into account specific scenarios when the same set of algorithms are used in the roadside perception and computing system in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0043] Figure 1 is a flow chart of an environment adaptation method provided by an embodiment of the present invention;

[0044] Figure 2 is a flowchart of step S102 in the environment adaptation method provided by an embodiment of the present invention;

[0045] Figure 3 It is a flowchart for implementing algorithm training and parameter tuning in the environment adaptation method provided by one embodiment of the present invention;

[0046] Figure 4 It is a flowchart for implementing the method for obtaining the calculation result of the demand to be calculated in the environment adaptation method provided by one embodiment of the present invention;

[0047] Figure 5 is a schematic diagram of the structure of an environment adaptive device provided by an embodiment of the present invention;

[0048] Figure 6 It is a schematic diagram of the composition of a roadside sensing and computing system provided by an embodiment of the present invention;

[0049] Figure 7is a schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] The embodiment of the present invention optimizes the all-weather adaptive capability of roadside computing equipment in different deployment environments, deploys multiple sets of algorithms for the same computing demand by traversing the computing requirements of the roadside computing equipment; configures algorithm strategies for multiple sets of algorithms corresponding to different computing requirements according to the environmental factors of the equipment; after the configuration is completed, obtains the requirements to be calculated, runs multiple sets of algorithms corresponding to the requirements to be calculated, and obtains algorithm results and confidence levels; obtains the calculation results of the requirements to be calculated according to the algorithm results and confidence levels of the multiple sets of algorithms; thereby effectively solving the problems of poor equipment applicability, easy degradation of accuracy, and inability to take into account specific scenarios when the same set of algorithms is used in the roadside perception and computing system in the prior art, and effectively improves the accuracy of detection, classification, and tracking of traffic participants, obstacles, and other target objects by the roadside perception and computing system under different deployment environments, different lighting conditions, different visibility, and different meteorological conditions, as well as the accuracy of traffic event detection.

[0052] The following is a detailed description of the environment adaptation method provided by this embodiment. Figure 1 As shown, the environment adaptation method includes:

[0053] In step S101, the computing requirements of the roadside computing equipment are traversed, and multiple groups of algorithms are deployed for the same computing requirement.

[0054] Here, the roadside computing device is an important component of the roadside perception and computing system. In order to provide intelligent connected vehicles with beyond-visual-range road condition perception capabilities, the roadside computing device synchronizes and fuses the information collected by the perception devices deployed on the roadside in time and space, extracts, classifies and tracks traffic participants, obstacles and other targets, detects traffic events, etc., and forms structured data. The perception devices include but are not limited to laser radar, millimeter-wave radar, camera, etc. The roadside computing device splits the above functions into multiple fine-grained computing requirements.

[0055] In the embodiment of the present invention, multiple groups of algorithms are deployed in the roadside computing equipment for the same computing requirement. The computing targets of each group of algorithms for the same computing requirement are the same, but the computing accuracy of each group of algorithms has its own advantages under different deployment environments or different lighting, visibility, and meteorological conditions. The present invention configures an algorithm combination for each computing requirement, and by deploying multiple groups of different algorithms with the same computing targets, it can effectively address the problem of accuracy degradation of a single algorithm when the equipment deployment environment is different or the weather environment changes.

[0056] In step S102, algorithm strategies are configured for multiple groups of algorithms corresponding to different computing requirements according to environmental factors of the device.

[0057] Here, the environmental factors of the equipment vary according to different computing requirements, including but not limited to equipment deployment factors and weather factors. The equipment deployment factors are basically fixed after the equipment is deployed and will not change, including but not limited to the equipment installation height and equipment pitch angle. The weather factors change in real time, including but not limited to illumination quantification indicators, visibility quantification indicators, and meteorological types. Weather factors can be obtained through sensing devices deployed on the roadside, or from a third-party meteorological platform.

[0058] As a preferred example of the present invention, Figure 2 As shown, the algorithm strategies for configuring multiple groups of algorithms corresponding to different computing requirements according to the environmental factors of the device in step S102 include:

[0059] In step S201, for each computing requirement, multiple groups of algorithms corresponding to the computing requirement are obtained.

[0060] In step S202, each group of algorithms is traversed, and the algorithm strategy corresponding to the algorithm is configured according to the equipment deployment factor and the weather factor.

[0061] Here, the embodiment of the present invention combines the input environmental factors to adapt each algorithm to an algorithm strategy that matches the current environmental factor combination, wherein the algorithm strategy corresponding to each algorithm includes an algorithm parameter set and an algorithm weight. The algorithm strategy set of multiple groups of algorithms for the same computing requirement constitutes the algorithm combination configuration strategy for the computing requirement.

[0062] Optionally, in order to avoid frequent changes in algorithm strategies, the embodiment of the present invention sets Kq threshold intervals for each environmental factor, where q represents the number of the environmental factor. Therefore, when the threshold interval corresponding to the device deployment factor or weather factor changes, the embodiment of the present invention will trigger the adjustment of the algorithm strategy.

[0063] Here, in order to reduce the complexity of algorithm training, Kq should not be set too high. For a certain computing requirement, if a certain environmental factor has little effect on the accuracy of the calculation result, the Kq corresponding to the environmental factor can be 1.

[0064] The algorithm parameter set and algorithm weight in the algorithm strategy are obtained by training and tuning the parameters of the algorithm. After deploying multiple sets of algorithms for different computing requirements, the embodiment of the present invention also trains and tunes the parameters of the multiple sets of algorithms corresponding to each computing requirement under the influence of different environmental factors. Figure 3 As shown, the training and parameter tuning of multiple groups of algorithms corresponding to each computing requirement includes:

[0065] In step S301, threshold intervals are set for environmental factors of the device respectively.

[0066] The threshold range is set according to the actual application of different environmental factors and is not limited here.

[0067] In step S302, when training multiple groups of algorithms corresponding to the same computing requirement, the environmental factors are combined according to different threshold intervals to obtain at least one environmental factor combination set.

[0068] The embodiment of the present invention combines preset environmental factors according to their respective threshold intervals to form different environmental factor combination sets. For example, environmental factor A has threshold interval A1, and environmental factor B has threshold interval B1 and threshold interval B2, then the environmental factor combination set formed by the two includes threshold interval A1, threshold interval B1, threshold interval B2, threshold interval A1+threshold interval B1, and threshold interval A1+threshold interval B2.

[0069] In step S303, the preset training samples are classified according to the environmental factor combination sets to obtain a training sample subset corresponding to each environmental factor combination set.

[0070] Here, the preset training samples are data sample sets used for training. The embodiment of the present invention classifies the training samples according to the environmental factor combination set, divides the training samples into a plurality of subsets, obtains a training sample subset corresponding to each environmental factor combination set, and is used to train and optimize parameters of the algorithm under the influence of the corresponding environmental factor combination set.

[0071] In step S304, for each combination of environmental factors, a corresponding training sample subset is used to train and optimize parameters of each group of algorithms.

[0072] Here, a set of environmental factor combinations corresponds to an application scenario of a specific environmental state, such as different deployment environments, different lighting, visibility, and environmental states determined by meteorological conditions. The training sample subset corresponding to the environmental factor combination set represents the data collected by the sensing device under the specific environmental state. For the same computing requirements, the embodiment of the present invention uses the corresponding training sample subset for each environmental factor combination set to train and optimize parameters of each group of algorithms to obtain the computing results of each group of algorithms under the specific environmental state.

[0073] Optionally, as a preferred example of the present invention, after completing the training of the algorithm and parameter tuning, the method includes:

[0074] In step S305, for each combination of environmental factors, the average accuracy of the algorithm results of each group of algorithms is calculated.

[0075] Here, for a computing requirement, under different sets of environmental factor combinations, each set of algorithms outputs an algorithm result each time it is run. The embodiment of the present invention calculates the accuracy of the algorithm result, and then calculates the average accuracy based on the accuracy of the algorithm result obtained after running the algorithm several times. Under a specific set of environmental factor combinations, each set of algorithms under a computing requirement corresponds to an average accuracy.

[0076] In step S306, the algorithm weight of the algorithm is obtained according to the average accuracy.

[0077] For the same computing requirement, the embodiment of the present invention compares the average accuracy rates corresponding to several groups of algorithms under the computing requirement, sorts them by accuracy rate, and then assigns corresponding algorithm weights. Here, the algorithm weight is used as a judgment factor for the roadside computing device to respond to the computing requirement and output the response result. Among them, the higher the average accuracy rate, the greater the algorithm weight, so as to ensure that the calculation results of the algorithm with higher accuracy rate in the training process under this environment are preferentially adopted.

[0078] Optionally, due to the specificity of the actual operating environment, the average accuracy of the algorithm results of the roadside perception and computing system in certain weather environments of certain deployment environments is lower than expected. The algorithm can be adaptively trained based on the data collected from the actual environment, and the algorithm combination configuration strategy in the specific environment can be further optimized. As a preferred example of the present invention, after calculating the average accuracy of the algorithm results of each group of algorithms for each combination of environmental factors in step S305, the method further includes:

[0079] In step S307, each group of algorithms is traversed to compare the average accuracy of the algorithm results with a preset accuracy threshold.

[0080] The preset accuracy threshold is used as a criterion for determining whether to trigger adaptive training, and can be set according to actual applications, and is not limited here. A group of algorithms corresponds to a preset accuracy threshold, and the preset accuracy thresholds corresponding to different groups of algorithms may be the same or different.

[0081] In step S308, if the average accuracy is less than the preset accuracy threshold, the perception information collected by the roadside perception device from the actual environment is obtained.

[0082] Here, if the average accuracy is less than the preset accuracy threshold, it indicates that the average accuracy of the calculation results of the roadside perception and computing system in certain weather environments in certain deployment environments is lower than expected. At this time, adaptive training is triggered to collect perception information from the actual environment.

[0083] In step S309, the algorithm is adaptively trained using the perception information.

[0084] The embodiment of the present invention utilizes the above-mentioned perception information collected from the actual environment to perform adaptive training on the algorithm, and further optimizes the algorithm combination configuration strategy under the specific environment, so as to adopt a personalized algorithm combination configuration strategy to improve the accuracy of the calculation results in the specific scenario, and then send the training results for algorithm deployment to ensure that the algorithm combination configuration strategy is more matched with the actual scenario.

[0085] Here, the results of adaptive training include but are not limited to new algorithm combinations, new environmental factor threshold interval divisions, new algorithm parameter sets, and new algorithm weight ratios, which enable maintenance and management of the algorithm combination configuration strategies deployed on all roadside computing devices to ensure that when the roadside computing equipment recovers from a failure or is replaced with a new device, the matching algorithm combination configuration strategy can be quickly obtained.

[0086] In step S103, after the configuration is completed, the requirements to be calculated are obtained, and multiple groups of algorithms corresponding to the requirements to be calculated are run to obtain the results and confidence levels of each group of algorithms.

[0087] Here, after the configuration is completed, each algorithm has been adapted to the algorithm strategy that matches the current combination of environmental factors. For the needs to be calculated, the embodiment of the present invention calls multiple groups of algorithms corresponding to the needs to be calculated, and uses the actual collected data of the perception device in the roadside perception and computing system or the output data of the previous calculation link as the algorithm input to execute the algorithm. After each group of algorithms runs, an algorithm result and a confidence level will be output. The algorithm confidence level refers to the confidence level corresponding to the algorithm result.

[0088] In step S104, the calculation results of the demand to be calculated are obtained according to the algorithm results and confidence levels of the multiple groups of algorithms.

[0089] Here, the embodiment of the present invention selects the final calculation result and confidence level based on the algorithm result, confidence level and algorithm weight output by the algorithm.

[0090] Optionally, the following description is made by taking target classification as an example. As a preferred example of the present invention, Figure 4 As shown, when the demand to be calculated is a target classification, the step S104 of obtaining the calculation result of the demand to be calculated according to the algorithm results and confidence of the multiple groups of algorithms includes:

[0091] In step S401, the election factor of each group of algorithms is calculated, where the election factor is the product of the confidence corresponding to the algorithm and the algorithm weight.

[0092] Here, it is assumed that there are a series of targets, and n groups of algorithms are matched to the target classification requirement in combination with environmental factors, and the corresponding algorithm combination configuration strategy is (algorithm i, algorithm parameter set i, algorithm weight i), i represents the algorithm number, i = 1 ~ n. When classifying one of the targets, these n groups of algorithms and the matching algorithm parameter set are used to obtain the calculation result (target type i, confidence i), the target type i represents the category to which the target object belongs calculated by algorithm i, and the confidence i is the credibility of the category to which the target object belongs calculated by algorithm i. Calculate the election factor Vi of each group of algorithms = confidence i * algorithm weight i.

[0093] In step S402, if the target types output by multiple groups of algorithms corresponding to the requirements to be calculated are all different, the algorithm corresponding to the election factor with the largest value is obtained as the target algorithm, the target type elected is the target type output by the target algorithm, and the confidence degree elected is the confidence degree corresponding to the target algorithm.

[0094] Here, if the target types calculated by all algorithms are different, the target type i calculated by the algorithm corresponding to the election factor Vi with the largest value is the elected target type, and the confidence i calculated by the algorithm is the elected confidence.

[0095] In step S403, if the target types output by multiple groups of algorithms corresponding to the calculation requirements are not exactly the same, the election factors of multiple algorithms that output the same target type are added together to obtain the election factor and value, and the election factors and values ​​of different target types are compared. The target type corresponding to the election factor and value with the largest numerical value is obtained as the elected target type, and the confidence of the election is the ratio of the election factor and value to the sum of the algorithm weights of the algorithms corresponding to the target type.

[0096] Here, if there are multiple sets of algorithms that calculate the same target type, add the Vi of the multiple sets of algorithms that calculate the same target type to obtain the election factor and value V Sum,j (j is the target type number), and the election factors and values ​​V of different target types are Sum,j Compare the largest election factor and value V Sum,j The corresponding target type j is the selected target type, and the selected confidence = V Sum,j / The calculation result is the sum of the algorithm weights of the algorithms of target type j.

[0097] For example, suppose there is a target classification that matches 3 algorithms:

[0098] The algorithm weight of Algorithm 1 is 0.6, the calculated target type is A, and the confidence is 0.95;

[0099] The algorithm weight of Algorithm 2 is 0.1, the calculated target type is A, and the confidence is 0.8;

[0100] The algorithm weight of Algorithm 3 is 0.3, and the calculated target type is B with a confidence level of 0.85.

[0101] Then the target type is the election factor and value V corresponding to A Sum,A =0.6*0.95+0.1*0.8=0.65; the election factor and value V corresponding to the target type B Sum,B =0.3*0.85=0.255. By comparison, V Sum,A >V Sum,B , so the final target type is A, confidence = V Sum,A / (0.6+0.1)=0.93, which is taken as the calculation result of the demand to be calculated.

[0102] In summary, the embodiments of the present invention deploy multiple groups of algorithms for the same computing requirement by traversing the computing requirements of the roadside computing equipment; configure algorithm strategies for the multiple groups of algorithms corresponding to different computing requirements according to the environmental factors of the equipment; after the configuration is completed, the requirements to be calculated are obtained, and the multiple groups of algorithms corresponding to the requirements to be calculated are run to obtain the algorithm results and confidence levels of each group of algorithms; and the computing results of the requirements to be calculated are obtained according to the algorithm results of the multiple groups of algorithms and their corresponding confidence levels. The present invention configures an algorithm combination for each computing requirement, and deploys multiple groups of different algorithms with the same computing objectives to cope with different equipment deployment environments or weather environments, and gives priority to algorithms that better match the actual environment, effectively solving the problems of poor equipment applicability, easy degradation of accuracy, and inability to take into account specific scenarios when the same set of algorithms is used in the roadside perception and computing system in the prior art.

[0103] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0104] In one embodiment, the present invention further provides an environment adaptation device, which corresponds one-to-one to the environment adaptation method in the above embodiment. Figure 5 As shown, the environment adaptation device includes a deployment module 51, a configuration module 52, a calculation module 53, and an acquisition module 54. The functional modules are described in detail as follows:

[0105] A deployment module 51, used to traverse the computing requirements of the roadside computing equipment and deploy multiple groups of algorithms for the same computing requirement;

[0106] Configuration module 52, used to configure algorithm strategies for multiple groups of algorithms corresponding to different computing requirements according to environmental factors of the device;

[0107] The calculation module 53 is used to obtain the requirements to be calculated after the configuration is completed, run multiple groups of algorithms corresponding to the requirements to be calculated, and obtain the algorithm results and confidence levels corresponding to each group of algorithms;

[0108] The acquisition module 54 is used to acquire the calculation results of the requirements to be calculated according to the algorithm results and confidence levels of the multiple groups of algorithms.

[0109] Optionally, the environmental factors include equipment deployment factors and weather factors;

[0110] The configuration module 52 includes:

[0111] An acquisition unit, configured to acquire, for each computing requirement, a plurality of groups of algorithms corresponding to the computing requirement;

[0112] A configuration unit, used to traverse each set of algorithms and configure an algorithm strategy corresponding to the algorithm according to a device deployment factor and a weather factor;

[0113] Among them, the algorithm strategy corresponding to each algorithm includes an algorithm parameter set and an algorithm weight.

[0114] Optionally, before configuring algorithm strategies for multiple groups of algorithms corresponding to different computing requirements according to environmental factors of the device, the apparatus includes a training and tuning module, wherein the training and tuning module 55 is used to:

[0115] Setting threshold intervals for environmental factors of the equipment respectively;

[0116] When training multiple groups of algorithms corresponding to the same computing requirement, the environmental factors are combined according to different threshold intervals to obtain at least one environmental factor combination set;

[0117] Classify the preset training samples according to the environmental factor combination set to obtain a training sample subset corresponding to each environmental factor combination set;

[0118] For each combination of environmental factors, the corresponding training sample subset is used to train and optimize the parameters of each group of algorithms.

[0119] Optionally, the training and tuning module is further used to: after completing the training and parameter tuning of the algorithm,

[0120] For each combination of environmental factors, calculate the average accuracy of the algorithm results of each group of algorithms;

[0121] Obtaining an algorithm weight of the algorithm according to the average accuracy;

[0122] Among them, the higher the average accuracy, the greater the algorithm weight.

[0123] Optionally, the training and tuning module is further used to:

[0124] Traverse each group of algorithms and compare the average accuracy of the algorithm results with the preset accuracy threshold;

[0125] If the average accuracy is less than the preset accuracy threshold, acquiring perception information collected by the roadside perception device from the actual environment;

[0126] The algorithm is adaptively trained using the perception information.

[0127] Optionally, the equipment deployment factor includes equipment installation height and equipment pitch angle;

[0128] The weather factors include quantitative indicators of light, quantitative indicators of visibility, and meteorological types.

[0129] Optionally, when the demand to be calculated is a target classification, the acquisition module 54 includes:

[0130] A calculation unit, used to calculate the election factor of each group of algorithms, where the election factor is the product of the confidence corresponding to the algorithm and the algorithm weight;

[0131] A first acquisition unit is used to acquire, if the target types of the outputs of the multiple groups of algorithms corresponding to the requirements to be calculated are different, the algorithm corresponding to the election factor with the largest value as the target algorithm, the target type obtained by election is the target type output by the target algorithm, and the confidence obtained by election is the confidence corresponding to the target algorithm;

[0132] The second acquisition unit is used to add the election factors of multiple algorithms that output the same target type to obtain the election factor and value if the target types output by multiple groups of algorithms corresponding to the calculation requirements are not exactly the same, compare the election factors and values ​​of different target types, and obtain the target type corresponding to the election factor and value with the largest numerical value as the elected target type. The confidence of the election is the ratio of the election factor and value to the sum of the algorithm weights of the algorithms corresponding to the target type.

[0133] For the specific definition of the environment adaptive device, please refer to the definition of the environment adaptive method above, which will not be repeated here. Each module in the above-mentioned environment adaptive device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0134] Optionally, as a preferred example of the present invention, Figure 6 The roadside perception and computing system provided for the embodiment of the present invention includes: a roadside computing device, a V2X platform, a laser radar, a millimeter-wave radar, a video camera, a clock synchronization device, a roadside communication unit RSU, and an intelligent connected vehicle / on-board unit OBU.

[0135] Among them, in order to provide intelligent connected vehicles with beyond-visual-range road condition perception capabilities, the roadside computing device synchronizes and fuses the information collected by the perception devices deployed on the roadside in time and space, extracts, classifies and tracks traffic participants, obstacles and other targets, detects traffic events, etc., and forms structured data. The roadside computing device splits the above functions into multiple fine-grained computing requirements, and each computing requirement can be implemented by multiple different algorithms at the same time. The computing target of each algorithm is the same, but the accuracy of the calculation has its own advantages in different deployment environments or different lighting, visibility, and meteorological conditions. In the embodiment of the present invention, multiple groups of algorithms can be deployed on the roadside computing device for the same computing requirement, and the configuration module 52, the computing module 53, and the acquisition module 54 cooperate with each other to adopt the algorithm that best matches the current actual environment for the computing requirement, so as to achieve all-weather adaptive capabilities in different deployment environments.

[0136] The V2X platform is composed of a deployment module 51 and a training and tuning module, which are used to train and deploy the algorithms deployed on the roadside computing devices. In addition, V2X can also provide information including road condition information and traffic guidance information to the intelligent networked vehicle / on-board unit OBU through 3G / 4G / 5G and other communication methods.

[0137] The sensing equipment deployed on the roadside includes but is not limited to laser radar, millimeter wave radar, camera, etc. The laser radar is used to obtain relevant information of the target to be measured, including but not limited to distance, direction, speed, shape. The millimeter wave radar is used to obtain the distance, direction, relative speed information of a single target to be measured, and to detect the average speed of traffic, traffic volume, lane occupancy, queue length and time analysis. The video camera is used to obtain audio and video information of the target to be measured.

[0138] The always-on synchronization device is used to provide the same reference time for each device in the system by setting a unique clock source, and assist each device in the system to calibrate its own clock time according to the provided reference time to achieve time synchronization.

[0139] The roadside communication unit RSU supports C-V2X communication to provide the intelligent connected vehicle / on-board unit OBU with a variety of information including road structure information, traffic participant information, obstacle information, traffic event information, sign information, traffic light state information, perception sharing messages, roadside coordination messages, etc.

[0140] The intelligent connected vehicle / on-board unit OBU supports C-V2X communication for obtaining information including road condition information, other traffic participant information and traffic guidance information from nearby roadside equipment RSU or other intelligent connected vehicles / on-board units OBU; and supports 3G / 4G / 5G communication from the V2X platform for obtaining information including road condition information and traffic guidance information; it also supports obtaining vehicle driving data through the CAN bus, including current vehicle speed, driving direction, etc.; and supports real-time acquisition of the vehicle's current position through the built-in GNSS function, and supports broadcasting the vehicle's status and driving intentions to nearby intelligent connected devices.

[0141] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an environment adaptation method is implemented.

[0142] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0143] Traverse the computing requirements of roadside computing equipment and deploy multiple sets of algorithms for the same computing requirement;

[0144] Configure algorithm strategies for multiple groups of algorithms corresponding to different computing requirements based on the environmental factors of the device;

[0145] After the configuration is completed, the requirements to be calculated are obtained, and multiple groups of algorithms corresponding to the requirements to be calculated are run to obtain the algorithm results and confidence levels corresponding to each group of algorithms;

[0146] The calculation results of the requirements to be calculated are obtained according to the algorithm results and confidence levels of the multiple groups of algorithms.

[0147] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0148] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0149] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An environment adaptation method, characterized in that: The method comprises: Traverse the computing requirements of roadside computing equipment and deploy multiple sets of algorithms for the same computing requirement; Configure algorithm strategies for multiple groups of algorithms corresponding to different computing requirements based on the environmental factors of the device; After the configuration is completed, the requirements to be calculated are obtained, and multiple groups of algorithms corresponding to the requirements to be calculated are run to obtain the algorithm results and confidence levels corresponding to each group of algorithms; Obtaining calculation results of the requirements to be calculated according to the algorithm results and confidence levels of the multiple groups of algorithms; The environmental factors include equipment deployment factors and weather factors; The algorithm strategies for configuring multiple groups of algorithms corresponding to different computing requirements according to the environmental factors of the device include: For each computing requirement, obtaining multiple groups of algorithms corresponding to the computing requirement; Traverse each set of algorithms and configure the algorithm strategy corresponding to the algorithm according to the equipment deployment factor and weather factor; Among them, the algorithm strategy corresponding to each algorithm includes an algorithm parameter set and an algorithm weight; Before configuring algorithm strategies for multiple groups of algorithms corresponding to different computing requirements according to environmental factors of the device, the method includes: Setting threshold intervals for environmental factors of the equipment respectively; When training multiple groups of algorithms corresponding to the same computing requirement, the environmental factors are combined according to different threshold intervals to obtain at least one environmental factor combination set; Classify the preset training samples according to the environmental factor combination set to obtain a training sample subset corresponding to each environmental factor combination set; For each combination of environmental factors, the corresponding training sample subset is used to train and optimize the parameters of each set of algorithms; After completing the training and parameter tuning of the algorithm, the method includes: For each combination of environmental factors, calculate the average accuracy of the algorithm results of each group of algorithms; Obtaining an algorithm weight of the algorithm according to the average accuracy; Among them, the higher the average accuracy, the greater the algorithm weight; The equipment deployment factors include equipment installation height and equipment pitch angle; The weather factors include quantitative indicators of light, quantitative indicators of visibility, and meteorological types.

2. The method for environmental adaptation according to claim 1, characterized in that: After calculating the average accuracy of the algorithm results of each group of algorithms for each combination of environmental factors, the method further includes: Traverse each group of algorithms and compare the average accuracy of the algorithm results with the preset accuracy threshold; If the average accuracy is less than the preset accuracy threshold, acquiring perception information collected by the roadside perception device from the actual environment; The algorithm is adaptively trained using the perception information.

3. The method for environmental adaptation according to claim 1, characterized in that: When the demand to be calculated is a target classification, obtaining the calculation result of the demand to be calculated according to the algorithm results and confidence levels of the multiple groups of algorithms includes: Calculate the election factor of each group of algorithms, where the election factor is the product of the confidence level corresponding to the algorithm and the algorithm weight; If the target types of the outputs of multiple groups of algorithms corresponding to the requirements to be calculated are all different, the algorithm corresponding to the election factor with the largest value is obtained as the target algorithm, the target type obtained by the election is the target type output by the target algorithm, and the confidence obtained by the election is the confidence corresponding to the target algorithm; If the target types output by multiple groups of algorithms corresponding to the calculation requirements are not exactly the same, the election factors of multiple algorithms that output the same target type are added together to obtain the election factor and value, and the election factors and values ​​of different target types are compared. The target type corresponding to the election factor and value with the largest numerical value is obtained as the elected target type, and the confidence of the election is the ratio of the election factor and value to the sum of the algorithm weights of the algorithms corresponding to the target type.

4. An environment adaptive device, characterized in that: The device comprises: A deployment module is used to traverse the computing requirements of roadside computing equipment and deploy multiple sets of algorithms for the same computing requirement; A configuration module is used to configure algorithm strategies for multiple groups of algorithms corresponding to different computing requirements according to the environmental factors of the device; The calculation module is used to obtain the requirements to be calculated after the configuration is completed, run multiple groups of algorithms corresponding to the requirements to be calculated, and obtain the algorithm results and confidence levels corresponding to each group of algorithms; An acquisition module, used for acquiring the calculation results of the requirements to be calculated according to the algorithm results and confidence levels of the multiple groups of algorithms; The environmental factors include equipment deployment factors and weather factors; The algorithm strategies for configuring multiple groups of algorithms corresponding to different computing requirements according to the environmental factors of the device include: For each computing requirement, obtaining multiple groups of algorithms corresponding to the computing requirement; Traverse each set of algorithms and configure the algorithm strategy corresponding to the algorithm according to the equipment deployment factor and weather factor; Among them, the algorithm strategy corresponding to each algorithm includes an algorithm parameter set and an algorithm weight; Before configuring algorithm strategies for multiple groups of algorithms corresponding to different computing requirements according to environmental factors of the device, the method includes: Setting threshold intervals for environmental factors of the equipment respectively; When training multiple groups of algorithms corresponding to the same computing requirement, the environmental factors are combined according to different threshold intervals to obtain at least one environmental factor combination set; Classify the preset training samples according to the environmental factor combination set to obtain a training sample subset corresponding to each environmental factor combination set; For each combination of environmental factors, the corresponding training sample subset is used to train and optimize the parameters of each set of algorithms; After completing the algorithm training and parameter tuning, including: For each combination of environmental factors, calculate the average accuracy of the algorithm results of each group of algorithms; Obtaining an algorithm weight of the algorithm according to the average accuracy; Among them, the higher the average accuracy, the greater the algorithm weight; The equipment deployment factors include equipment installation height and equipment pitch angle; The weather factors include quantitative indicators of light, quantitative indicators of visibility, and meteorological types.

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the environment adaptation method according to any one of claims 1 to 3 is implemented.

6. A roadside perception and computing system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the environment adaptation method according to any one of claims 1 to 3 is implemented.

Images