Motion monitoring method and device based on ARM industrial control motherboard
By adjusting the direction, compressing and behavior recognition of human motion data on the ARM industrial control motherboard, combining the dual-stream hybrid graph convolution network and bone recognition equipment, the problem of low accuracy in motion monitoring in vehicles is solved, and more efficient pedestrian movement is achieved Monitoring ensures safe driving.
Patent Information
- Application Number
- CN202310381272.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-11
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2043-04-11
AI Technical Summary
During vehicle driving, the accuracy of motion monitoring in the prior art is low, making it difficult to effectively monitor the movement status of pedestrians, affecting safe driving.
The motion monitoring method based on the ARM industrial control motherboard is adopted, by obtaining human body motion data, direction adjustment, spatial compression and time compression are performed, behavior recognition is used using a dual-stream hybrid graph convolution network and bone recognition equipment, and motion monitoring is performed in combination with a pre-trained model.
It improves the accuracy of motion monitoring, reduces data redundancy, enhances the real-time monitoring ability of pedestrian movement status, and ensures safe driving.
Smart Images

Figure CN116740752B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a motion monitoring method and device based on an ARM industrial control motherboard. Background Art
[0002] As people's living standards improve, the proportion of households owning cars is gradually increasing, and car driving is the issue that everyone is most concerned about. With the development of science and technology, more intelligent automation technologies have been added to vehicle driving, making it possible to achieve intelligent driving and automated driving. For example, it is possible to detect and identify vehicles, pedestrians and all obstacles, predict the position, posture or motion state of obstacles such as vehicles and pedestrians, and ensure the safe and stable operation of autonomous driving vehicles through a reasonable multi-sensor layout configuration solution.
[0003] Pedestrian motion monitoring is of utmost importance during vehicle driving, and it is necessary to always pay attention to the motion of pedestrians around to ensure safe driving. Therefore, it is urgent to propose a more accurate motion monitoring method. Summary of the invention
[0004] The present invention provides a motion monitoring method and device based on an ARM industrial control mainboard, the main purpose of which is to solve the problem of low accuracy in motion monitoring.
[0005] To achieve the above purpose, the present invention provides a motion monitoring method based on an ARM industrial control motherboard, comprising:
[0006] The ARM industrial control motherboard in the target vehicle is used to obtain a preset reference pedestrian human motion data set during motion, and the human motion data in the human motion data set is directionally adjusted according to a preset direction adjustment formula to obtain a standard motion data set. The preset direction adjustment formula is:
[0007]
[0008] Among them, R z (θ) is the direction rotation matrix, θ is the motion angle;
[0009] Projecting the standard motion data set to a preset linear subspace to perform spatial compression processing to obtain an initial compressed data set, and performing temporal compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set;
[0010] Constructing a two-stream hybrid graph convolution network according to the hetero-kernel graph convolution module and the traditional graph convolution module, inputting the final compressed data set into the two-stream hybrid graph convolution network for skeleton behavior recognition, and obtaining a first behavior recognition result;
[0011] Using a skeleton recognition device to identify the skeleton position data in the final compressed data set, performing feature extraction processing on the skeleton position data to obtain human skeleton features, and performing behavior recognition on the human skeleton data features based on a pre-trained behavior recognition model to obtain a second behavior recognition result;
[0012] The motion of a preset reference pedestrian corresponding to the target vehicle is monitored according to the first behavior recognition result and the second behavior recognition result.
[0013] Optionally, performing direction adjustment processing on the human motion data in the human motion data set according to a preset direction adjustment formula to obtain a standard motion data set includes:
[0014] Selecting any motion data in the human motion data set as target motion data, and taking the direction of the arbitrary motion data as the human motion direction;
[0015] Constructing a two-dimensional coordinate axis plane, mapping the target motion data onto the two-dimensional coordinate axis plane, and obtaining projected motion data;
[0016] Calculate the corresponding motion angle according to the projected motion data and a preset motion angle calculation formula;
[0017] The human body motion data is adjusted to move along the horizontal axis direction in the two-dimensional coordinate axis plane using the motion angle and the preset direction adjustment formula to obtain a standard motion data set.
[0018] Optionally, the preset motion angle calculation formula is:
[0019]
[0020] Among them, θ is the motion angle, XAxis represents the unit vector of the target motion data in the horizontal axis direction, v' is the projected motion data, |XAxis| is the modulus corresponding to the unit vector, and |v'| is the modulus corresponding to the projected motion data.
[0021] Optionally, projecting the standard motion data set to a preset linear subspace to perform spatial compression processing to obtain an initial compressed data set includes:
[0022] Performing dimensionality reduction processing on the standard motion data set based on the linear subspace to obtain a dimensionality reduction data set;
[0023] Identify the reconstruction accuracy corresponding to the dimension reduction data set, and determine whether the reconstruction accuracy meets the preset accuracy requirement;
[0024] If the reconstruction accuracy does not meet the preset accuracy requirement, the dimension parameter corresponding to the linear subspace is adjusted, and the standard motion data set is subjected to dimensionality reduction processing according to the linear subspace after the dimensional parameter adjustment to obtain a further dimensionality reduction data set, until the dimensionality reduction data set meets the accuracy requirement;
[0025] If the reconstruction accuracy meets the preset accuracy requirement, the dimension reduction data set is used as the initial compressed data set.
[0026] Optionally, performing time compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set includes:
[0027] Using a preset resampling algorithm to perform sampling processing on the initial compressed data set to obtain a resampled data set;
[0028] The resampled data set is interpolated, and an error threshold corresponding to the interpolated data is identified. When the error threshold meets the accuracy requirement, the resampled data set is quantized, sorted, and entropy encoded to obtain a final compressed data set.
[0029] Optionally, the performing feature extraction processing on the bone position data to obtain human bone features includes:
[0030] Performing vector processing on the bone position data using a preset vector conversion formula to obtain bone vector data;
[0031] Calculating the joint angle feature corresponding to the bone position data based on the bone vector data and a preset joint angle calculation formula;
[0032] Calculating the joint angle feature corresponding to the bone position data according to the joint angle feature and a preset joint angle calculation formula;
[0033] The joint angle features and the joint included angle features are summarized as human skeleton features.
[0034] Optionally, the preset vector conversion formula is:
[0035]
[0036]
[0037]
[0038]
[0039] in, and All are bone vector data. and represents the bone position data. m and n are preset fixed parameters.
[0040] Optionally, the preset joint angle calculation formula is:
[0041]
[0042] Among them, cosθ is the joint angle feature, is the first angle feature in the joint angle features, is the second angle feature in the joint angle feature, is the modulus of the first angle feature in the joint angle feature, It is the modulus of the second angle feature in the joint angle feature.
[0043] Optionally, inputting the final compressed data set into the two-stream hybrid graph convolutional network for skeleton behavior recognition to obtain a first behavior recognition result includes:
[0044] Extracting joint data and skeleton data from the final compressed data set, inputting the joint data and the skeleton data into the two-stream hybrid graph convolutional network to obtain a skeleton prediction result and a joint prediction result;
[0045] The skeleton prediction result and the joint prediction result are fused to obtain a final fusion result, and the final fusion result is used as the first behavior recognition result.
[0046] In order to solve the above problems, the present invention also provides a motion monitoring device based on an ARM industrial control motherboard, the device comprising:
[0047] A direction adjustment module is used to obtain a human motion data set of a preset reference pedestrian in motion by using the ARM industrial control mainboard in the target vehicle, and to perform direction adjustment processing on the human motion data in the human motion data set according to a preset direction adjustment formula to obtain a standard motion data set;
[0048] A data compression module, used for projecting the standard motion data set to a preset linear subspace to perform spatial compression processing to obtain an initial compressed data set, and performing temporal compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set;
[0049] A first recognition module is used to construct a two-stream hybrid graph convolution network according to the hetero-kernel graph convolution module and the traditional graph convolution module, input the final compressed data set into the two-stream hybrid graph convolution network for skeleton behavior recognition, and obtain a first behavior recognition result;
[0050] A second recognition module is used to use a skeleton recognition device to recognize the skeleton position data in the final compressed data set, perform feature extraction processing on the skeleton position data to obtain human skeleton features, and perform behavior recognition on the human skeleton data features based on a pre-trained behavior recognition model to obtain a second behavior recognition result;
[0051] A motion monitoring module is used to monitor the motion of a preset reference pedestrian corresponding to the target vehicle according to the first behavior recognition result and the second behavior recognition result.
[0052] The embodiment of the present invention uses the ARM industrial control mainboard in the target vehicle to obtain the human motion data set of the preset reference pedestrian during the movement process, so that the obtained data is more accurate, and the direction of the human motion data is adjusted so that the obtained standard motion data sets are all in the same direction, which is convenient for data analysis. The standard motion data set is compressed in space and time to avoid data redundancy and excessive space occupation, and the first behavior recognition and second behavior recognition processing are performed on the compressed final compressed data set respectively, and the preset reference pedestrian corresponding to the target vehicle is monitored for motion according to the first behavior recognition result and the second behavior recognition result. Therefore, the motion monitoring method and device based on the ARM industrial control mainboard proposed in the present invention can solve the problem of low accuracy in motion monitoring. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 A schematic diagram of a flow chart of a motion monitoring method based on an ARM industrial control motherboard provided in one embodiment of the present invention;
[0054] Figure 2 for Figure 1 A detailed implementation flow chart of one of the steps in the process;
[0055] Figure 3 for Figure 1 A detailed implementation flow chart of one of the steps in the process;
[0056] Figure 4 This is a functional module diagram of a motion monitoring device based on an ARM industrial control motherboard provided in one embodiment of the present invention.
[0057] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0058] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0059] The embodiment of the present application provides a motion monitoring method based on an ARM industrial control motherboard. The execution subject of the motion monitoring method based on the ARM industrial control motherboard includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiment of the present application. In other words, the motion monitoring method based on the ARM industrial control motherboard can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms.
[0060] Reference Figure 1 FIG. 1 is a flow chart of a motion monitoring method based on an ARM industrial control motherboard provided by an embodiment of the present invention. In this embodiment, the motion monitoring method based on an ARM industrial control motherboard includes:
[0061] S1. Using the ARM industrial control motherboard in the target vehicle to obtain a preset reference pedestrian human motion data set during movement, and performing direction adjustment processing on the human motion data in the human motion data set according to a preset direction adjustment formula to obtain a standard motion data set.
[0062] In the embodiment of the present invention, the ARM industrial control motherboard refers to an industrial control module, which uses an ARM architecture chip as the main control unit and also includes peripheral industrial control modules such as network, serial port, audio or video. The preset reference pedestrian refers to a human body that is walking or doing other movements around the target vehicle. The motion data generated by each body part of the preset reference pedestrian during the movement can be obtained through the ARM industrial control motherboard in the target vehicle, and the motion data is aggregated and processed to obtain a human motion data set.
[0063] Specifically, the direction adjustment processing is performed on the human motion data in the human motion data set according to a preset direction adjustment formula to obtain a standard motion data set, including:
[0064] S11, selecting any motion data in the human motion data set as target motion data, and taking the direction of the arbitrary motion data as the human motion direction;
[0065] S12, constructing a two-dimensional coordinate axis plane, mapping the target motion data onto the two-dimensional coordinate axis plane, and obtaining projected motion data;
[0066] S13, calculating the corresponding motion angle according to the projection motion data and a preset motion angle calculation formula;
[0067] S14. Using the movement angle and the preset direction adjustment formula, the human body movement data is adjusted to move along the horizontal axis direction in the two-dimensional coordinate axis plane to obtain a standard movement data set.
[0068] In detail, the two-dimensional coordinate axis plane is the xy plane, let the target motion data be v, then the projected motion data is v', and the corresponding motion angle θ is calculated according to the projected motion data v' and a preset motion angle calculation formula.
[0069] Furthermore, the preset motion angle calculation formula is:
[0070]
[0071] Among them, θ is the motion angle, XAxis represents the unit vector of the target motion data in the horizontal axis direction, v' is the projected motion data, |XAxis| is the modulus corresponding to the unit vector, and |v'| is the modulus corresponding to the projected motion data.
[0072] Specifically, the preset direction adjustment formula is:
[0073]
[0074] Among them, R z (θ) is the direction rotation matrix, and θ is the motion angle.
[0075] In detail, because the visual effects of the target human body in different directions during movement are completely different, it is necessary to adjust the direction of the human body motion data in the human body motion data set to obtain a standard motion data set to facilitate subsequent processing of the motion data in the same motion direction.
[0076] S2. Projecting the standard motion data set to a preset linear subspace to perform spatial compression processing to obtain an initial compressed data set, and performing temporal compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set.
[0077] In the embodiment of the present invention, the step of projecting the standard motion data set to a preset linear subspace for spatial compression to obtain an initial compressed data set includes:
[0078] S21, performing dimensionality reduction processing on the standard motion data set based on the linear subspace to obtain a dimensionality reduced data set;
[0079] S22, identifying the reconstruction accuracy corresponding to the dimension reduction data set, and determining whether the reconstruction accuracy meets the preset accuracy requirement;
[0080] S23, if the reconstruction accuracy does not meet the preset accuracy requirement, adjusting the dimension parameter corresponding to the linear subspace, and performing dimensionality reduction processing on the standard motion data set according to the linear subspace after the dimensional parameter adjustment to obtain a further dimensionality reduction data set, until the dimensionality reduction data set meets the accuracy requirement;
[0081] S24: If the reconstruction accuracy meets the preset accuracy requirement, the dimension reduction data set is used as the initial compressed data set.
[0082] In detail, the standard motion data set is [M1, M2, …, Mn]. The standard motion data set is projected into a linear subspace with a lower dimensionality. The dimensionality parameter is used to determine the dimension of the low-dimensional subspace, and also determines the accuracy of data reconstruction. The data is fed back to the reconstruction module after dimensionality reduction. If the data does not meet the accuracy requirements after reconstruction, the dimensionality parameter is adjusted to reduce the data again until the accuracy requirements are met.
[0083] Specifically, performing time compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set includes:
[0084] Using a preset resampling algorithm to perform sampling processing on the initial compressed data set to obtain a resampled data set;
[0085] The resampled data set is interpolated, and an error threshold corresponding to the interpolated data is identified. When the error threshold meets the accuracy requirement, the resampled data set is quantized, sorted, and entropy encoded to obtain a final compressed data set.
[0086] In detail, the error threshold is used to control data reconstruction accuracy.
[0087] S3. Construct a two-stream mixed graph convolution network according to the hetero-kernel graph convolution module and the traditional graph convolution module, input the final compressed data set into the two-stream mixed graph convolution network for skeleton behavior recognition, and obtain a first behavior recognition result.
[0088] In an embodiment of the present invention, the Exclusive Kernel Graph Convolution Block (EKB) has a more flexible and efficient convolution kernel allocation strategy compared to the traditional graph convolution, and can better adapt to the skeleton-based behavior recognition task. The traditional graph convolution block (GCB) is suitable for the extraction and aggregation of high-level semantic information. Therefore, a dual-stream hybrid graph convolution network is constructed based on the exclusive kernel graph convolution module and the traditional graph convolution module, so that the exclusive kernel graph convolution module and the traditional graph convolution module can give full play to their respective advantages in convolution layers of different depths.
[0089] Specifically, the inputting the final compressed data set into the two-stream hybrid graph convolutional network for skeleton behavior recognition to obtain a first behavior recognition result includes:
[0090] Extracting joint data and skeleton data from the final compressed data set, inputting the joint data and the skeleton data into the two-stream hybrid graph convolutional network to obtain a skeleton prediction result and a joint prediction result;
[0091] The skeleton prediction result and the joint prediction result are fused to obtain a final fusion result, and the final fusion result is used as the first behavior recognition result.
[0092] In detail, the hetero-kernel graph convolution module includes 2 hetero-kernel graph convolution units, the traditional graph convolution module includes 8 traditional graph convolution units, the output channels of each unit in the hetero-kernel graph convolution module are 8 and 8 respectively, and the number of output channels of each unit in the traditional graph convolution module is 64, 64, 128, 128, 256, 256 and 256 respectively. The joint data and the skeleton data are input into the two-stream hybrid graph convolution network, and the two-stream hybrid graph convolution network uses the hetero-kernel graph convolution module in the shallower layer of the graph convolution network and the traditional graph convolution module in the deeper layer for result prediction.
[0093] S4. Use a skeleton recognition device to identify the skeleton position data in the final compressed data set, perform feature extraction processing on the skeleton position data to obtain human skeleton features, perform behavior recognition on the human skeleton data features based on a pre-trained behavior recognition model, and obtain a second behavior recognition result.
[0094] In the embodiment of the present invention, the bone recognition device may include a depth sensor, an infrared transmitter, an RGB color camera, and a microphone array. It can perform real-time human bone tracking, voice recognition, identity recognition, etc., and can clearly identify the position of human bones, can identify 25 joints, and can simultaneously display the bone structure diagrams of six people.
[0095] Specifically, the feature extraction process is performed on the bone position data to obtain human bone features, including:
[0096] Performing vector processing on the bone position data using a preset vector conversion formula to obtain bone vector data;
[0097] Calculating the joint angle feature corresponding to the bone position data based on the bone vector data and a preset joint angle calculation formula;
[0098] Calculating the joint angle feature corresponding to the bone position data according to the joint angle feature and a preset joint angle calculation formula;
[0099] The joint angle features and the joint included angle features are summarized as human skeleton features.
[0100] In detail, the bone position data includes the three-dimensional coordinates and quaternion of each node of the human skeleton.
[0101] Furthermore, the preset vector conversion formula is:
[0102]
[0103]
[0104]
[0105]
[0106] in, and All are bone vector data. and represents the bone position data. m and n are preset fixed parameters.
[0107] Specifically, the preset joint angle calculation formula is:
[0108]
[0109]
[0110] in, is the first angle feature in the joint angle features, is the second angle feature in the joint angle feature, and Indicates joint points and Indicates joint points and Indicates joint points
[0111] Specifically, the preset joint angle calculation formula is:
[0112]
[0113] Among them, cosθ is the joint angle feature, is the first angle feature in the joint angle features, is the second angle feature in the joint angle feature, is the modulus of the first angle feature in the joint angle feature, It is the modulus of the second angle feature in the joint angle feature.
[0114] Furthermore, behavior recognition is performed on the human skeleton data features based on a pre-trained behavior recognition model to obtain a second behavior recognition result, wherein the pre-trained behavior recognition model refers to a model constructed and trained using pre-acquired big data, wherein the behavior recognition model can be a convolutional neural network model, a vector machine model or a random forest model, and the pre-acquired big data is data related to behavior.
[0115] S5. Perform motion monitoring on a preset reference pedestrian corresponding to the target vehicle according to the first behavior recognition result and the second behavior recognition result.
[0116] In an embodiment of the present invention, the first behavior recognition result and the second behavior recognition result are obtained from different angles and using different data analysis methods. The first behavior recognition result and the second behavior recognition result are used to perform motion monitoring on a preset reference pedestrian corresponding to the target vehicle, thereby improving the accuracy of motion monitoring.
[0117] The embodiment of the present invention uses the ARM industrial control mainboard in the target vehicle to obtain the human motion data set of the preset reference pedestrian during the movement process, so that the obtained data is more accurate, and the direction of the human motion data is adjusted so that the obtained standard motion data sets are all in the same direction, which is convenient for data analysis. The standard motion data set is compressed in space and time to avoid data redundancy and excessive space occupation, and the first behavior recognition and second behavior recognition processing are performed on the compressed final compressed data set respectively, and the preset reference pedestrian corresponding to the target vehicle is monitored for motion according to the first behavior recognition result and the second behavior recognition result. Therefore, the motion monitoring method and device based on the ARM industrial control mainboard proposed in the present invention can solve the problem of low accuracy in motion monitoring.
[0118] like Figure 4, which is a functional module diagram of a motion monitoring device based on an ARM industrial control motherboard provided by an embodiment of the present invention.
[0119] The motion monitoring device 100 based on the ARM industrial control motherboard of the present invention can be installed in an electronic device. According to the functions to be implemented, the motion monitoring device 100 based on the ARM industrial control motherboard can include a direction adjustment module 101, a data compression module 102, a first identification module 103, a second identification module 104 and a motion monitoring module 105. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, which are stored in the memory of the electronic device.
[0120] In this embodiment, the functions of each module / unit are as follows:
[0121] The direction adjustment module 101 is used to use the motion data acquisition device to obtain a human motion data set of a target human body during motion, and perform direction adjustment processing on the human motion data in the human motion data set according to a preset direction adjustment formula to obtain a standard motion data set;
[0122] The data compression module 102 is used to project the standard motion data set to a preset linear subspace to perform spatial compression processing to obtain an initial compressed data set, and perform temporal compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set;
[0123] The first recognition module 103 is used to construct a two-stream hybrid graph convolution network according to the hetero-kernel graph convolution module and the traditional graph convolution module, input the final compressed data set into the two-stream hybrid graph convolution network to perform skeleton behavior recognition, and obtain a first behavior recognition result;
[0124] The second recognition module 104 is used to use a skeleton recognition device to recognize the skeleton position data in the final compressed data set, perform feature extraction processing on the skeleton position data to obtain human skeleton features, and perform behavior recognition on the human skeleton data features based on a pre-trained behavior recognition model to obtain a second behavior recognition result;
[0125] The motion monitoring module 105 is used to monitor the motion of the target human body according to the first behavior recognition result and the second behavior recognition result.
[0126] In detail, each module described in the motion monitoring device 100 based on the ARM industrial control motherboard in the embodiment of the present invention is used in the same manner as described above. Figures 1 to 3 The same technical means as the motion monitoring method based on the ARM industrial control motherboard described in the previous section can produce the same technical effects, so I will not go into details here.
[0127] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.
[0128] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0129] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.
[0130] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.
[0131] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.
[0132] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
[0133] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.
[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.
Claims
1. A motion monitoring method based on ARM industrial control motherboard, It is characterized in that The method comprises: The ARM industrial control motherboard in the target vehicle is used to obtain a preset reference pedestrian human motion data set during motion, and the human motion data in the human motion data set is directionally adjusted according to a preset direction adjustment formula to obtain a standard motion data set. The preset direction adjustment formula is: in, is the direction rotation matrix, is the movement angle; The step of performing direction adjustment processing on the human motion data in the human motion data set according to a preset direction adjustment formula to obtain a standard motion data set includes: Selecting any motion data in the human motion data set as target motion data, and taking the direction of the arbitrary motion data as the human motion direction; Constructing a two-dimensional coordinate axis plane, mapping the target motion data onto the two-dimensional coordinate axis plane, and obtaining projected motion data; Calculate the corresponding motion angle according to the projected motion data and a preset motion angle calculation formula; Using the motion angle and the preset direction adjustment formula, the human body motion data is adjusted to move along the horizontal axis direction in the two-dimensional coordinate axis plane to obtain a standard motion data set; The visual effects of the target human body in different directions during the movement are completely different, so the direction of the human body movement data in the human body movement data set is adjusted to obtain a standard movement data set, which is convenient for subsequent processing of the movement data in the same movement direction; Projecting the standard motion data set to a preset linear subspace to perform spatial compression processing to obtain an initial compressed data set, and performing temporal compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set; Constructing a two-stream hybrid graph convolution network according to the hetero-kernel graph convolution module and the traditional graph convolution module, inputting the final compressed data set into the two-stream hybrid graph convolution network for skeleton behavior recognition, and obtaining a first behavior recognition result; Using a skeleton recognition device to identify the skeleton position data in the final compressed data set, performing feature extraction processing on the skeleton position data to obtain human skeleton features, and performing behavior recognition on the human skeleton data features based on a pre-trained behavior recognition model to obtain a second behavior recognition result; The step of performing feature extraction processing on the bone position data to obtain human bone features includes: Performing vector processing on the bone position data using a preset vector conversion formula to obtain bone vector data; Calculating the joint angle feature corresponding to the bone position data based on the bone vector data and a preset joint angle calculation formula; Calculating the joint angle feature corresponding to the bone position data according to the joint angle feature and a preset joint angle calculation formula; Summarizing the joint angle features and the joint included angle features into human skeleton features; The preset vector conversion formula is: in, , , and All are bone vector data. , , and represents the bone position data, and are preset fixed parameters; The preset joint angle calculation formula is: in, is the first angle feature in the joint angle features, is the second angle feature in the joint angle feature, , and Indicates joint points , , and Indicates joint points , , and Indicates joint points ; Performing motion monitoring on a preset reference pedestrian corresponding to the target vehicle according to the first behavior recognition result and the second behavior recognition result; The first behavior recognition result and the second behavior recognition result are obtained from different angles and using different data analysis methods, and the first behavior recognition result and the second behavior recognition result are used to perform motion monitoring on a preset reference pedestrian corresponding to the target vehicle, thereby improving the accuracy of motion monitoring; The preset joint angle calculation formula is: in, is the joint angle feature, is the first angle feature in the joint angle features, is the second angle feature in the joint angle feature, is the modulus of the first angle feature in the joint angle feature, It is the modulus of the second angle feature in the joint angle feature.
2. The motion monitoring method based on the ARM industrial control motherboard as claimed in claim 1, It is characterized in that The preset motion angle calculation formula is: in, is the motion angle, represents the unit vector of the target motion data in the horizontal direction, is the projection motion data, is the modulus corresponding to the unit vector, is the modulus corresponding to the projected motion data.
3. The motion monitoring method based on the ARM industrial control motherboard as claimed in claim 1, It is characterized in that The projecting the standard motion data set to a preset linear subspace to perform spatial compression processing to obtain an initial compressed data set includes: Performing dimensionality reduction processing on the standard motion data set based on the linear subspace to obtain a dimensionality reduction data set; Identify the reconstruction accuracy corresponding to the dimension reduction data set, and determine whether the reconstruction accuracy meets the preset accuracy requirement; If the reconstruction accuracy does not meet the preset accuracy requirement, the dimension parameter corresponding to the linear subspace is adjusted, and the standard motion data set is subjected to dimensionality reduction processing according to the linear subspace after the dimensional parameter adjustment to obtain a further dimensionality reduction data set, until the dimensionality reduction data set meets the accuracy requirement; If the reconstruction accuracy meets the preset accuracy requirement, the dimension reduction data set is used as the initial compressed data set.
4. The motion monitoring method based on the ARM industrial control motherboard as claimed in claim 3, It is characterized in that The step of performing time compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set includes: Using a preset resampling algorithm to perform sampling processing on the initial compressed data set to obtain a resampled data set; The resampled data set is interpolated, and an error threshold corresponding to the interpolated data is identified. When the error threshold meets the accuracy requirement, the resampled data set is quantized, sorted, and entropy encoded to obtain a final compressed data set.
5. The motion monitoring method based on the ARM industrial control motherboard as claimed in claim 1, It is characterized in that The step of inputting the final compressed data set into the two-stream hybrid graph convolutional network for skeleton behavior recognition to obtain a first behavior recognition result includes: Extracting joint data and skeleton data from the final compressed data set, inputting the joint data and the skeleton data into the two-stream hybrid graph convolutional network to obtain a skeleton prediction result and a joint prediction result; The skeleton prediction result and the joint prediction result are fused to obtain a final fusion result, and the final fusion result is used as the first behavior recognition result.
6. A motion monitoring device based on ARM industrial control motherboard, It is characterized in that The device comprises: A direction adjustment module is used to obtain a human motion data set of a preset reference pedestrian in motion by using the ARM industrial control mainboard in the target vehicle, and to perform direction adjustment processing on the human motion data in the human motion data set according to a preset direction adjustment formula to obtain a standard motion data set; The step of performing direction adjustment processing on the human motion data in the human motion data set according to a preset direction adjustment formula to obtain a standard motion data set includes: Selecting any motion data in the human motion data set as target motion data, and taking the direction of the arbitrary motion data as the human motion direction; Constructing a two-dimensional coordinate axis plane, mapping the target motion data onto the two-dimensional coordinate axis plane, and obtaining projected motion data; Calculate the corresponding motion angle according to the projected motion data and a preset motion angle calculation formula; Using the motion angle and the preset direction adjustment formula, the human body motion data is adjusted to move along the horizontal axis direction in the two-dimensional coordinate axis plane to obtain a standard motion data set; The visual effects of the target human body in different directions during the movement are completely different, so the direction of the human body movement data in the human body movement data set is adjusted to obtain a standard movement data set, which is convenient for subsequent processing of the movement data in the same movement direction; A data compression module, used for projecting the standard motion data set to a preset linear subspace to perform spatial compression processing to obtain an initial compressed data set, and performing temporal compression processing on the initial compressed data set according to a preset resampling algorithm to obtain a final compressed data set; A first recognition module is used to construct a two-stream hybrid graph convolution network according to the hetero-kernel graph convolution module and the traditional graph convolution module, input the final compressed data set into the two-stream hybrid graph convolution network for skeleton behavior recognition, and obtain a first behavior recognition result; A second recognition module is used to use a skeleton recognition device to recognize the skeleton position data in the final compressed data set, perform feature extraction processing on the skeleton position data to obtain human skeleton features, and perform behavior recognition on the human skeleton data features based on a pre-trained behavior recognition model to obtain a second behavior recognition result; The step of performing feature extraction processing on the bone position data to obtain human bone features includes: Performing vector processing on the bone position data using a preset vector conversion formula to obtain bone vector data; Calculating the joint angle feature corresponding to the bone position data based on the bone vector data and a preset joint angle calculation formula; Calculating the joint angle feature corresponding to the bone position data according to the joint angle feature and a preset joint angle calculation formula; Summarizing the joint angle features and the joint included angle features into human skeleton features; The preset vector conversion formula is: in, , , and All are bone vector data. , , and represents the bone position data, and are preset fixed parameters; The preset joint angle calculation formula is: in, is the first angle feature in the joint angle features, is the second angle feature in the joint angle feature, , and Indicates joint points , , and Indicates joint points , , and Indicates joint points ; A motion monitoring module, used for performing motion monitoring on a preset reference pedestrian corresponding to the target vehicle according to the first behavior recognition result and the second behavior recognition result; The first behavior recognition result and the second behavior recognition result are obtained from different angles and using different data analysis methods, and the first behavior recognition result and the second behavior recognition result are used to perform motion monitoring on a preset reference pedestrian corresponding to the target vehicle, thereby improving the accuracy of motion monitoring; The preset joint angle calculation formula is: in, is the joint angle feature, is the first angle feature in the joint angle features, is the second angle feature in the joint angle feature, is the modulus of the first angle feature in the joint angle feature, It is the modulus of the second angle feature in the joint angle feature.
