A method for determining the components of an object to be identified and a method for recognizing non-specific features.

Abstract

This invention relates to a method for determining components of an object to be identified and a method for recognizing non-specific features. It fuses various two-dimensional perceptual data about the object to be identified with corresponding points in an original three-dimensional point cloud. The two-dimensional perceptual data is used as attributes added to the corresponding original three-dimensional point cloud to construct an attributed three-dimensional point cloud. By utilizing the attributes of each point in the attributed three-dimensional point cloud and the correlation between these attributes, the relationships between the points in the three-dimensional point cloud are determined. Furthermore, the correspondence between the regions to which the points with correlated attributes belong and each component or related component of the 3D appearance of the object to be identified is determined. This enables the differentiation of different components or parts of the target, assists in constructing a more accurate result model, and significantly improves the accuracy of the recognition results.

Description

Technical Field

[0001] This invention relates to the field of object recognition technology, and more specifically, to a method for determining the components of an object to be identified, and a non-specific feature recognition method based on the method for determining the components of the object to be identified. Background Technology

[0002] Existing radar technologies, including lidar, operate by identifying targets using reflected waves. However, the reflected signals only contain location information. Consequently, the resulting model constructed from radar reflections (usually in the form of point clouds) is difficult to distinguish between the various components or parts of the target. This usually requires manual disassembly of the model, which is extremely labor-intensive. Moreover, the accuracy of the disassembly results is low, and they differ significantly from the actual situation of the target. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for determining the components of an object to be identified, and a non-specific feature recognition method based on the method for determining the components of the object to be identified. This method integrates two-dimensional perception data and three-dimensional point clouds to distinguish different components or parts of the target, assists in building a more accurate result model, and significantly improves the accuracy of the recognition results.

[0004] The technical solution of the present invention is as follows:

[0005] A method for determining the components of an object to be identified, comprising the following steps:

[0006] 1) Acquire the perception data of the object to be identified, wherein the perception data includes one or more two-dimensional perception data and the original three-dimensional point cloud;

[0007] 2) Use the two-dimensional sensing data as an attribute attached to the corresponding original three-dimensional point cloud, perform joint calculations, and construct an attributed three-dimensional point cloud with attributes.

[0008] 3) Determine the correspondence between each region of the attribute 3D point cloud and each component or associated component of the 3D appearance of the object to be identified.

[0009] As a preferred option, step 3) specifically involves: using the attributes of each cloud point in the attribute 3D point cloud and the correlation between the attributes of each cloud point, determining the relationship between each cloud point in the attribute 3D point cloud, and the correspondence between each region to which each cloud point with correlated attributes belongs and each component or associated component of the 3D appearance of the object to be identified.

[0010] As a preferred option, for sensing data collected sequentially from different collection locations, if the feature values ​​of one or more types of two-dimensional sensing data at different collection locations meet preset similarity conditions or are determined by a specific model to have a correlation threshold, and are unique at each collection location, then it is determined that the two-dimensional sensing data at different collection locations are correlated.

[0011] As a preferred approach, for sensing data collected simultaneously from different acquisition locations, if the same spatial field is collected from different acquisition locations, and there is only one object to be identified in the spatial field, or the collected sensing data can correctly point to one of the multiple objects to be identified, then for a certain object to be identified, one or more two-dimensional sensing data collected from different acquisition locations are correlated.

[0012] As a preferred approach, for sensing data collected simultaneously from different acquisition locations, if the same spatial field is acquired from different acquisition locations, the acquired sensing data can be calibrated using the sensing data of known objects to establish a calibration model. When performing joint calculations, the calibration model is used to handle the differences between sensing data collected simultaneously from different acquisition locations.

[0013] Preferably, for a certain type of two-dimensional sensing data of a certain object to be identified, if the features of the two-dimensional sensing data are not detected at other subsequent acquisition locations, but it can be determined from the features of other types of two-dimensional sensing data that the features of the undetected two-dimensional sensing data still belong to the object to be identified, then the features of the undetected two-dimensional sensing data will continue to be designated as belonging to the object to be identified.

[0014] As a preferred method, the acquisition device for acquiring sensing data is time-synchronized, and the acquired two-dimensional sensing data and the original three-dimensional point cloud are recorded with timestamps. Based on the timestamps, the state of the object to be identified during the acquisition interval is inferred to obtain inferred sensing data. When performing joint calculation, the inferred sensing data is assigned a different weight than the acquired sensing data.

[0015] Preferably, in step 1), the two-dimensional sensing data includes one or more combinations of color, spectral combination, contour, line, odor, sound, light reflectivity, material, motion trend, electromagnetic characteristics, temperature, temperature change trend, vibration amplitude, rigidity, vibration frequency, and vibration frequency change trend.

[0016] As a preferred approach, by combining electromagnetic induction, temperature patterns, vibration frequency variation characteristics, motion correlation, and light reflectivity, the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified is determined.

[0017] Preferably, an object image of the object to be identified is acquired, and the object image and spectrum are processed in combination with light intensity, atmospheric conditions, light direction, and object material. Based on the color and contour of the object image and spectrum, parts of the object image belonging to the same component are determined. Then, in combination with electromagnetic induction, temperature law, vibration frequency change characteristics, motion correlation, and light reflectivity, it is determined whether the cloud points in the three-dimensional point cloud are the same rigid or flexible component.

[0018] Preferably, the motion correlation includes the different motion correlations exhibited by different materials.

[0019] Preferably, in step 1), the original 3D point cloud is obtained by LiDAR, or it is generated based on image data collected by multiple image acquisition devices.

[0020] A non-specific feature recognition method based on the aforementioned component determination method is provided, comprising a peer-to-peer network including multiple node devices, all of which are independent of each other; each node device is equipped with a data acquisition device and a computing module, the data acquisition device including a lidar and at least one other type of sensor, used to acquire different corresponding types of sensing data, the sensing data including two-dimensional sensing data and raw three-dimensional point cloud; node devices located at different acquisition positions acquire at least one point sample of the object to be identified, the point sample being the raw three-dimensional point cloud and the two-dimensional sensing data of the corresponding sensor type.

[0021] For a given node device, the collected sensing data is processed to obtain result data, which is then propagated to other node devices. Other node devices that receive the result data use it as one of the original data collected, and the result data influences the result data of other node devices. Based on this, without needing to obtain the identity information of the object to be identified, multiple node devices in the peer-to-peer network perform collaborative computation to determine that each unique object to be identified is itself, thus achieving non-specific feature recognition.

[0022] Preferably, the current node device receives the result data output by other node devices; for the current node device, the collected sensing data is combined with the result data from other node devices to calculate the result data of the current node device, and then sent to other node devices; the node devices in the peer-to-peer network perform collaborative calculations as sensing data is collected and result data is calculated.

[0023] Preferably, in a peer-to-peer network, for a certain type of two-dimensional sensing data of a certain object to be identified, in the result data transmitted from the node device that collected the two-dimensional sensing data to other node devices, the subsequent node devices adjust their sensing attention based on the features of the two-dimensional sensing data, or report the features of the two-dimensional sensing data for subsequent node devices to adjust their sensing attention; if the subsequent other node devices do not detect the features of the two-dimensional sensing data, but can determine from the features of other types of two-dimensional sensing data that the undetected features of the two-dimensional sensing data still belong to the object to be identified, then the undetected features of the two-dimensional sensing data are continued to be represented in the result data of the current node device and transmitted to other node devices.

[0024] Preferably, the method for reporting the features of the two-dimensional sensing data for subsequent node devices to adjust the sensing attention is as follows: adjusting the parameters of the data processing model of the subsequent node devices based on the result data expressing the features of the two-dimensional sensing data provided by the preceding node devices, or the features of the two-dimensional sensing data, so that the subsequent node devices can improve the computing power of the subsequent node devices to identify the features of the two-dimensional sensing data; or, the subsequent node devices use the sensing attention model to match the features of the received two-dimensional sensing data or the result data expressing the features of the two-dimensional sensing data to adjust the computing power.

[0025] As a preferred embodiment, when processing the result data output by several preceding node devices, based on the data processing model, if the object to be identified described by several preceding node devices can be determined as the same target through certain common features of the two-dimensional sensing data, the features and other information of the two-dimensional sensing data described by each node device are merged into the same target.

[0026] Preferably, when the result data received by the node device indicates that the flag used by the current node device to identify the object to be identified before the current receipt of result data is different from the flag used by other node devices to identify the object to be identified, and the flags assigned to the object to be identified by other node devices have been updated, then the flag used by the current node device to identify the object to be identified before the current receipt of result data is converted.

[0027] As a preferred method, the method for converting the flag used by the current node device to identify the object to be identified before the current received result data is as follows:

[0028] Replace the flag used by the current node device to identify the object to be identified before the current reception of result data with the latest flag assigned to the object by other node devices;

[0029] Alternatively, record the conversion relationship between the flag used by the current node device to identify the object to be identified before the current reception of result data and the updated flags assigned to the object by other node devices, and perform the conversion when it is necessary to reference the result data received by the current node device in the current reception.

[0030] Alternatively, node devices can deploy conversion models to perform corresponding conversions on the labels of multiple objects to be identified based on the input raw data or result data.

[0031] Preferably, when it is determined that the identity information of the object to be identified needs to be obtained, an identity information acquisition command is triggered. The identity information acquisition command is used as one of the inputs to participate in the calculation of the result data of the node device. By driving the node device in the peer-to-peer network that is connected to the barrier-free data collection conditions that can obtain the identity information of the object to be identified to respond to the corresponding result data, the identity information of the object to be identified can be obtained.

[0032] As a preferred approach, the peer-to-peer network verifies the authenticity of the identity information of the object to be identified, thereby determining its permissions. In this approach, the node device in the peer-to-peer network that can obtain identity information does not provide the identity information itself, but only expresses the verification result in the result data of the node device based on the verification requirements for the authenticity of the identity information in the received result data.

[0033] Preferably, in a peer-to-peer network, the node device capable of obtaining identity information does not provide identity information. Instead, the information source device that drives the provision of identity information establishes an encrypted information transmission channel with the node device input terminal that needs to obtain identity information, or establishes an encrypted information transmission channel using other network communication modes, and uses the identity information as one of the inputs to the node device.

[0034] As a preferred option, a component attribute knowledge base is also set up, which records the attribute parameters corresponding to each component of a known type of target. The attribute parameters of each component in the component attribute knowledge base are combined with the attributes of the cloud points in the attribute 3D point cloud for collaborative calculation to help determine the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified.

[0035] As a preferred method, the scene is reconstructed in 3D by using images from cameras located at different acquisition positions that capture data of the same space, along with their installation positions and camera parameters, to obtain a 3D scene model. The on-site perception data is used as input to adjust the environment of the 3D scene. By comparing the states of the 3D scene model in the 3D scene at different time points and under different environments, static facilities and dynamic targets, as well as the various components of dynamic targets, are distinguished. The state data of the 3D data of the dynamic targets after 3D reconstruction in the scene under the corresponding environmental parameters are collaboratively calculated with the 3D point clouds of each attribute to determine the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified.

[0036] Among them, the camera parameters, installation location and image are used, or the camera parameters, installation location and image are combined with radar parameters, radar installation location and detection data to perform three-dimensional reconstruction of the current scene where the object to be identified is located;

[0037] When reconstructing a scene, the initial 3D scene model is constructed by using the perception data collected on-site as input, adjusting the environment of the 3D scene, and using optical, acoustic, and physical models to cross-verify the calculation results of the light and appearance state in the scene with the result data of each node device. The scene is further optimized based on the verification results.

[0038] As a preferred option, the perception data collected on-site serves as the input to the optical model, acoustic model, and physical model in the 3D scene model after the scene is reconstructed using the images from each camera, the installation location, and the camera parameters. The output includes one or more data such as the color, light reflectivity, shadow shape and density, and smooth surface reflection of the dynamic target. The output dynamic target data is then used in conjunction with the 3D point cloud of each attribute to determine the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified.

[0039] One or more of the attribute 3D point cloud, physical parameters of the clearly defined target in the knowledge base, and on-site collected data such as illumination, weather, air quality, air composition, and wind speed are used as training parameters to accumulate samples and update the training of the optical model, acoustic model, and physical model.

[0040] The beneficial effects of this invention are as follows:

[0041] The component determination method for an object to be identified described in this invention fuses various two-dimensional perceptual data about the object to be identified with corresponding points in the original three-dimensional point cloud. The two-dimensional perceptual data is used as attributes added to the corresponding original three-dimensional point cloud to construct an attributed three-dimensional point cloud. By utilizing the attributes of each point in the attributed three-dimensional point cloud and the correlations between these attributes, the relationships between the points in the three-dimensional point cloud are determined, as well as the correspondence between the regions to which the points with correlated attributes belong and each component or related component of the 3D appearance of the object to be identified. Furthermore, when constructing a three-dimensional result model from the attributed three-dimensional point cloud, components inferred to be of the same type are constructed as independent component models based on the attributes of the points in the attributed three-dimensional point cloud; ultimately, a result model with independent component models is obtained. In the result model of this invention, different components can be naturally distinguished through independent component models, greatly reducing the workload of model splitting. Moreover, the independent component models have high accuracy in corresponding to the actual situation of the target, resulting in a more precise result model.

[0042] The non-specific feature recognition method described in this invention, based on the component determination method of the object to be identified and with a more accurate result model, can significantly improve the accuracy of the recognition results.

[0043] This invention utilizes a peer-to-peer network for non-specific feature identification of objects. In this network, there is no hierarchy among nodes, no fixed connection paths, and each node only receives calculation results from other nodes and transmits its own results. The detection of events and / or the response of corresponding execution devices do not rely on a single node for identification and control, but rather on collaborative calculations and confirmation by multiple nodes in the network. Without requiring specific features or specific identity information, each unique object can be identified as itself, achieving non-specific feature identification. This invention achieves high accuracy and precise location identification through non-specific feature identification for object recognition, identity verification, or event monitoring. This invention can identify and verify objects without relying on specific features, protecting privacy while simultaneously addressing issues related to transportation, education, healthcare, epidemic prevention, public services, emergency response, public security, counter-terrorism, community management and services, market behavior, production safety, and civilized behavior.

[0044] This invention employs non-specific feature recognition, effectively preventing risks caused by theft or counterfeiting of specific features, thus significantly enhancing security. It utilizes a non-contact, passive method for seamless identification of the object, greatly improving ease of execution. Based on the aforementioned peer-to-peer network, this invention can be easily deployed over coverage areas ranging from hundreds of meters to hundreds of kilometers, making it suitable for various geographical areas. Detailed Implementation

[0045] The present invention will be further described in detail below with reference to the embodiments.

[0046] To address the shortcomings of existing technologies, such as the need for manual model decomposition, which is labor-intensive and results in low accuracy, this invention provides a method for determining the components of an object to be identified. Based on this component determination method, a non-specific feature recognition method is also provided. This invention combines two-dimensional sensing data with the original three-dimensional point cloud, adding different attributes to the cloud points in the original three-dimensional point cloud corresponding to the radar reflection wave. This allows for the determination of different components of the object to be identified, thereby constructing a more accurate result model and significantly improving the recognition results.

[0047] The method for determining the components of an object to be identified according to the present invention mainly includes the following steps:

[0048] 1) Acquire perceptual data of the object to be identified. The perceptual data includes one or more two-dimensional perceptual data and an original three-dimensional point cloud. Typically, multiple perceptual data about the object to be identified can be acquired simultaneously, including corresponding types of two-dimensional perceptual data collected using different types of sensors, and an original three-dimensional point cloud. The original three-dimensional point cloud is obtained through LiDAR acquisition, or generated based on image data collected by multiple image acquisition devices. In this embodiment, the two-dimensional perceptual data includes one or more combinations of color, spectral combination, contour, lines, odor, sound, light reflectivity, material, motion trend, electromagnetic characteristics, temperature, temperature change trend, vibration amplitude, rigidity (derived from vibration amplitude, industrial parameters of the identified target, or known targets), vibration frequency, and vibration frequency change trend.

[0049] 2) Two-dimensional sensing data is used as an attribute appended to the corresponding original three-dimensional point cloud. Specifically, by combining the two-dimensional sensing data, its acquisition location, effective sensing distance field data, and the original three-dimensional point cloud, joint calculations are performed to construct an attributed three-dimensional point cloud. That is, corresponding two-dimensional sensing data is appended to the cloud points of the original three-dimensional point cloud. Thus, the cloud points of the attributed three-dimensional point cloud can easily distinguish cloud points belonging to different components by reflecting multiple different types of two-dimensional sensing data. In specific implementation, a cloud point of the original three-dimensional point cloud may be appended with multiple types of two-dimensional sensing data simultaneously, or multiple two-dimensional sensing data of the same type may be acquired simultaneously from different locations.

[0050] 3) Based on the attributes of the cloud points, determine the correspondence between each region of the attributed 3D point cloud (composed of cloud points with attributes) and each component of the 3D appearance of the object to be identified, or the correspondence between each region of the attributed 3D point cloud and related components (i.e., different components are related). Specifically, using the attributes of each cloud point in the attributed 3D point cloud and the correlation between the attributes of each cloud point, determine the relationship between the cloud points in the 3D point cloud, and the correspondence between each region to which each cloud point with correlated attributes belongs and each component or related component of the 3D appearance of the object to be identified. When determining the correspondence between each region of the attributed 3D point cloud and each component or related component of the 3D appearance of the object to be identified, when constructing a 3D result model of the object to be identified using the attributed 3D point cloud, the region corresponding to each component (the region of the attributed 3D point cloud) is constructed as an independent component model, or the region corresponding to a related component (the region of the attributed 3D point cloud) can also be associated and constructed as an independent component model. In this embodiment, based on the type of two-dimensional sensing data recorded in step 1), the correspondence between each region of the attribute three-dimensional point cloud and each component or related component of the 3D appearance of the object to be identified can be determined by combining electromagnetic induction (temperature and electromagnetic waves), temperature law, vibration frequency change characteristics, motion correlation, and light reflectivity.

[0051] Regarding the aforementioned correlation, in this embodiment, for sensing data collected sequentially from different collection locations, if the feature values ​​of one or more two-dimensional sensing data at different collection locations respectively meet the preset similarity conditions or are determined by a specific model to have a correlation reaching a threshold, and are unique at each collection location, then it is determined that the two-dimensional sensing data at different collection locations have a correlation.

[0052] On the other hand, for sensing data collected simultaneously from different acquisition locations, if the same spatial field is collected from different acquisition locations, and there is only one object to be identified in the spatial field, or the collected sensing data can correctly point to one of the multiple objects to be identified, then for a certain object to be identified, one or more two-dimensional sensing data collected from different acquisition locations are correlated.

[0053] When multiple types of two-dimensional sensing data or multiple identical two-dimensional sensing data are collected simultaneously from different locations on the same object to be identified, unavoidable environmental factors (including light intensity, light direction, atmospheric conditions, etc.) and equipment operating status may cause significant errors in the two-dimensional sensing data collected at different times. For example, the same component of the same object to be identified may have color differences, spectral data differences, etc., thus affecting consistency and ultimately the accuracy of the identification result. Therefore, this invention further addresses the issue that, for sensing data collected simultaneously from different locations, if the same spatial field is collected from different locations, the acquired sensing data is calibrated using the sensing data of the known object to establish a calibration model. During joint calculations, the calibration model is used to handle the differences between the sensing data collected simultaneously from different locations. For example, when three cameras at different angles, under different lighting and atmospheric conditions, photograph the same component of the same target, the resulting color differences and spectral data differences are accumulated and a calibration model is established to consider the data differences caused by positional differences when processing the two-dimensional sensing data acquired by these cameras.

[0054] Because the same component will appear differently under different light intensities, atmospheric conditions, and light directions (e.g., in the presence of shadows or reflections), traditional image recognition methods can easily misidentify a single component as multiple different components. To address this issue, this invention utilizes image, color, contour, line, and spectrum to identify the same component (especially when the object to be identified is a person and the component is an external accessory). Specifically, an image of the object to be identified is acquired, and the image and spectrum are processed in conjunction with light intensity, atmospheric conditions, light direction, and object material. Considering the influence of these factors, the color and contour of the object image and spectrum are used to determine which parts of the object image belong to the same component (i.e., which parts of the object image belong to the same component). This invention processes the object image and spectrum to obtain the correspondence between the image and spectrum of the same component under different environments, including the correspondence between color and contour, thereby determining which parts (parts of the image) belong to the same component. Then, by combining electromagnetic induction, temperature patterns, vibration frequency variation characteristics, motion correlation, and light reflectivity, it is determined whether the cloud points in the three-dimensional point cloud belong to the same rigid or flexible component. In this embodiment, the motion correlation includes the different motion correlations exhibited by different materials such as ropes and fabrics. In specific implementations, the present invention can also utilize the peer-to-peer network provided by the present invention to simultaneously use multiple sets of cameras and sensors to determine the same component from multiple targets.

[0055] To address the potential issue of excessively long time intervals between consecutive sensing data acquisitions due to the inability to continuously acquire sensing data (not necessarily all sensing data, but possibly one or more types of 2D sensing data), thus affecting the final identification result, this invention employs correlation inference to acquire speculative sensing data, thereby avoiding excessive impact on the final identification result. Specifically, for a particular type of 2D sensing data of a target object, if subsequent acquisition locations do not detect the features of that 2D sensing data, but the features of other types of 2D sensing data can determine that the undetected features of that 2D sensing data still belong to the target object, then the undetected features of that 2D sensing data are still designated as belonging to the target object. For example, if the color of target object A was acquired in the previous sensing data acquisition, but the color of target object A was not acquired in the subsequent sensing data acquisition, but sensing data acquired from other locations can determine that target object A exists in addition to other target objects, then the color of the unacquired target object A is still attached to the cloud point as an attribute.

[0056] Furthermore, the acquisition device for acquiring sensory data is time-synchronized. The acquired 2D sensory data and the original 3D point cloud are recorded with timestamps. Based on the timestamps, the state of the object to be identified during the acquisition interval is inferred (inference of the sensory data), resulting in inferred sensory data. During joint calculation, the inferred sensory data is assigned a different weight than the acquired sensory data. For example, the sensory data of the object to be identified during the acquisition interval between two sets of sensory data can be inferred based on motion inertia, or, in the case of occlusion, the sensory data of the object to be identified within the occluded area can be inferred based on physical parameters. In practice, the weight of the inferred sensory data can be chosen to be less than the weight of the acquired sensory data to avoid the inferred sensory data having an excessive impact on the recognition result. When the error between the inferred sensory data and the actual situation is large, if the weight of the inferred sensory data is too large, it will affect the accuracy of the recognition result.

[0057] Based on the component determination method described above, this invention also provides a non-specific feature recognition method for an object to be identified. The term "non-specific feature recognition," strictly defined, differs from the common understanding of "recognition." Commonly, "recognition" refers to identifying the concrete form or specific identity information of a target, such as who it is (including name, specific information indicating the target's identity), or what it is (e.g., a car, a person). However, the "recognition" in this invention refers to identifying each unique target (i.e., a user) as itself; that is, for any given object to be identified, its existence is unique. After implementing "non-specific feature recognition," this invention determines that the object to be identified (i.e., the user who has not been identified or had their identity confirmed) is itself, and not another object to be identified. The result of "non-specific feature recognition" does not require determining the specific features of the object to be identified, nor does it require determining the object's identity information or concrete form. For example, if a person is considered object A to be verified, and an object is considered object B to be verified, then after implementing "non-specific feature recognition," it is not necessary to identify whether object A is a person or what their specific identity is, nor is it necessary to identify whether object B is an object or what kind of object it is; rather, it is necessary to determine that object A is object A itself, and object B is object B itself. Then, corresponding services or controls can be provided for object A or object B.

[0058] In this invention, a peer-to-peer network is established, comprising multiple node devices. There is no hierarchy among these nodes, forming a decentralized network and computing architecture. Unlike traditional single-point aggregation computing models, the data transmission direction between node devices in this invention does not follow a fixed, predetermined path. In this peer-to-peer network, a single node device processes the collected raw data to obtain result data, which is then propagated to other node devices. Other node devices receiving the result data use it as one of their collected raw data sets, thus influencing the result data of other node devices. For ease of description, the aforementioned "single node device" is referred to as the "current node device," and the "other node devices" as "subsequent node devices." One aspect of this influence is that the result data calculated by subsequent node devices is not entirely determined by their own collected raw data, but rather by the result data output by the current node device. Specifically, the result data output by the current node device may alter the data processing model and parameters used by subsequent node devices to calculate their result data, thereby affecting the result data of subsequent node devices. For example, if the output data of the current node device is correlated with the raw data collected by subsequent node devices, it is necessary to consider the impact of the output data of the current node device on the accuracy of the output data of the subsequent node devices. Specifically, for the perception of a specific target, if the result data is calculated based solely on the raw data collected by subsequent node devices, it can only reflect the real-time (including real-time location and time) single-point result judgment of the target within the perception range of the subsequent node devices. However, the output data of the current node device reflects the direct perception data and result judgment of the target at other locations and at other times, or other indirectly related perception data and result judgments, which helps to improve the accuracy and comprehensiveness of the result data of the subsequent node devices, including superimposed calculations of the same dimension and correlation references of different dimensions.

[0059] Because there is no master-slave relationship between nodes in a peer-to-peer network, point-to-point transmission is possible. Therefore, for a given calculation result corresponding to a specific perceived data point of a target, as reflected in the output data of one node, the information is relatively symmetrical among other nodes receiving that result data. Other nodes use the received result data as input, combining it with their own sensor data to calculate their own result data. Their own result data naturally encompasses both the received result data and the information reflected by their own sensors, and is transmitted to other nodes in the next layer. Thus, for a specific perceived data point of a target, the information is relatively symmetrical across all nodes. This prevents the impact of tampering or falsification of the calculation process and results of a single node on the result data. It also serves as a means to detect faulty, tampered, or non-compliant node devices. This fundamentally solves the inherent hidden dangers of traditional information technology, namely, the false, falsified, and erroneous information caused by information asymmetry, which becomes a point of entry for fraud and cyberattacks. It also addresses the problems of poor accuracy, excessive time consumption, low credibility, and poor responsiveness in complex integrated applications. Therefore, it can truly become the information infrastructure for comprehensive management of large areas and the infrastructure for the digital economy. Unlike blockchain technology, which relies on independent computation by each node to determine the result and emphasizes the preservation of original data, this invention focuses on peer-to-peer collaborative computation among node devices. Through this collaborative computation, each node device can adjust its own data processing model (i.e., the algorithm for calculating the result data) and parameters when processing data. This adjustment is a feedback mechanism from all node devices, transforming the computation of all node devices into a unified whole. Instead of individual nodes performing calculations independently, all node devices collaboratively complete the computation. The adjustments to the node device's data processing model are objectively real and will impact subsequent data processing iterations.

[0060] This invention also provides auxiliary means for identifying components. Specifically, it establishes a component attribute knowledge base, which records the attribute parameters corresponding to each component of a known type of target. By combining the attribute parameters of each component in the component attribute knowledge base with the attributes of cloud points in the attribute 3D point cloud, a collaborative calculation is performed to assist in determining the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified. For example, the component attribute knowledge base records the color, appearance, optical refractive index, vibration performance, and rigidity parameters of the boom for a certain model of crane; and the color, appearance, optical refractive index, vibration performance, and rigidity parameters of the hoisting rope, etc. This can be used to assist in distinguishing the appearance attribution of different components corresponding to different regions in the attribute 3D point cloud. Specifically, for the identification of a specific crane model, non-specific feature recognition can be performed through the peer-to-peer network provided by this invention, combined with accessibility data collection, to jointly calculate the specific crane model for the object to be identified.

[0061] In the collaborative computing process of this invention, collaborative computing and the acquisition of sensing data are actually carried out simultaneously and continuously. To assist collaborative computing, in this embodiment, the scene is reconstructed in three dimensions using images from cameras located at different acquisition positions that capture data of the same space, along with their installation positions and camera parameters, to obtain a three-dimensional scene model. Sensing data collected on-site (i.e., detection data collected by various sensors on-site) is used as input to adjust the environment of the three-dimensional model (lighting, weather, air quality, wind speed, etc.). By comparing the states of the three-dimensional scene model in different time points and environments, static facilities and dynamic targets, as well as the various components of dynamic targets, are distinguished. The state data of the three-dimensional data of the dynamic target after 3D reconstruction in the scene, under the corresponding environmental parameters, is collaboratively calculated with the three-dimensional point clouds of each attribute to determine the correspondence between each region of the attribute three-dimensional point cloud and each component or related component of the 3D appearance of the object to be identified. Specifically, the current scene of the object to be identified is reconstructed in three dimensions using camera parameters, installation positions, and images, or by combining camera parameters, installation positions, and images with radar parameters, radar installation positions, and detection data.

[0062] When reconstructing a scene, the initial 3D scene model is constructed by using on-site collected perception data as input. The environment of the 3D scene is adjusted, and the calculated results of lighting and appearance (e.g., shadows, effects between models, such as snow) are cross-validated with the results from each node device using optical, acoustic, and physical models (e.g., optical models built based on ray tracing technology), and other relevant data. Based on the verification results, the scene is further optimized to approximate the real-world scene. Specifically, the on-site collected perception data serves as input to the optical, acoustic, and physical models in the 3D scene model after reconstruction using camera images, installation locations, and camera parameters. The output includes one or more data points such as the color, light reflectivity, shadow shape and density of dynamic targets, and reflections of smooth surfaces. The output dynamic target data is then collaboratively calculated with the 3D point clouds of various attributes to determine the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified.

[0063] To improve the recognition accuracy of optical, acoustic, and physical models, this embodiment uses one or more of the attribute 3D point cloud, physical parameters of the clearly defined target in the knowledge base, and on-site collected data on illumination, weather, air quality, air composition, and wind speed as training parameters to accumulate samples and update the optical, acoustic, and physical models. The calculation results of the updated optical, acoustic, and physical models are closer to the real situation.

[0064] The peer-to-peer network of this invention is based on collaborative computing, does not rely on single-point identification, and distributes computing functions across the entire network, reducing the hardware and software requirements of single-point computing, resulting in high execution efficiency and greatly improving anti-attack capabilities. The network maintains a relatively symmetrical information state among node devices, making it immune to illegal data tampering. Even if a single node device is physically compromised and its transmitted data is altered, the network-wide computing is a highly redundant and complex calculation with extensive multi-dimensional verification. Therefore, the alteration of data transmitted by a single node device does not affect the overall network computing results. Furthermore, it can quickly locate the faulty and tampered node device, ensuring the reliability of the overall network computing results. This, in turn, can resolve the conflict between data sharing and information security between departments.

[0065] The result data transmitted between node devices can be the processing result of information rather than the information itself. Therefore, the raw data collected (i.e., the perceived data) does not need to be stored. Node devices only receive the calculation results output by other node devices and send out their own calculation results. The amount of information contained in a single calculation result is insufficient to reconstruct any event or target information. A definite result can only be obtained by joint calculation of the calculation results of the entire peer-to-peer network, multi-dimensional data matrix elements, and physical space and facility correspondence. The collaborative calculation has less dependence on the information transmitted by a few node devices, which can fundamentally change the nature of traditional information technology's single-point security sensitivity.

[0066] In this invention, since the output data of each node device reflects the state evolution of the output data of the preceding node devices, the behavior, attributes, state, or events of the target when it was perceived by the preceding node devices can be inferred based on the output data received by the current node device. For example, when it is necessary to find the location of target 'a' 15 minutes ago, the location of the node device that perceived target 'a' can be obtained at the current moment, thus inferring the location of target 'a'. Then, based on the transmission path of the output data, it can be inferred back to 15 minutes ago to estimate the location of target 'a' 15 minutes ago (determined by the node device that perceived target 'a'). Furthermore, the node device does not need to store the original data about target 'a'. That is, based on this invention, it is not necessary to identify the original data to find target 'a', but rather to first infer the node device that perceived target 'a', and if necessary, obtain the original data about target 'a' at the time when it needs to be found from the storage device connected to the node device.

[0067] Node devices are equipped with data acquisition devices (in specific implementations, these may include one or more of the following: image acquisition devices, audio acquisition devices, temperature measurement devices, vibration frequency sensing devices, lidar, chemical sensors, electromagnetic induction devices, etc., selected to match the required type of sensing data) and a computing module. The data acquisition devices include lidar and at least one other type of sensor, used to collect sensing data of different corresponding types. The sensing data includes two-dimensional sensing data and raw three-dimensional point clouds. The computing module calculates the result data based on a data processing model. Node devices located at different acquisition positions (i.e., at different physical installation locations) collect at least one point sample of the object to be identified. The point sample consists of the raw three-dimensional point cloud and the two-dimensional sensing data of the corresponding sensor type. Based on this, without needing to obtain the identity information of the object to be identified, multiple node devices in the peer-to-peer network perform collaborative calculations to determine that each unique object to be identified is itself, achieving non-specific feature recognition.

[0068] Specifically, taking a given node device as the current node device, and considering the data transmission between its preceding and subsequent node devices (in this invention, preceding and subsequent node devices are only used to describe their sequential relationship with the current node device in the current calculation and data transmission process, and do not imply any necessary sequential or priority relationship between them), the current node device receives the result data output by other node devices (including preceding node devices), and subsequent node devices receive the result data output by other node devices (including the current node device). For the current node device, the collected sensing data is combined with the result data from other node devices (including preceding node devices) to calculate the result data of the current node device, and this result data is sent to other node devices (including subsequent node devices). Similarly, the working process of subsequent node devices is the same as that of the current node device, and preceding node devices also receive the result data from the preceding node devices of their predecessors and perform the same working process as the current node device; that is, the node devices in the peer-to-peer network perform the same working process. Furthermore, the node devices in the peer-to-peer network perform collaborative calculations as sensing data is collected and result data is calculated. In this process, the output data of a certain node device is only received and used as input by the subsequent layer of node devices, and the output data of the subsequent layer of node devices will cover the output data of the preceding layer of node devices (including the aforementioned node device).

[0069] In a peer-to-peer network, all events are processed synchronously, and it is not necessary to explicitly produce phased results such as what event was discovered or what the specific content of the event is. In a peer-to-peer network, only the sensor's perception and the corresponding execution device's response are explicit. All other intermediate processes are processed simultaneously through collaborative computing. That is, during the operation of this invention, the intermediate process of event discovery is imperceptible. As collaborative computing proceeds and the node device obtains the result data, the corresponding execution device automatically responds and executes.

[0070] To further ensure the trustworthiness of the data source and computation process, in this invention, all node devices encrypt their computational results based on an encrypted consensus mechanism, obtaining encrypted results, which are then sent to other node devices. The encrypted consensus mechanism includes one or more consensus mechanisms, with different mechanisms corresponding to changes in the encryption algorithm structure and parameters of the node devices.

[0071] Node devices communicate using standard-sized data packets (i.e., result data or calculation results). In this invention, the node devices in the peer-to-peer network are similar to human neurons. Just as each neuron does not transmit specific data directly describing external events, the node devices do not output raw data. Instead, they process the raw data acquired by connected sensors and data acquisition devices into standard-sized data packets (i.e., result data or calculation results, similar to nerve impulses in neurons) based on their own data processing model (similar to the biological characteristics of nerve cells). The information contained in a single data packet is insufficient to reconstruct any event or target information. A definite result can only be obtained through collaborative computation involving the calculation results across the entire peer-to-peer network, multi-dimensional data matrix elements, and the correspondence between physical space and facilities. Collaborative computation has little dependence on the data output by a few node devices, and it simultaneously processes all requests received or initiated by all node devices. It is a collaborative verification computation of highly multi-dimensional related information, thereby fundamentally changing the traditional single-point security sensitivity of information systems.

[0072] To ensure data integrity and the effective execution of collaborative computing, this invention deploys a QoS mechanism in the peer-to-peer network, which prioritizes the transmission quality of result data between node devices.

[0073] In practical implementation, the peer-to-peer network can be configured using one or a combination of 4G, 5G, or MESH modes to suit different application scenarios. The optimal solution is achieved by considering factors such as feasibility and cost. The MESH mode is based on the LTE standard, communicating at the LTE physical layer. Data is carried by a customized frame structure, and interaction is performed using a dedicated wireless communication protocol. Customizing the frame structure for peer-to-peer network computing and employing a proprietary wireless communication protocol developed for urban cluster peer-to-peer network computing further enhances its security and reliability. Furthermore, the wireless algorithm is fully adaptable to the multipath channel environment controlled by a consensus mechanism required for peer-to-peer network computing. Communication distances range from 100 meters to 10 kilometers within cities, and up to 120 kilometers in the field using omnidirectional antennas. In this embodiment, the Mesh network communication distance is 50-150 meters between indoor nodes and 50 meters to 120 kilometers between outdoor nodes, with each node capable of connecting to 65,535 nodes. In addition, when networking in 4G and 5G modes, there is no limit to the communication distance, and the number of node devices that can be connected depends on the computing power of the computing chip and the communication latency.

[0074] In a peer-to-peer network, for a specific type of two-dimensional sensing data of an object to be identified, the subsequent node devices adjust their perceptual attention based on the features of the two-dimensional sensing data in the result data transmitted from the node device that collected the data to other node devices. (This doesn't necessarily require the features of the two-dimensional sensing data to be included in the result data; rather, the features of the two-dimensional sensing data participate in the calculation of the preceding node device, so that the result data of the preceding node device can be used as input to the data processing model of the subsequent node device, allowing the subsequent node device's data processing model to achieve the effect of adjusting perceptual attention during calculation.) Alternatively, the features of the two-dimensional sensing data can be reported for subsequent node devices to adjust their perceptual attention (the features of the two-dimensional sensing data are directly described in the result data). If other subsequent node devices do not detect the features of the two-dimensional sensing data, but can determine from the features of other types of two-dimensional sensing data that the undetected features of the two-dimensional sensing data still belong to the object to be identified, then the undetected features of the two-dimensional sensing data are continued to be described in the result data of the current node device and transmitted to other node devices. For example, if a preceding node device senses the color of an object A to be identified, but the current node device does not sense the color of the object A to be identified, but it can be determined from the sensing data of other node devices that there is another object A to be identified besides other objects to be identified, then the color of the object A to be identified that has not been sensed will still be represented in the result data of the current node device.

[0075] In this embodiment, the method for reporting the features of the two-dimensional sensing data for subsequent node devices to adjust the sensing attention is as follows: adjusting the parameters of the data processing model of the subsequent node devices based on the result data expressing the features of the two-dimensional sensing data provided by the preceding node devices, or the features of the two-dimensional sensing data, so that the subsequent node devices can improve the computing power of the subsequent node devices to identify the features of the two-dimensional sensing data; or, the subsequent node devices use the sensing attention model to match the features of the received two-dimensional sensing data or the result data expressing the features of the two-dimensional sensing data to adjust the computing power.

[0076] The “feature” mentioned above has a different meaning from the “feature recognition” in the prior art. The “feature recognition” in the prior art usually refers to information that can determine the identity of a target, while the “feature” in this invention represents a kind of perceived data belonging to the object to be identified, such as coordinates, colors belonging to the object to be identified, etc. The “non-specific feature recognition” of the object to be identified cannot be directly completed by the “feature” perceived by a single point.

[0077] In this embodiment, the method for reporting the features of the point sample for subsequent node devices to adjust the perceptual attention is as follows: The parameters of the data processing model of the subsequent node devices are adjusted based on the result data expressing the features of the two-dimensional perceptual data provided by the preceding node device (in this invention, the features of the two-dimensional perceptual data are usually not provided themselves, but rather expressed in the result data), or the features of the two-dimensional perceptual data (i.e., the features of the two-dimensional perceptual data themselves), so that the subsequent node devices improve their computing power for identifying the features of the two-dimensional perceptual data; or, the subsequent node devices use the perceptual attention model to match the features of the received two-dimensional perceptual data or the result data expressing the features of the two-dimensional perceptual data for computing power adjustment.

[0078] When a node device processes the output data from several preceding node devices, based on a data processing model, if the objects to be identified described by several preceding node devices can be determined as the same target through certain common features of the two-dimensional sensing data, the features and other information of the two-dimensional sensing data described by each node device are merged into the same target. For example, the features of two-dimensional sensing data in physical spaces that almost completely overlap at the same time can be used to determine that they are the same target.

[0079] When the result data received by a node device indicates that the flag used by the current node device to identify the object to be identified before the current reception of result data is different from the flags used by other node devices to identify the object to be identified, and the flags assigned to the object by other node devices have been updated, then the flag used by the current node device to identify the object to be identified before the current reception of result data is converted. Specifically, the method for converting the flag used by the current node device to identify the object to be identified before the current reception of result data is as follows:

[0080] Replace the flag used by the current node device to identify the object to be identified before the current reception of result data with the latest flag assigned to the object by other node devices;

[0081] Alternatively, record the conversion relationship between the flag used by the current node device to identify the object to be identified before the current reception of result data and the updated flags assigned to the object by other node devices, and perform the conversion when it is necessary to reference the result data received by the current node device in the current reception.

[0082] Alternatively, node devices can deploy conversion models to perform corresponding conversions on the labels of multiple objects to be identified based on the input raw data or result data.

[0083] In this invention, in order to improve the effectiveness of "non-specific feature recognition", for sensing data collected successively from different collection locations, if the feature values ​​of one or more two-dimensional sensing data at different collection locations meet the preset similarity conditions or are determined by a specific model to have a correlation threshold, and are unique at each collection location, then it is determined that the two-dimensional sensing data at different collection locations are correlated.

[0084] On the other hand, for sensing data collected simultaneously from different acquisition locations, if the same spatial field is collected from different acquisition locations, and there is only one object to be identified in the spatial field, or the collected sensing data can correctly point to one of the multiple objects to be identified, then for a certain object to be identified, one or more two-dimensional sensing data collected from different acquisition locations are correlated.

[0085] In the process of "non-specific feature recognition," this invention can also acquire the identity information of the object to be identified when necessary. Specifically, when it is determined that the identity information of the object to be identified needs to be acquired, an identity information acquisition command is triggered. This command is used as one of the inputs in the calculation of the result data of the node device. By driving the node device connected to the peer-to-peer network with barrier-free data collection conditions capable of acquiring the identity information of the object to be identified to respond with the corresponding result data, the identity information of the object to be identified is acquired. The acquisition of identity information is also the result of collaborative calculation; that is, the determination that identity information needs to be acquired triggers the acquisition of identity information, rather than being additionally triggered by a specific request command. Based on this invention, if permission calculation is triggered by a request command, in most cases, it can be completed without acquiring identity information. Only in a few cases, when it is found that permission calculation cannot be completed without acquiring identity information, is the determination that identity information needs to be acquired generated according to implementation requirements. For example, if collaborative computing reveals that a person's identity information exists in several location-specific QR code registration systems, package pickup registration systems, or consumer registration systems, and prior authorization from the person or legal access to these systems is obtained, then the peer-to-peer network can drive node devices connected to these systems via barrier-free data collection. The obtained information is then transmitted to the peer-to-peer network through each node device for information comparison and to provide accurate identity information. Based on this, the present invention can also minimize the possibility of identity tampering with a system.

[0086] Specifically, the peer-to-peer network determines the permissions of the target by verifying the authenticity of the identity information. Nodes in the peer-to-peer network capable of acquiring identity information may choose not to provide the identity information (or may provide it depending on implementation requirements), but instead express the verification result in their own result data based solely on the verification requirements for the identity information's authenticity found in the received result data. In other words, in this invention, even when a node capable of acquiring identity information does not provide it, the verification result is expressed in its own result data based solely on the verification requirements for the identity information's authenticity found in the received result data.

[0087] When a node device in a peer-to-peer network that can obtain identity information does not provide identity information, the information source device that drives the provision of identity information establishes an encrypted file transmission channel with the input terminal of the node device that needs to obtain identity information, or establishes an encrypted information transmission channel using other network communication modes; and uses the identity information as one of the inputs of the node device.

[0088] When necessary, in order to meet the needs of other traditional computing modes for raw data, such as the need for evidence preservation in traditional evidence presentation, in this embodiment, the node device can be equipped with a data storage device for storing the raw data sensed by the sensor.

[0089] In practical implementation, the node device can also be equipped with leakage protection and other functions in its power supply. The node device can also provide various communication interfaces, including fiber optic interfaces and wireless communication interfaces; it can also provide a data interface for connecting external storage devices. The node device can be powered by solar energy or mains power. When implemented outdoors, the node device can be installed on poles such as streetlights (without crossarms, mounted on the main pole, or integrated into the lampshade); in pole-less areas, if implemented indoors, it can be wall-mounted or integrated into the ceiling.

[0090] When this invention is implemented indoors and outdoors, the node devices, as artificial intelligence facilities installed in public spaces, can serve as digital economy infrastructure for urban clusters, providing 24 / 7 seamless coverage. Through collaborative computing across node devices, vehicle identification at any location within the coverage area can achieve near 100% accuracy, with location identification accuracy related to sensor accuracy.

[0091] In this invention, the architecture of a peer-to-peer computing network consists of nodes of the same type and function. Each node dynamically adjusts its data processing model in real time according to the network's consensus mechanism. The raw data collected by the data acquisition devices (including sensors, cameras, etc.) connected to each node is processed and encrypted by the node according to its own data processing model, generating byte-level processing and encryption results (i.e., result data). This result data is then sent to other node devices (the computational and encryption results received by the current node from other node devices are also considered part of the raw data collected by the current node). Therefore, the effect of the raw data sensed by each sensor will propagate exponentially among a massive number of peer-to-peer node devices. If each node sends its result data to 100 surrounding node devices, after four units of time, hundreds of millions of node devices will be affected by the event sensed by that sensor. In this computing model, information is relatively symmetrical and immune to tampering and forgery. It fundamentally solves the inherent hidden dangers of traditional information technology, namely, the false, forged, and erroneous information caused by information asymmetry, which in turn become entry points for fraud and cyberattacks, as well as the problems of long cycles, poor accuracy, and poor adaptability in complex and integrated applications. In turn, it truly becomes an information infrastructure for comprehensive management of large areas and a digital economy infrastructure.

[0092] The above embodiments are merely illustrative of the present invention and are not intended to limit the invention. Any changes or modifications to the above embodiments based on the technical essence of the present invention will fall within the scope of the claims of the present invention.

Claims

1. A method for determining components of an object to be identified, characterized in that, The steps are as follows: 1) Acquire the perception data of the object to be identified, wherein the perception data includes one or more two-dimensional perception data and the original three-dimensional point cloud; 2) Use the two-dimensional sensing data as an attribute attached to the corresponding original three-dimensional point cloud, perform joint calculations, and construct an attributed three-dimensional point cloud with attributes. 3) Based on the attributes of each cloud point in the attribute 3D point cloud, each region of the attribute 3D point cloud is constructed by the cloud points with attributes. The correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified is determined. Specifically, by using the attributes of each cloud point in the attribute 3D point cloud and the correlation between the attributes of each cloud point, the relationship between each cloud point in the attribute 3D point cloud is determined, as well as the correspondence between each region to which each cloud point with attribute correlation belongs and each component or related component of the 3D appearance of the object to be identified. When the correspondence between each region of the attribute 3D point cloud and each component of the 3D appearance of the object to be identified or the correspondence between related components is determined, when constructing a 3D result model of the object to be identified through the attribute 3D point cloud, the region corresponding to each component is constructed as an independent component model, or the regions corresponding to related components are associated and constructed as an independent component model.

2. The method for determining the components of an object to be identified according to claim 1, characterized in that, For sensing data collected sequentially from different collection locations, if the feature values ​​of one or more types of two-dimensional sensing data at different collection locations meet the preset similarity conditions or are determined by a specific model to have a correlation threshold, and are unique at each collection location, then it is determined that the two-dimensional sensing data at different collection locations are correlated.

3. The method for determining the components of an object to be identified according to claim 1, characterized in that, If sensing data collected simultaneously from different locations are collected from the same spatial field at different locations, and there is only one object to be identified in the spatial field, or the collected sensing data can correctly point to one of the multiple objects to be identified, then for a certain object to be identified, one or more two-dimensional sensing data collected from different locations are correlated.

4. The method for determining the components of an object to be identified according to claim 1, characterized in that, For sensing data collected simultaneously from different acquisition locations, if the same spatial field is collected from different acquisition locations, the acquired sensing data can be calibrated using the sensing data of known objects to establish a calibration model. When performing joint calculations, the calibration model is used to handle the differences between sensing data collected simultaneously from different acquisition locations.

5. The method for determining the components of an object to be identified according to claim 1, characterized in that, For a certain type of two-dimensional sensing data of a certain object to be identified, if the features of the two-dimensional sensing data are not detected at other subsequent acquisition locations, but it can be determined from the features of other types of two-dimensional sensing data that the features of the undetected two-dimensional sensing data still belong to the object to be identified, then the features of the undetected two-dimensional sensing data will continue to be designated as belonging to the object to be identified.

6. The method for determining the components of an object to be identified according to claim 5, characterized in that, The acquisition device for acquiring sensing data is time-synchronized, and the acquired two-dimensional sensing data and the original three-dimensional point cloud are recorded with timestamps. Based on the timestamps, the state of the object to be identified during the acquisition interval is inferred to obtain inferred sensing data. When performing joint calculation, the inferred sensing data is assigned a different weight than the acquired sensing data.

7. The method for determining the components of an object to be identified according to any one of claims 1 to 6, characterized in that, In step 1), the two-dimensional sensing data includes one or more combinations of color, spectral combination, contour, line, smell, sound, light reflectivity, material, motion trend, electromagnetic characteristics, temperature, temperature change trend, vibration amplitude, rigidity, vibration frequency, and vibration frequency change trend.

8. The method for determining the components of an object to be identified according to claim 7, characterized in that, By combining electromagnetic induction, temperature patterns, vibration frequency variation characteristics, motion correlation, and light reflectivity, the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified is determined.

9. The method for determining the components of an object to be identified according to claim 8, characterized in that, The object image to be identified is acquired, and the object image and spectrum are processed in combination with light intensity, atmospheric conditions, light direction and object material. Based on the color and contour of the object image and spectrum, the parts of the object image belonging to the same component are determined. Then, in combination with electromagnetic induction, temperature law, vibration frequency change characteristics, motion correlation and light reflectivity, it is determined whether the cloud points in the three-dimensional point cloud are the same rigid or flexible component.

10. The method for determining the components of an object to be identified according to claim 8, characterized in that, The aforementioned motion correlation includes the different motion correlations exhibited by different materials.

11. The method for determining components of an object to be identified according to any one of claims 1 to 6, characterized in that, In step 1), the original 3D point cloud is obtained by LiDAR, or generated based on image data collected by multiple image acquisition devices.

12. A non-specific feature recognition method based on the component determination method according to any one of claims 1 to 11, characterized in that, A peer-to-peer network is set up, comprising multiple node devices, with no hierarchy among them. Each node device is equipped with a data acquisition device and a computing module. The data acquisition device includes a lidar and at least one other type of sensor, used to acquire different corresponding types of sensing data. The sensing data includes two-dimensional sensing data and raw three-dimensional point clouds. Node devices located at different acquisition positions acquire at least one point sample of the object to be identified. The point sample is the raw three-dimensional point cloud and the two-dimensional sensing data of the corresponding sensor type. For a given node device, the collected sensing data is processed to obtain result data, which is then propagated to other node devices. Other node devices that receive the result data use it as one of the original data collected, and the result data influences the result data of other node devices. Based on this, without needing to obtain the identity information of the object to be identified, multiple node devices in the peer-to-peer network perform collaborative computation to determine that each unique object to be identified is itself, thus achieving non-specific feature recognition.

13. The non-specific feature recognition method according to claim 12, characterized in that, The current node device receives the result data output by other node devices; for the current node device, it combines the collected sensing data with the result data from other node devices to calculate the result data of the current node device, and then sends it to other node devices; In a peer-to-peer network, node devices perform collaborative computation as they collect sensing data and calculate result data.

14. The non-specific feature recognition method according to claim 12, characterized in that, In a peer-to-peer network, for a specific type of two-dimensional sensing data of an object to be identified, the resulting data transmitted from the node device that collected the two-dimensional sensing data to other node devices allows subsequent node devices to adjust their sensing attention based on the features of the two-dimensional sensing data, or report the features of the two-dimensional sensing data for subsequent node devices to adjust their sensing attention. If subsequent node devices do not detect the features of the two-dimensional sensing data, but can determine from the features of other types of two-dimensional sensing data that the undetected features still belong to the object to be identified, then the undetected features of the two-dimensional sensing data are continued to be represented in the result data of the current node device and transmitted to other node devices.

15. The non-specific feature recognition method according to claim 12, characterized in that, The method for reporting the features of the two-dimensional sensing data to subsequent node devices for adjusting sensing attention is as follows: Adjusting the parameters of the data processing model of subsequent node devices based on the result data expressing the features of the two-dimensional sensing data provided by the preceding node device, or the features of the two-dimensional sensing data, so that the subsequent node devices improve their computing power for identifying the features of the two-dimensional sensing data; or, the subsequent node devices use the sensing attention model to match the features of the received two-dimensional sensing data or the result data expressing the features of the two-dimensional sensing data for computing power adjustment.

16. The non-specific feature recognition method according to claim 14, characterized in that, When a node device processes the output data of several preceding node devices, based on the data processing model, if the object to be identified described by several preceding node devices can be determined as the same target through certain common features of the two-dimensional sensing data, the features and other information of the two-dimensional sensing data described by each node device are merged into the same target.

17. The non-specific feature recognition method according to claim 16, characterized in that, If the result data received by a node device indicates that the flag used by the current node device to identify the object to be identified before the current receipt of result data is different from the flag used by other node devices to identify the object to be identified, and the flags assigned to the object to be identified by other node devices have been updated, then the flag used by the current node device to identify the object to be identified before the current receipt of result data is converted.

18. The non-specific feature recognition method according to claim 16, characterized in that, The method for converting the flag used by the current node device to identify the object to be identified before the current reception result data is as follows: Replace the flag used by the current node device to identify the object to be identified before the current reception of result data with the latest flag assigned to the object by other node devices; Alternatively, record the conversion relationship between the flag used by the current node device to identify the object to be identified before the current reception of result data and the updated flags assigned to the object by other node devices, and perform the conversion when it is necessary to reference the result data received by the current node device in the current reception. Alternatively, node devices can deploy conversion models to perform corresponding conversions on the labels of multiple objects to be identified based on the input raw data or result data.

19. The non-specific feature recognition method according to claim 12, characterized in that, When it is determined that the identity information of the object to be identified needs to be obtained, an identity information acquisition command is triggered. The identity information acquisition command is used as one of the inputs to participate in the calculation of the result data of the node device. By driving the node device in the peer-to-peer network that is connected to the barrier-free data collection conditions that can obtain the identity information of the object to be identified, the corresponding result data is responded to, thereby realizing the acquisition of the identity information of the object to be identified.

20. The non-specific feature recognition method according to claim 19, characterized in that, Peer-to-peer networks determine the permissions of an object by verifying the authenticity of its identity information. In this process, the node devices in the peer-to-peer network that can obtain identity information do not provide the identity information itself, but only express the verification results in the result data of the node device based on the verification requirements for the authenticity of the identity information in the received result data.

21. The non-specific feature recognition method according to claim 20, characterized in that, In a peer-to-peer network, the node device capable of obtaining identity information does not provide the identity information itself. Instead, the information source device that drives the provision of identity information establishes an encrypted information transmission channel with the node device input terminal that needs to obtain the identity information, or establishes an encrypted information transmission channel using other network communication modes, and uses the identity information as one of the inputs to the node device.

22. The method for determining components of an object to be identified according to any one of claims 12 to 21, characterized in that, A component attribute knowledge base is also set up, which records the attribute parameters corresponding to each component of a known type of target. The attribute parameters of each component in the component attribute knowledge base are combined with the attributes of the cloud points in the attribute 3D point cloud for collaborative calculation to help determine the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified.

23. The non-specific feature recognition method according to any one of claims 12 to 21, characterized in that, The scene is reconstructed in 3D by using images from cameras located at different acquisition positions that capture images of the same space, along with their installation locations and camera parameters, to obtain a 3D scene model. The environment of the 3D scene is adjusted by using the perception data collected on-site as input. By comparing the states of the 3D scene model in the 3D scene at different time points and under different environments, static facilities and dynamic targets, as well as the various components of the dynamic targets, can be distinguished. The state data of the three-dimensional data after the dynamic target is reconstructed in the scene under the corresponding environmental parameters are calculated together with the three-dimensional point cloud of each attribute to determine the correspondence between each region of the attribute three-dimensional point cloud and each part or related part of the 3D appearance of the object to be identified. Among them, the camera parameters, installation location and image are used, or the camera parameters, installation location and image are combined with radar parameters, radar installation location and detection data to perform three-dimensional reconstruction of the current scene where the object to be identified is located; When reconstructing a scene, the initial 3D scene model is constructed by using the perception data collected on-site as input, adjusting the environment of the 3D scene, and using optical, acoustic, and physical models to cross-verify the calculation results of the light and appearance state in the scene with the result data of each node device. The scene is further optimized based on the verification results.

24. The non-specific feature recognition method according to claim 23, characterized in that, The perception data collected on-site serves as the input to the optical, acoustic, and physical models in the 3D scene model after the scene is reconstructed using the images from each camera, the installation location, and the camera parameters. The output includes one or more data such as the color, light reflectivity, shadow shape and density, and smooth surface reflection of the dynamic target. The output dynamic target data is then used in conjunction with the 3D point cloud of each attribute to determine the correspondence between each region of the attribute 3D point cloud and each component or related component of the 3D appearance of the object to be identified. One or more of the attribute 3D point cloud, physical parameters of the clearly defined target in the knowledge base, and on-site collected data such as illumination, weather, air quality, air composition, and wind speed are used as training parameters to accumulate samples and update the training of the optical model, acoustic model, and physical model.