Riding risk intelligent early warning system and method based on intelligent helmet

Abstract

The invention relates to the field of intelligent early warning, and particularly discloses a riding risk intelligent early warning system and method based on an intelligent helmet. The method comprises the following steps of: performing data element and local time sequence coding on a time sequence queue of strong monitoring data (relative distance and relative speed) by using a data analysis and coding technology based on artificial intelligence; and performing core time sequence feature significant convergence on each coded relative distance local time sequence implicit correlation feature and each coded relative speed local time sequence implicit correlation feature, so as to intelligently judge whether a collision risk exists or not according to interaction analysis between the relative distance and relative speed time sequence feature significant convergence representations. And the analysis result is sent to the second processor so as to control the vibration module to carry out early warning on the user. Therefore, the reliability of early warning can be improved, and the riding safety is effectively guaranteed.

Description

Technical Field

[0001] This application relates to the field of intelligent warning, and more specifically, to an intelligent warning system and method for riding risks based on an intelligent helmet. Background Art

[0002] Riding, as a green travel mode and fitness exercise, is becoming increasingly popular among the public. However, during the riding process, there are multiple potential threats such as motor vehicles, pedestrians, and complex road conditions, and the safety issue has become increasingly prominent. As the most important safety equipment for riders, a helmet, through its physical structure design, disperses and absorbs the impact force during a collision, thus reducing the injury to the head. Nevertheless, ordinary helmets have a single function and can only provide protection during an accident. They cannot monitor the surrounding environment in real time and give early warnings of potential dangers, making it difficult to meet the needs of modern riding safety.

[0003] In this regard, Patent CN114424851A proposes an intelligent helmet and a collision warning method. It monitors the distance and speed of surrounding moving objects through a radar and sends the data to a first processing chip. This chip analyzes whether there is a collision risk, such as the relative distance being less than a preset threshold or the moving trajectories intersecting. If there is a risk, the first processing chip notifies a second processing chip, which controls the vibration module to vibrate to physically remind the user of the potential danger. This process ensures that riders can obtain alerts in time to avoid accidents.

[0004] In the above patent, it is to judge whether there is a collision risk by comparing the relative distance or speed with a preset threshold. However, a fixed preset threshold is difficult to dynamically adapt to different traffic environments and riding conditions. For example, the requirements for safety distances are significantly different between busy urban roads and open rural roads. In addition, the reaction times and habits of different riders are also different, and a fixed threshold may not be applicable to all users. If the threshold is set too sensitively, it will lead to frequent false alarms, disturbing the users and possibly causing them to ignore real alarm signals.

[0005] Therefore, an optimized intelligent warning scheme for riding risks based on an intelligent helmet is desired. Summary of the Invention

[0006] To solve the above technical problems, this application is proposed.

[0007] According to one aspect of this application, an intelligent warning system for riding risks based on an intelligent helmet is provided, which includes:

[0008] A monitoring data acquisition module, configured to obtain monitoring data of moving objects within the detection range through a radar sensor deployed on the helmet to obtain a time series queue of the monitoring data, where the monitoring data includes the relative distance and relative speed between the user and the moving object;

[0009] A cycling collision risk analysis module for sending the time series queue of the monitoring data to a first processor for cycling collision risk analysis to obtain a warning analysis result, where the warning analysis result is used to indicate whether there is a collision risk. Among them, the first processor is used to perform time series clustering and significant interaction analysis of the core time series data of the monitoring data to perform the cycling collision risk analysis;

[0010] A warning module for sending the warning analysis result to a second processor, and the second processor controls a vibration module to vibrate based on the warning analysis result to give a warning prompt to the user.

[0011] According to another aspect of the present application, a cycling risk intelligent warning method based on an intelligent helmet is provided, including:

[0012] Obtaining the monitoring data of the moving objects within the detection range through a radar sensor deployed on the helmet to obtain the time series queue of the monitoring data, where the monitoring data includes the relative distance and relative speed between the user and the moving objects;

[0013] Sending the time series queue of the monitoring data to a first processor for cycling collision risk analysis to obtain a warning analysis result, where the warning analysis result is used to indicate whether there is a collision risk. Among them, the first processor is used to perform time series clustering and significant interaction analysis of the core time series data of the monitoring data to perform the cycling collision risk analysis;

[0014] Sending the warning analysis result to a second processor, and the second processor controls a vibration module to vibrate based on the warning analysis result to give a warning prompt to the user.

[0015] Compared with the prior art, the present application provides a smart early warning system and method for riding risk based on a smart helmet, which obtains the monitoring data of the moving object within the detection range through the radar sensor deployed on the helmet to obtain the time series queue of the monitoring data (relative distance and relative speed), and sends the time series queue of the monitoring data to the first processor, in which the time series queue of the monitoring data is subjected to data division and local time series encoding using the data analysis and encoding technology based on artificial intelligence, and then the encoded local time series implicit correlation features of each relative distance and each local time series implicit correlation features of each relative speed are respectively subjected to significant convergence of core time series features, so as to intelligently judge whether there is a collision risk according to the interactive analysis between the significant convergence representation of the relative distance time series features and the significant convergence representation of the relative speed time series features, and send the analysis results to the second processor to control the vibration module to warn the user. Compared with the fixed threshold in the patent, the present application can adaptively adjust the risk assessment standard according to the specific scenario, thereby effectively reducing the false alarm rate and improving the credibility of the warning, thereby effectively ensuring riding safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 It is a block diagram of a cycling risk intelligent warning system based on a smart helmet according to an embodiment of the present application.

[0018] Figure 2 The present invention is a block diagram of a riding collision risk analysis module in a riding risk intelligent warning system based on a smart helmet according to an embodiment of the present application.

[0019] Figure 3 This is a block diagram of a monitoring data core time series feature aggregation unit in a smart helmet-based riding risk intelligent early warning system according to an embodiment of the present application.

[0020] Figure 4 The present invention is a flowchart of a method for intelligent early warning of cycling risks based on a smart helmet according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Instead, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0022] It should be noted that in this application, all actions of obtaining signals, information, or data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where the location is located and obtaining authorization from the owner of the corresponding device.

[0023] In response to the problems in the above background art, the present application proposes a smart warning system for riding risks based on a smart helmet. It obtains a time series queue of monitoring data (relative distance and relative speed) of moving objects within the detection range through a radar sensor deployed on the helmet, and sends the time series queue of the monitoring data to a first processor. In the first processor, data grouping and local time series encoding are performed on the time series queue of the monitoring data by using artificial intelligence-based data analysis and encoding techniques. Then, significant aggregation of core time series features is respectively performed on each local time series implicit correlation feature of the relative distance and each local time series implicit correlation feature of the relative speed after encoding. Based on this, an intelligent judgment is made on whether there is a collision risk through the interaction analysis between the significant aggregation representation of the relative distance time series features and the significant aggregation representation of the relative speed time series features, and the analysis result is sent to a second processor to control a vibration module to give a warning to the user. Compared with the fixed threshold in the patent, the present application can adaptively adjust the risk assessment standard according to the specific scenario, thereby effectively reducing the false alarm rate and improving the credibility of the warning, thus effectively ensuring riding safety.

[0024] Figure 1 It is a block diagram of a smart warning system for riding risks based on a smart helmet according to an embodiment of the present application. Specifically, as Figure 1As shown in the figure, the intelligent warning system 100 for riding risks based on an intelligent helmet according to an embodiment of the present application includes: a monitoring data acquisition module 110, configured to obtain monitoring data of moving objects within a detection range through a radar sensor deployed on the helmet to obtain a time series queue of the monitoring data, where the monitoring data includes the relative distance and relative speed between the user and the moving object; a riding collision risk analysis module 120, configured to send the time series queue of the monitoring data to a first processor for riding collision risk analysis to obtain a warning analysis result, where the warning analysis result is used to indicate whether there is a collision risk. Among them, the first processor is configured to perform time series clustering and significant interaction analysis of monitoring core time series data on the time series queue of the monitoring data to perform the riding collision risk analysis; a warning module 130, configured to send the warning analysis result to a second processor, and the second processor controls a vibration module to vibrate based on the warning analysis result to give a warning prompt to the user.

[0025] In an embodiment of the present application, the monitoring data acquisition module 110 is configured to obtain monitoring data of moving objects within a detection range through a radar sensor deployed on the helmet to obtain a time series queue of the monitoring data, where the monitoring data includes the relative distance and relative speed between the user and the moving object. It should be understood that the monitoring data refers to the relevant data of moving objects within the detection range obtained through the radar sensor deployed on the helmet, specifically including the relative distance and relative speed between the user (rider) and the moving object. Specifically, the relative distance refers to the spatial interval between the user and other moving objects within the detection range. It is one of the basic parameters for evaluating collision risks. The relative speed indicates the speed at which the user and other moving objects approach or move away from each other. Combining the relative distance can more accurately predict the possibility and urgency of potential collisions. By obtaining these two data, a comprehensive understanding of the dynamic environment where the rider is located can be achieved, including which moving objects are around, their distances from the rider, and the relative movement speeds, so as to timely detect targets that may pose a threat to the rider and take corresponding measures to avoid accidents.

[0026] The radar sensor is a key device for obtaining monitoring data. It is carefully deployed on the helmet to ensure that it can detect surrounding moving objects to the greatest extent. The radar sensor mainly works based on the principle of electromagnetic wave reflection. It emits electromagnetic wave signals into the surrounding space. When these electromagnetic waves encounter a moving object, they will be reflected, and the reflected wave is received by the radar sensor. By measuring the time delay between the transmitted wave and the reflected wave and combining the propagation speed of electromagnetic waves in the air, the relative distance between the user and the moving object can be accurately calculated. For example, if the time difference between the transmitted signal and the received reflected signal is t and the propagation speed of electromagnetic waves is c, according to the formula d = c×t / 2 (dividing by 2 because the electromagnetic wave travels a round-trip distance), the relative distance d can be obtained.

[0027] For the measurement of relative speed, the radar sensor utilizes the Doppler effect. When there is relative motion between the moving object and the helmet, the frequency of the reflected electromagnetic wave will change. By detecting this frequency change (i.e., Doppler frequency shift), the radar sensor can calculate the speed of the moving object relative to the user. Assuming the transmitted signal frequency is f0 and the received signal frequency is f1, the Doppler frequency shift Δf = f1 - f0. According to the Doppler effect formula v = c×Δf / (2×f0) (where c is the propagation speed of the electromagnetic wave), the relative speed v can be obtained.

[0028] In the process of obtaining relative distance and relative speed data, the real-time and continuity of the data also need to be considered. The radar sensor continuously emits and receives electromagnetic wave signals at a certain sampling frequency, so that the position and speed information of the moving object can be continuously obtained at different time points. Each piece of data obtained is recorded and arranged in chronological order, thus forming a time-series queue of monitoring data. The selection of the sampling frequency is crucial. If the sampling frequency is too low, important motion information may be missed, resulting in untimely warnings. If the sampling frequency is too high, although more accurate data can be obtained, it will increase the burden of data processing. Usually, a suitable sampling frequency will be selected according to the actual application scenario and hardware performance to balance the accuracy and processing efficiency of the data.

[0029] To ensure the accuracy and reliability of the obtained data, the radar sensor also needs to be calibrated and error compensated. In actual use, it will be affected by various factors, such as weather conditions (rain, snow, fog, etc.), electromagnetic interference in the surrounding environment, etc. These factors may affect the measurement accuracy of the radar sensor. Therefore, the radar sensor needs to be calibrated regularly. By measuring and comparing with a standard target with known distance and speed, the parameters of the sensor are adjusted to improve the measurement accuracy. At the same time, some error compensation algorithms will also be used to correct the errors in the measurement data. For example, filtering algorithms are used to remove noise interference and improve the stability and reliability of the data.

[0030] In addition, the design of the helmet also needs to consider the deployment position and angle of the radar sensor. To obtain a wider detection range, the radar sensor is generally installed at the front, side, etc. of the helmet, and its angle is carefully adjusted to ensure that the main areas in front of and on the side of the cyclist can be covered. At the same time, the structural design of the helmet should not cause occlusion or interference to the electromagnetic wave emission and reception of the radar sensor, otherwise it will affect the quality of data acquisition.

[0031] In terms of data transmission, the monitoring data obtained by the radar sensor needs to be transmitted to the subsequent processing unit (such as the first processor) in a timely manner. This is usually achieved through wired or wireless communication technologies. The wired communication method has the advantages of stable transmission and strong anti-interference ability, but it may limit the flexibility and wearing comfort of the helmet; the wireless communication method (such as Bluetooth, Wi-Fi, etc.) is more convenient and can improve the user experience, but attention needs to be paid to the signal stability and transmission distance. When choosing a communication method, various factors will be comprehensively considered to ensure that the data can be transmitted to the processing unit quickly and accurately.

[0032] In the embodiment of the present application, the riding collision risk analysis module 120 is configured to send the time series queue of the monitoring data to the first processor for riding collision risk analysis to obtain a warning analysis result, where the warning analysis result is used to indicate whether there is a collision risk. Among them, the first processor is configured to perform time series clustering and significant interaction analysis of the monitoring core time series data on the time series queue of the monitoring data to perform the riding collision risk analysis. Correspondingly, considering that the time series queue of the monitoring data contains a large amount of raw data information, these data need to be processed and analyzed through specific methods to be converted into valuable information to determine whether there is a collision risk. The first processor has powerful data processing capabilities and corresponding algorithms, and can process these complex time series data, so as to extract the implicit features and correlation information in the data and provide a basis for the judgment of the collision risk.

[0033] Figure 2 It is a block diagram of the riding collision risk analysis module in the riding risk intelligent warning system based on an intelligent helmet according to the embodiment of the present application. Specifically, as Figure 2 shown, the riding collision risk analysis module 120 includes: a monitoring data sub-division local time series extraction unit 121, configured to perform data sub-division and local time series sequence coding on the time series queue of the monitoring data to obtain the time series distribution of the relative distance local time series implicit correlation feature vector and the time series distribution of the relative speed local time series implicit correlation feature vector; a monitoring data core time series feature aggregation unit 122, configured to perform significant aggregation of the monitoring data based on the core time series features on the time series distribution of the relative distance local time series implicit correlation feature vector and the time series distribution of the relative speed local time series implicit correlation feature vector respectively to obtain a relative distance time series feature significant aggregation coding vector and a relative speed time series feature significant aggregation coding vector; a monitoring data feature interaction analysis unit 123, configured to perform feature interaction analysis on the relative distance time series feature significant aggregation coding vector and the relative speed time series feature significant aggregation coding vector to obtain a relative motion parameter time series fusion coding vector; a warning analysis result generation unit 124, configured to obtain the warning analysis result based on the relative motion parameter time series fusion coding vector.

[0034] Specifically, the local time series extraction unit 121 of the monitoring data sub-queue is used to perform data sub-queueing and local time series encoding on the time series queue of the monitoring data to obtain the time series distribution of the relative distance local time series implicit correlation feature vector and the time series distribution of the relative speed local time series implicit correlation feature vector. Specifically, in the embodiment of the present application, the local time series extraction unit of the monitoring data sub-queue is used to: sub-queue the time series queue of the monitoring data according to the parameter sample dimension to obtain the time queue of the relative distance and the time queue of the relative speed; use the sequence encoder based on the time convolutional network to perform local time series encoding on the time queue of the relative distance and the time queue of the relative speed to obtain the time series distribution of the relative distance local time series implicit correlation feature vector and the time series distribution of the relative speed local time series implicit correlation feature vector. It should be understood that considering that the relative distance and the relative speed are two different dimensions with different physical meanings and variation laws. In the time series queue of the monitoring data, their mixing is not conducive to analyzing and mining their features separately. That is, both the relative distance and the relative speed have their own time series change trends and implicit feature information. Therefore, in order to perform more detailed processing and analysis according to the respective characteristics of the relative distance and the relative speed, in the technical solution of the present application, the time series queue of the monitoring data is sub-queued according to the parameter sample dimension to obtain the time queue of the relative distance and the time queue of the relative speed. That is, by obtaining the time queue of the relative distance and the time queue of the relative speed, the changes of the relative distance and the relative speed over time can be analyzed in depth respectively. For example, analyzing the change of the relative distance can understand the approaching or moving away trend between the moving object and the cyclist, and analyzing the change of the relative speed can master the change of the relative movement speed between the two, providing richer feature information for accurately judging the collision risk. Correspondingly, considering that both the time queue of the relative distance and the time queue of the relative speed have time series feature changes at different local time scales, such as whether the speed is too fast in the short term and whether the distance continues to decrease. Based on this, in the technical solution of the present application, the time queue of the relative distance and the time queue of the relative speed are subjected to local time series encoding by using the sequence encoder based on the time convolutional network to capture the local patterns and short-term dependencies in the time series, and obtain the time series distribution of the relative distance local time series implicit correlation feature vector and the time series distribution of the relative speed local time series implicit correlation feature vector. It is worth mentioning that the time convolutional network is particularly suitable for processing data with time series properties, such as the time queues of relative distance and relative speed. It can effectively capture the local dependencies and patterns of the data in the time dimension.Through convolution operations, the Temporal Convolutional Network (TCN) can extract local temporal features at different time steps to better process the temporal information in the data, uncover the implicit laws and correlations of relative distance and relative speed over time, highlight the features that are crucial for collision risk judgment, and thus provide a better data basis for subsequent analysis.

[0035] Specifically, the monitoring data core temporal feature aggregation unit 122 is configured to perform significant aggregation of the monitoring data based on core temporal features on the time series distributions of the relative distance local temporal implicit correlation feature vector and the relative speed local temporal implicit correlation feature vector respectively, to obtain a relative distance temporal feature significantly aggregated coding vector and a relative speed temporal feature significantly aggregated coding vector. Further, considering that the time series distributions of the relative distance local temporal implicit correlation feature vector and the relative speed local temporal implicit correlation feature vector respectively represent the dynamic changes of the relative distance and speed between the cyclist and moving objects (such as other cyclists, vehicles, obstacles, etc.) within a local period of time. To more comprehensively and accurately determine whether the cyclist is facing a collision risk, it is necessary to aggregate these local temporal features based on the full time domain. In complex time series data, not all features are equally important for collision risk judgment. For example, features such as a sudden decrease in relative distance or a sharp change in relative speed may be important signals of an impending collision. Therefore, in order to globally condense these features while retaining local significant features, the present application performs significant aggregation of the monitoring data based on core temporal features on the time series distributions of the relative distance local temporal implicit correlation feature vector and the relative speed local temporal implicit correlation feature vector respectively, to obtain a relative distance temporal feature significantly aggregated coding vector and a relative speed temporal feature significantly aggregated coding vector. This approach can effectively reconcile the conflict between global feature compression and local key feature retention when obtaining a concise representation of global features. It can not only highlight important features but also enable the generated significantly aggregated coding vectors to cover both the overall macroscopic information and not miss local significant features, making it possible to more efficiently utilize these features in subsequent collision risk judgment.

[0036] Particularly, the processing process of the time series distribution of the relative distance local temporal implicit correlation feature vector is taken as an example for specific illustration here.

[0037] Figure 3 It is a block diagram of the monitoring data core temporal feature aggregation unit in the cycling risk intelligent warning system based on an intelligent helmet according to an embodiment of the present application. Specifically, as Figure 3As shown, the monitoring data core timing feature aggregation unit 122 includes: a coarse and fine-grained aggregation feature calculation subunit 1221, configured to calculate a coarse-grained aggregation feature and a fine-grained compensation aggregation feature of the time series distribution of the relative distance local timing implicit association feature vector respectively to obtain a relative distance timing implicit association feature coarse-grained aggregation coding vector and a relative distance timing implicit association feature fine-grained compensation aggregation coding vector; a relative distance timing multi-scale interaction subunit 1222, configured to perform weighted interaction processing on the relative distance timing implicit association feature coarse-grained aggregation coding vector and the relative distance timing implicit association feature fine-grained compensation aggregation coding vector to obtain the relative distance timing feature significant aggregation coding vector.

[0038] Specifically, in the embodiment of the present application, the coarse and fine-grained aggregation feature calculation subunit 1221 includes: a relative distance coarse-grained aggregation secondary subunit, configured to perform relative distance information kernel coarse-grained aggregation on the time series distribution of the relative distance local timing implicit association feature vector to obtain the relative distance timing implicit association feature coarse-grained aggregation coding vector; a relative distance aggregation compensation factor calculation secondary subunit, configured to calculate a kernel aggregation compensation factor of each relative distance local timing implicit association feature vector in the time series distribution of the relative distance local timing implicit association feature vector relative to the relative distance timing implicit association feature coarse-grained aggregation coding vector to obtain a time series distribution of the relative distance timing feature kernel aggregation compensation factor; a relative distance compensation weight calculation secondary subunit, configured to perform gated function explicit compensation on the time series distribution of the relative distance timing feature kernel aggregation compensation factor to obtain a time series distribution of the relative distance timing feature kernel aggregation compensation weight factor; a relative distance fine-grained aggregation secondary subunit, configured to perform relative distance timing fine-grained dynamic compensation aggregation on the time series distribution of the relative distance timing feature kernel aggregation compensation weight factor, the relative distance timing implicit association feature coarse-grained aggregation coding vector, and the time series distribution of the relative distance local timing implicit association feature vector to obtain the relative distance timing implicit association feature fine-grained compensation aggregation coding vector.

[0039] Specifically, the relative distance coarse-grained aggregation secondary subunit is configured to perform relative distance information kernel coarse-grained aggregation on the time series distribution of the relative distance local timing implicit association feature vector to obtain the relative distance timing implicit association feature coarse-grained aggregation coding vector, and this process is represented by the formula:

[0040]

[0041] Wherein, is the time series distribution of the relative distance local timing implicit association feature vector, , , and are respectively the 1st, 2nd, th, and th relative distance local temporal implicit correlation feature vectors in the time series distribution of the relative distance local temporal implicit correlation feature vectors, and are respectively the maximum value and the minimum value of taking , is the th relative distance local temporal implicit correlation feature aggregation value in the time series distribution of the relative distance local temporal implicit correlation feature aggregation value, is the normalization function, is the th normalized relative distance local temporal implicit correlation feature aggregation value in the time series distribution of the normalized relative distance local temporal implicit correlation feature aggregation value, is the number of vectors in, is the relative distance temporal implicit correlation feature coarse-grained aggregation coding vector.

[0042] It should be understood that the original relative distance local temporal implicit correlation feature vector data is huge and complex. Therefore, in order to capture the overall global trend and at the same time extract the key temporal characteristics, the relative distance information kernel coarse-grained aggregation is performed on the time series distribution of the relative distance local temporal implicit correlation feature vectors to quickly extract the general features and overall trends in the data. Through the non-linear mapping constructed by the information kernel, the complex relationships between the relative distance features in the time series are accurately modeled in the implicit high-dimensional space, the main patterns are extracted, and the relative distance temporal implicit correlation feature coarse-grained aggregation coding vector is obtained.

[0043] More specifically, the relative distance aggregation compensation factor calculation secondary subunit is used for: a feature enhancement tertiary subunit, which is used to perform feature enhancement on the relative distance local temporal implicit association feature vector and the relative distance temporal implicit association feature coarse-grained aggregation coding vector based on point convolution and the sigmoid function respectively to obtain an enhanced relative distance local temporal implicit association feature vector and an enhanced relative distance temporal implicit association feature coarse-grained aggregation coding vector; a temporal difference calculation tertiary subunit, which is used to calculate the absolute difference vector between the enhanced relative distance local temporal implicit association feature vector and the enhanced relative distance temporal implicit association feature coarse-grained aggregation coding vector to obtain a relative distance temporal difference coding vector; a difference compensation tertiary subunit, which is used to obtain the relative distance temporal feature kernel aggregation compensation factor based on the relative distance temporal difference coding vector. More specifically, in the embodiment of the present application, the difference compensation tertiary subunit is used for: multiplying the relative distance temporal difference coding vector by the relative distance weight matrix and then performing element-wise addition with the relative distance compensation bias value to obtain a relative distance temporal compensation correction vector; multiplying the relative distance temporal compensation correction vector by the scoring weight vector to obtain the relative distance temporal feature kernel aggregation compensation factor corresponding to the relative distance local temporal implicit association feature vector. The processing process of the relative distance aggregation compensation factor calculation secondary subunit is expressed by the formula:

[0044]

[0045] Wherein, is the th relative distance local temporal implicit association feature vector in the time series distribution of the relative distance local temporal implicit association feature vector, is the relative distance temporal implicit association feature coarse-grained aggregation coding vector, is the point convolution coding, is the activation function, and are respectively and corresponding weight matrices, is corresponding enhanced relative distance local temporal implicit association feature vector, is the enhanced relative distance temporal implicit association feature coarse-grained aggregation coding vector, is element-wise subtraction by position, is the absolute value operation, is and relative distance temporal difference coding vector between, is corresponding relative distance weight matrix, is matrix multiplication, for calculating the two-norm of a vector, is the value of the logarithmic function with base 2, is the corresponding relative distance compensation offset value, is the corresponding scoring weight vector, is the th relative distance temporal feature kernel aggregation compensation factor in the time series distribution of the relative distance temporal feature kernel aggregation compensation factor.

[0046] Correspondingly, considering that although the coarse-grained aggregation coding vector can reflect the overall features, in the process of global feature compression and aggregation, some local personalized information of relative distance features may be weakened or lost. Therefore, calculating the kernel aggregation compensation factor of each relative distance local temporal implicit correlation feature vector relative to the relative distance temporal implicit correlation feature coarse-grained aggregation coding vector can measure the deviation between each local temporal implicit correlation feature and the coarse-grained aggregation coding vector, so as to dynamically generate the compensation factor for each local feature, retain the local details of the relative distance features, obtain the time series distribution of the relative distance temporal feature kernel aggregation compensation factor, and provide a basis for subsequent more refined compensation modeling.

[0047] Specifically, in response to the two-norm of the enhanced relative distance local temporal implicit correlation feature vector being less than the two-norm of the enhanced relative distance temporal implicit correlation feature coarse-grained aggregation coding vector, the logarithmic function value obtained by taking the logarithm with base 2 of the ratio between the two-norm of the enhanced relative distance local temporal implicit correlation feature vector and the two-norm of the enhanced relative distance temporal implicit correlation feature coarse-grained aggregation coding vector plus one is used as the relative distance compensation offset value; in response to the two-norm of the enhanced relative distance local temporal implicit correlation feature vector being greater than or equal to the two-norm of the enhanced relative distance temporal implicit correlation feature coarse-grained aggregation coding vector, the ratio obtained by dividing the two-norm of the enhanced relative distance local temporal implicit correlation feature vector by the two-norm of the enhanced relative distance temporal implicit correlation feature coarse-grained aggregation coding vector is used as the relative distance compensation offset value. This process is expressed by the formula:

[0048]

[0049] where, is the corresponding enhanced relative distance local temporal implicit correlation feature vector, is the enhanced relative distance temporal implicit correlation feature coarse-grained aggregation coding vector, is for calculating the two-norm of a vector, is the value of the logarithmic function with base 2, For the corresponding relative distance compensation offset value.

[0050] Specifically, here, for the deviation compensation between the relative distance local temporal implicit correlation feature vector and the relative distance temporal implicit correlation feature coarse-grained aggregation coding vector, the performance deviation of the kernel aggregation strategy as a scenario strategy can be measured by quantifying the regret metric based on the information kernel compression hypothesis in the kernel aggregation decision-making process, that is, the game-theoretic counterfactual regret value. Specifically, the vector norm representation is used to provide a normalized decision point loss description of the counterfactual regret value based on the policy actions, that is, the vector norm representations of the relative distance local temporal implicit correlation feature vector and the relative distance temporal implicit correlation feature coarse-grained aggregation coding vector. Then, for the possible differences in the vector distribution action game scenarios, the compensation rule correction of the node personalized information is carried out respectively with the information distribution degree of the regret value and the relative distribution amplitude of the regret value, so as to consider the relative distance local temporal implicit correlation feature personalized information as the un-taken action in the decision-making, and perform the offset compensation in the way of assuming its potential benefit based on the information kernel aggregation hypothesis.

[0051] Specifically, the relative distance compensation weight calculation secondary subunit is used to perform explicit compensation of the gating function on the time series distribution of the relative distance temporal feature kernel aggregation compensation factor to obtain the time series distribution of the relative distance temporal feature kernel aggregation compensation weight factor. This process is expressed by the formula:

[0052]

[0053] where is the -th relative distance temporal feature kernel aggregation compensation factor in the time series distribution of the relative distance temporal feature kernel aggregation compensation factor, is the gating compensation for , is the preset threshold, is the -th relative distance temporal feature kernel aggregation compensation weight factor in the time series distribution of the relative distance temporal feature kernel aggregation compensation weight factor.

[0054] It should be understood that considering that the time series distribution of the nuclear convergence compensation factor contains a lot of information about the deviation between local features and global features, but not all information is crucial for accurately describing the relative distance features, and there may be a lot of irrelevant or redundant content. Therefore, by performing explicit compensation on the time series distribution of the nuclear convergence compensation factor of the relative distance time series feature through a gating function, that is, using the gating function for explicit compensation, the intensity of the action of the compensation factor can be screened under the condition of non-linear constraints, highlighting the local features that have a significant impact on the relative distance features and suppressing irrelevant or redundant information, so as to more accurately capture the dynamic interaction relationship between the global and local features, and obtain the time series distribution of the nuclear convergence compensation weight factor of the relative distance time series feature.

[0055] Specifically, the relative distance fine-grained convergence secondary subunit is used to perform relative distance time series fine-grained dynamic compensation convergence on the time series distribution of the nuclear convergence compensation weight factor of the relative distance time series feature, the coarse-grained convergence coding vector of the relative distance time series implicit correlation feature, and the time series distribution of the relative distance local time series implicit correlation feature vector to obtain the fine-grained compensation convergence coding vector of the relative distance time series implicit correlation feature. This process is expressed by the formula:

[0056]

[0057] Wherein, is the th relative distance local time series implicit correlation feature vector in the time series distribution of the relative distance local time series implicit correlation feature vector, is the coarse-grained convergence coding vector of the relative distance time series implicit correlation feature, is the number of vectors in, is the th relative distance time series feature nuclear convergence compensation weight factor in the time series distribution of the relative distance time series feature nuclear convergence compensation weight factor, is the fine-grained compensation convergence coding vector of the relative distance time series implicit correlation feature.

[0058] Accordingly, considering that the coarse-grained aggregation coding vector, the time series distribution of local feature vectors, and the screened weight factors have been obtained in the previous steps, it is necessary to organically combine this information to accurately characterize the relative distance features. By performing relative distance time series fine-grained dynamic compensation aggregation on the time series distribution of the relative distance time series feature kernel aggregation compensation weight factor, the coarse-grained aggregation coding vector of the relative distance time series implicit correlation feature, and the time series distribution of the relative distance local time series implicit correlation feature vector, it is possible to dynamically adjust the relative distance local time series implicit correlation feature vector using the kernel aggregation compensation weight factor, while considering the global feature constraints represented by the coarse-grained aggregation coding vector, so that the compensated features can accurately reflect the local details of the relative distance and adapt to the dynamic changes of the global features.

[0059] Specifically, in the embodiment of the present application, the relative distance time series multi-scale interaction sub-unit 1222 is used to perform weighted interaction processing on the coarse-grained aggregation coding vector of the relative distance time series implicit correlation feature and the fine-grained compensation aggregation coding vector of the relative distance time series implicit correlation feature to obtain the relative distance time series feature significant aggregation coding vector. This process is expressed by the formula:

[0060]

[0061] where, is the coarse-grained aggregation coding vector of the relative distance time series implicit correlation feature, is the fine-grained compensation aggregation coding vector of the relative distance time series implicit correlation feature, and are weighted hyperparameters, is the relative distance time series feature significant aggregation coding vector.

[0062] Finally, perform weighted interaction processing on the coarse-grained aggregation coding vector of the relative distance time series implicit correlation feature and the fine-grained compensation aggregation coding vector of the relative distance time series implicit correlation feature obtained after compensation, so that the features after fine-grained adjustment contain the global view of the coarse-grained features, while retaining local significant details, forming a high-quality feature expression form that combines global and local, and obtaining the relative distance time series feature significant aggregation coding vector.

[0063] Specifically, the monitoring data feature interaction analysis unit 123 is configured to perform feature interaction analysis on the relative distance time-series feature significantly converged coding vector and the relative speed time-series feature significantly converged coding vector to obtain a relative motion parameter time-series fusion coding vector. It should be understood that considering that the relative distance time-series feature significantly converged coding vector and the relative speed time-series feature significantly converged coding vector respectively reflect the relative motion situation between the rider and the surrounding moving objects from different perspectives. The relative distance mainly reflects the spatial interval between the two, while the relative speed reflects the speed of change of this spatial interval over time. In order to fuse the information of the two and achieve information complementarity, so as to obtain richer and more comprehensive relative motion information, the present application performs feature interaction analysis on the relative distance time-series feature significantly converged coding vector and the relative speed time-series feature significantly converged coding vector to obtain a relative motion parameter time-series fusion coding vector. Specifically, in an actual riding scenario, there is a complex mutual relationship between the relative distance and the relative speed. For example, the change in relative speed will directly affect the change trend of the relative distance, and the magnitude of the relative distance will also impose certain restrictions on the change in relative speed. This complex mutual relationship is crucial for judging the collision risk. Feature interaction analysis can deeply explore this complex relationship between the relative distance and the relative speed, capture the key information that is easily overlooked when analyzed separately, so as to more comprehensively reflect the state and trend of relative motion, and provide a more powerful basis for risk judgment. In particular, in a specific example of the present application, the relative distance time-series feature significantly converged coding vector and the relative speed time-series feature significantly converged coding vector can be obtained by calculating the element-wise multiplication between the relative distance time-series feature significantly converged coding vector and the relative speed time-series feature significantly converged coding vector.

[0064] Specifically, the warning analysis result generation unit 124 is configured to: input the relative motion parameter time-series fusion encoded vector into a collision risk analyzer based on a classifier to obtain the warning analysis result. That is to say, classification processing is performed on the relative motion parameter time-series fusion encoded vector obtained by performing feature interaction analysis using the relative distance time-series feature significantly converged encoded vector and the relative velocity time-series feature significantly converged encoded vector, so as to intelligently determine whether there is a collision risk. Specifically, a classifier is a mature and effective data processing tool that can perform pattern recognition and classification on the input data. The relative motion parameter time-series fusion encoded vector synthesizes key information such as relative distance and relative velocity. Inputting it into the classifier can utilize the powerful pattern recognition ability of the classifier to analyze and process these complex feature information, thereby determining different motion states and risk levels. For example, classifiers such as support vector machine (SVM) and decision tree can learn from a large amount of labeled training data to establish a mapping relationship between relative motion parameters and collision risks, identify patterns and rules related to collision risks, and thus obtain reliable analysis results.

[0065] In an embodiment of the present application, the warning module 130 is configured to send the warning analysis result to a second processor, and the second processor controls the vibration module to vibrate based on the warning analysis result to give a warning prompt to the user. It should be understood that during the cycling process, the user may not be able to notice visual or auditory warning signals in a timely manner due to various reasons (such as distraction of attention, noisy environment, etc.). Vibration is a perception method that can directly act on the user's body and can effectively attract the user's attention, enabling the user to quickly realize the dangerous situation around and take corresponding measures to avoid collisions. And the control of the vibration module needs to respond in a timely manner to ensure that the user can receive the warning information in the first time. The second processor can specifically perform rapid processing and response on the warning analysis result. Compared with letting the first processor that undertakes complex analysis tasks also take into account controlling the vibration module, the second processor can control the start and stop of the vibration module more timely and reliably, meeting the real-time requirements of the warning prompt. In addition, separating the control function to the second processor can also avoid problems such as control delay or instability caused by the first processor processing other complex tasks. In this way, vibration warning, as an effective safety assistance means, can provide important safety guarantees for users at critical moments and reduce the occurrence of traffic accidents.

[0066] After the early warning analysis result is generated by the first processor, it needs to be sent to the second processor through a specific data transmission mechanism. Usually, this process is achieved by means of an internal communication line or short-range wireless communication technology. If an internal communication line is adopted, the digitally encoded early warning analysis result can be transmitted to the second processor quickly and stably in the form of an electrical signal. During the transmission process, some data verification and error correction technologies are adopted to ensure the accuracy and integrity of the data. For example, a check code is added at the data sending end. After the second processor at the receiving end receives the data, it checks the data according to the verification rules. Once an error is found, it requests a retransmission to ensure the reliable transmission of the data. If short-range wireless communication technology such as Bluetooth is used, its working principle is based on wireless signal transmission in the 2.4GHz ISM band. The first processor encodes the early warning analysis result into a data frame that conforms to the Bluetooth protocol for broadcasting, and the second processor continuously listens on this band. Once a matching signal is received, it parses the data frame to obtain the early warning analysis result. In this wireless transmission mode, in order to avoid signal interference, a frequency hopping technology is adopted, that is, the transmission frequency is quickly switched between different frequency channels to ensure the stability of data transmission.

[0067] After the second processor receives the early warning analysis result, it will parse and judge it. Corresponding control programs and logical judgment algorithms are pre-written inside the second processor. When the received early warning analysis result indicates a collision risk, the control program will be triggered. This control program is like the "command center" of the second processor. It will determine the vibration mode of the vibration module according to different levels or types of risks. For example, if the risk level is high, it may control the vibration module to generate strong and frequent vibrations; if the risk level is low, the intensity and frequency of the vibration will be reduced accordingly. At the same time, the second processor also has the ability of real-time processing and fast response, and can complete the parsing and judgment of the early warning analysis result in an extremely short time to ensure the timeliness of the early warning.

[0068] The vibration module is the direct execution component for implementing user warning prompts. Common vibration modules mostly use eccentric rotating mass motors (ERM) or linear resonant actuators (LRA). Taking the eccentric rotating mass motor as an example, its working principle is that the motor drives an eccentric heavy object to rotate, generating an unbalanced centrifugal force during rotation, thereby causing vibration. The second processor adjusts the motor speed by controlling the current or voltage of the motor, and then controls the intensity and frequency of the vibration. When the second processor determines that a warning needs to be issued, it sends a corresponding electrical signal command to the vibration module. The voltage magnitude and pulse width of this electrical signal determine the motor speed and rotation time. If strong vibration is required, the second processor sends an electrical signal with a high voltage and a long pulse width, causing the motor to rotate at high speed and generating a strong vibration effect; conversely, if only slight vibration is needed, an electrical signal with a low voltage and a short pulse width is sent. For the linear resonant actuator, it uses electromagnetic force to drive a mass block to reciprocate at a specific frequency to generate vibration. The second processor adjusts the output electrical signal frequency to match the inherent resonant frequency of the linear resonant actuator, thereby achieving efficient vibration output. At the same time, the intensity of the vibration can also be adjusted by controlling the amplitude of the electrical signal.

[0069] Throughout the implementation process, in order to ensure the stable operation of the vibration module and the consistency of the warning effect, the vibration module also needs to be calibrated and tested. In the production process of the smart helmet, each vibration module is calibrated individually. By inputting electrical signals with different intensities and frequencies into the vibration module, measuring its actual vibration output, and then fine-tuning the control parameters according to the measurement results, it is ensured that each vibration module can produce a consistent vibration effect when receiving the same control signal. In addition, during the use of the smart helmet, a self-check program is also performed regularly. The second processor sends a specific test signal to the vibration module to detect whether the vibration module can work normally. If the vibration module fails, the smart helmet system will inform the user through other means (such as indicator light flashing or sound prompt) for timely repair or replacement.

[0070] In addition, considering the user experience and different riding scenarios, the vibration mode and intensity of the vibration module can also be personalized. Users can set different vibration modes through a mobile application or operation buttons on the helmet according to their own needs and preferences. For example, some users may prefer strong vibration prompts, while some users hope the vibration is relatively gentle. This personalized setting function increases the adaptability and user satisfaction of the smart helmet.

[0071] In summary, the intelligent early warning system 100 for riding risk based on the smart helmet according to the embodiment of the present application is explained, which obtains the monitoring data of the moving object within the detection range through the radar sensor deployed on the helmet to obtain the time series queue of the monitoring data (relative distance and relative speed), and sends the time series queue of the monitoring data to the first processor, in which the time series queue of the monitoring data is subjected to data division and local time series encoding using the data analysis and encoding technology based on artificial intelligence, and then, the encoded local time series implicit correlation features of each relative distance and each local time series implicit correlation features of each relative speed are respectively subjected to significant convergence of core time series features, so as to intelligently judge whether there is a collision risk according to the interactive analysis between the significant convergence representation of the relative distance time series features and the significant convergence representation of the relative speed time series features, and send the analysis results to the second processor to control the vibration module to warn the user. Compared with the fixed threshold in the patent, the present application can adaptively adjust the risk assessment standard according to the specific scenario, thereby effectively reducing the false alarm rate, improving the credibility of the warning, and effectively ensuring riding safety.

[0072] As described above, the intelligent warning system 100 for cycling risks based on a smart helmet according to the embodiment of the present application can be implemented in various wireless terminals, such as a server having an intelligent warning algorithm for cycling risks based on a smart helmet. In one possible implementation, the intelligent warning system 100 for cycling risks based on a smart helmet according to the embodiment of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the intelligent warning system 100 for cycling risks based on a smart helmet can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the intelligent warning system 100 for cycling risks based on a smart helmet can also be one of the many hardware modules of the wireless terminal.

[0073] Alternatively, in another example, the smart helmet-based riding risk intelligent warning system 100 and the wireless terminal may also be separate devices, and the smart helmet-based riding risk intelligent warning system 100 may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0074] Figure 4 FIG. 1 is a flow chart of a method for intelligent early warning of cycling risks based on a smart helmet according to an embodiment of the present application. Figure 4As shown, the intelligent riding risk early warning method based on an intelligent helmet according to an embodiment of the present application includes: S110, obtaining monitoring data of moving objects within a detection range through a radar sensor deployed on the helmet to obtain a time series queue of the monitoring data, where the monitoring data includes the relative distance and relative speed between the user and the moving object; S120, sending the time series queue of the monitoring data to a first processor for riding collision risk analysis to obtain an early warning analysis result, where the early warning analysis result is used to indicate whether there is a collision risk. Among them, the first processor is used to perform time series clustering and significant interaction analysis of monitoring core time series data on the time series queue of the monitoring data to perform the riding collision risk analysis; S130, sending the early warning analysis result to a second processor, and the second processor controls a vibration module to vibrate based on the early warning analysis result to give an early warning prompt to the user.

[0075] Here, those skilled in the art can understand that the specific operations of each step in the above intelligent riding risk early warning method based on an intelligent helmet have been introduced in detail in the description of the Figures 1 to 3 intelligent riding risk early warning system based on an intelligent helmet above, and therefore, its repeated description will be omitted.

[0076] The above specific implementation manners have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific implementation manners of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cycling risk intelligent early warning system based on a smart helmet, characterized in that: include: A monitoring data acquisition module, used to acquire monitoring data of a moving object within a detection range through a radar sensor deployed on the helmet to obtain a time series queue of the monitoring data, wherein the monitoring data includes a relative distance and a relative speed between the user and the moving object; A riding collision risk analysis module, used for sending the time series queue of the monitoring data to the first processor for riding collision risk analysis to obtain a warning analysis result, wherein the warning analysis result is used to indicate whether there is a collision risk, wherein the first processor is used for performing time series clustering on the time series queue of the monitoring data and monitoring core time series data significant interaction analysis to perform the riding collision risk analysis; The early warning module is used to send the early warning analysis result to the second processor, and the second processor controls the vibration module to vibrate based on the early warning analysis result to give an early warning prompt to the user.

2. The intelligent early warning system for cycling risks based on a smart helmet according to claim 1 is characterized in that: The riding collision risk analysis module includes: A monitoring data team local time series extraction unit is used to perform data teaming and local time series sequence encoding on the time series queue of the monitoring data to obtain a time series distribution of a relative distance local time series implicit correlation feature vector and a time series distribution of a relative speed local time series implicit correlation feature vector; A monitoring data core time series feature aggregation unit is used to perform significant aggregation of monitoring data based on core time series features on the time series distribution of the relative distance local time series implicit correlation feature vector and the time series distribution of the relative speed local time series implicit correlation feature vector to obtain a relative distance time series feature significant aggregation coding vector and a relative speed time series feature significant aggregation coding vector; A monitoring data feature interactive analysis unit, used for performing feature interactive analysis on the relative distance time series feature significant convergence coding vector and the relative speed time series feature significant convergence coding vector to obtain a relative motion parameter time series fusion coding vector; The early warning analysis result generating unit is used to obtain the early warning analysis result based on the relative motion parameter time series fusion coding vector.

3. The intelligent early warning system for cycling risks based on smart helmets according to claim 2 is characterized in that: The monitoring data team local time series extraction unit is used to: The time series queue of the monitoring data is divided into data queues according to the parameter sample dimension to obtain a time queue of relative distance and a time queue of relative speed; A sequence encoder based on a temporal convolutional network is used to perform local temporal sequence encoding on the time queue of the relative distance and the time queue of the relative speed to obtain the time series distribution of the local temporal implicit correlation feature vector of the relative distance and the time series distribution of the local temporal implicit correlation feature vector of the relative speed.

4. The intelligent early warning system for cycling risks based on a smart helmet according to claim 3 is characterized in that: The monitoring data core time series feature aggregation unit includes: A coarse-grained and fine-grained converged feature calculation subunit, used to respectively calculate the coarse-grained converged features and the fine-grained compensated converged features of the time series distribution of the relative distance local temporal implicit association feature vector to obtain a coarse-grained converged coding vector of the relative distance temporal implicit association feature and a fine-grained compensated converged coding vector of the relative distance temporal implicit association feature; The relative distance time series multi-scale interaction subunit is used to perform weighted interaction processing on the coarse-grained aggregation coding vector of the relative distance time series implicit correlation feature and the fine-grained compensation aggregation coding vector of the relative distance time series implicit correlation feature to obtain the significant aggregation coding vector of the relative distance time series feature.

5. The intelligent early warning system for cycling risks based on smart helmets according to claim 4 is characterized in that: The coarse-grained and fine-grained convergence feature calculation subunit includes: A relative distance coarse-grained aggregation secondary subunit is used to perform relative distance information kernel coarse-grained aggregation on the time series distribution of the relative distance local temporal implicit correlation feature vector to obtain the relative distance temporal implicit correlation feature coarse-grained aggregation encoding vector; A relative distance convergence compensation factor calculation secondary subunit is used to calculate the kernel convergence compensation factor of each relative distance local temporal implicit association feature vector in the time series distribution of the relative distance local temporal implicit association feature vector relative to the relative distance temporal implicit association feature coarse-grained convergence encoding vector to obtain a time series distribution of the relative distance temporal feature kernel convergence compensation factor; A relative distance compensation weight calculation secondary subunit is used to perform gate function explicit compensation on the time series distribution of the relative distance time series feature core convergence compensation factor to obtain the time series distribution of the relative distance time series feature core convergence compensation weight factor; The relative distance fine-grained aggregation secondary sub-unit is used to perform relative distance time series fine-grained dynamic compensation aggregation on the time series distribution of the relative distance time series feature core aggregation compensation weight factor, the relative distance time series implicit association feature coarse-grained aggregation coding vector and the relative distance local time series implicit association feature vector to obtain the relative distance time series implicit association feature fine-grained compensation aggregation coding vector.

6. The intelligent early warning system for cycling risks based on smart helmets according to claim 5 is characterized in that: The relative distance convergence compensation factor calculation secondary subunit is used to: The feature enhancement three-level sub-unit is used to perform feature enhancement based on point convolution and sigmoid function on the relative distance local temporal implicit association feature vector and the relative distance temporal implicit association feature coarse-grained aggregation coding vector to obtain the enhanced relative distance local temporal implicit association feature vector and the enhanced relative distance temporal implicit association feature coarse-grained aggregation coding vector; The third-level subunit of temporal difference calculation is used to calculate the difference absolute value vector between the enhanced relative distance local temporal implicit association feature vector and the enhanced relative distance temporal implicit association feature coarse-grained convergence coding vector to obtain the relative distance temporal difference coding vector; The difference compensation tertiary subunit is used to obtain the relative distance time series feature kernel convergence compensation factor based on the relative distance time series difference coding vector.

7. The intelligent early warning system for cycling risks based on a smart helmet according to claim 6 is characterized in that: The difference compensation tertiary subunit is used for: The relative distance time series difference encoding vector is multiplied by the relative distance weight matrix, and then the relative distance compensation offset value is added position by position to obtain a relative distance time series compensation correction vector; The relative distance time series compensation correction vector is multiplied by the scoring weight vector to obtain the relative distance time series feature core convergence compensation factor corresponding to the relative distance local time series implicit correlation feature vector; In which, in response to the fact that the second norm of the enhanced relative distance local temporal implicit association feature vector is less than the second norm of the enhanced relative distance temporal implicit association feature coarse-grained aggregation coding vector, a ratio between the second norm of the enhanced relative distance local temporal implicit association feature vector and the second norm of the enhanced relative distance temporal implicit association feature coarse-grained aggregation coding vector is added with a constant one, and a logarithmic function value obtained by taking the logarithm with base 2 is used as the relative distance compensation bias value; In response to the fact that the binary norm of the enhanced relative distance local temporal implicit association feature vector is greater than or equal to the binary norm of the enhanced relative distance temporal implicit association feature coarse-grained aggregation coding vector, the ratio of the binary norm of the enhanced relative distance local temporal implicit association feature vector and the binary norm of the enhanced relative distance temporal implicit association feature coarse-grained aggregation coding vector is used as the relative distance compensation bias value.

8. The intelligent early warning system for cycling risks based on smart helmets according to claim 7 is characterized in that: The warning analysis result generating unit is used to: input the relative motion parameter time series fusion coding vector into a classifier-based collision risk analyzer to obtain the warning analysis result.

9. A method for intelligent early warning of cycling risks based on a smart helmet, characterized in that: include: Acquire monitoring data of a moving object within a detection range through a radar sensor deployed on the helmet to obtain a time series queue of the monitoring data, wherein the monitoring data includes a relative distance and a relative speed between the user and the moving object; Sending the time series queue of the monitoring data to the first processor for performing a riding collision risk analysis to obtain a warning analysis result, wherein the warning analysis result is used to indicate whether there is a collision risk, wherein the first processor is used to perform time series clustering on the time series queue of the monitoring data and significant interaction analysis on the monitoring core time series data to perform the riding collision risk analysis; The warning analysis result is sent to the second processor, and the second processor controls the vibration module to vibrate based on the warning analysis result to give a warning prompt to the user.

Images