State detection method and system based on intelligent riding helmet

Abstract

The invention discloses a state detection method and system based on an intelligent riding helmet, and relates to the field of intelligent detection, and the technical concept of the application is that head motion data of a user is collected in real time to obtain a time queue of the head motion data (acceleration and angular velocity), and the time queue is transmitted to a micro-control processor; in the micro-control processor, a data processing and coding technology based on deep learning is used to carry out data arrangement and time sequence coding on head data. Therefore, whether the riding abnormity exists or not is intelligently judged according to the hidden key clue interactive representation between the acceleration time correlation coding characteristics and the angular speed time correlation coding characteristics after time sequence coding, and when the abnormity exists, an alarm is given and a position notification is sent at the same time. The detection accuracy can be improved, and the riding safety and the emergency response efficiency are greatly enhanced.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to a state detection method and system based on an intelligent cycling helmet. Background Art

[0002] As people's living standards improve and their environmental awareness increases, cycling, as a healthy and environmentally friendly way of travel and exercise, has been loved by more and more people. However, there are many safety risks in the cycling process, such as complex road conditions, irregular behaviors of other traffic participants, and the rider's own mistakes, all of which may lead to cycling accidents. Therefore, it is very important to perform effective status detection during cycling.

[0003] Traditional cycling helmets are mainly composed of an outer shell, an inner lining and straps. They are designed to disperse and absorb the impact force through the rigid structure of the outer shell and the cushioning material of the inner lining in the event of a collision, thereby reducing damage to the head. However, this passive protection has obvious limitations. It can only work at the moment of an accident, and cannot warn of potential dangers in advance, nor can it send out a distress signal to the outside world in time after an accident. Moreover, traditional helmets have a single function and lack the ability to monitor and analyze the rider's movement status in real time, and cannot meet the rider's needs for safe and intelligent riding.

[0004] Therefore, a state detection solution based on a smart cycling helmet is desired. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed.

[0006] According to one aspect of the present application, a state detection method based on a smart cycling helmet is provided, which includes:

[0007] The user's head movement data is collected in real time through the acceleration sensor and gyroscope built into the smart cycling helmet to obtain the time queue of the head movement data;

[0008] The time queue of the head motion data is transmitted to a microcontroller for riding state abnormality detection to obtain a detection result indicating whether riding abnormality exists, wherein the microcontroller is used to perform key clue deep interactive correlation on the time queue of the head motion data to obtain a temporal interactive coding feature of the head motion data and perform the riding state abnormality detection based on the temporal interactive coding feature of the head motion data;

[0009] In response to the detection result that there is riding abnormality, an alarm mechanism is triggered and a notification containing location information is sent to a preset emergency contact via Bluetooth.

[0010] According to another aspect of the present application, a state detection system based on a smart cycling helmet is provided, which includes:

[0011] A head motion data acquisition module is used to acquire the user's head motion data in real time through an acceleration sensor and a gyroscope built into the smart cycling helmet to obtain a time queue of the head motion data;

[0012] A riding anomaly detection module, used for transmitting the time queue of the head motion data to the microcontroller to perform riding state anomaly detection to obtain a detection result indicating whether there is riding anomaly, wherein the microcontroller is used for performing key clue deep interactive correlation on the time queue of the head motion data to obtain a temporal interactive coding feature of the head motion data and performing the riding state anomaly detection based on the temporal interactive coding feature of the head motion data;

[0013] The abnormal result response module is used to trigger an alarm mechanism and send a notification containing location information to a preset emergency contact via Bluetooth in response to the detection result that there is a riding abnormality.

[0014] Compared with the prior art, the present application provides a state detection method and system based on a smart cycling helmet, which collects the user's head motion data in real time through an acceleration sensor and a gyroscope built into the smart cycling helmet to obtain a time queue of the head motion data (acceleration and angular velocity), and transmits it to a microcontroller, in which a data processing and coding technology based on deep learning is used to sort and time-series encode the head data, so as to intelligently judge whether there is a riding abnormality according to the implicit key clue interaction between the acceleration time-related coding features and the angular velocity time-related coding features after the time-series coding, and when there is an abnormality, an alarm is issued and a location notification is sent. The present application can capture and analyze the time-series changes in acceleration and angular velocity in real time, thereby improving the accuracy of detection. In addition, once a potential riding abnormality is detected, the system will not only issue a local alarm, but also send a notification containing location information to ensure that the outside world can be informed of the situation and provide help in a timely manner, thereby greatly enhancing the safety of riding and the efficiency of emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] 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.

[0016] Figure 1The present invention is a flowchart of a method for detecting a state of a smart cycling helmet according to an embodiment of the present application.

[0017] Figure 2 This is a flowchart of step S120 in the state detection method based on the smart cycling helmet according to an embodiment of the present application.

[0018] Figure 3 This is a flowchart of step S122 in the state detection method based on the smart cycling helmet according to an embodiment of the present application.

[0019] Figure 4 It is a block diagram of a state detection system based on a smart cycling helmet according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0021] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0022] It is worth noting that in this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0023] As people's living standards improve and their environmental awareness increases, cycling, as a healthy and environmentally friendly way of travel and exercise, has been loved by more and more people. However, there are many safety risks in the cycling process, such as complex road conditions, irregular behaviors of other traffic participants, and the rider's own mistakes, all of which may lead to cycling accidents. Therefore, it is very important to perform effective status detection during cycling.

[0024] Traditional cycling helmets are mainly composed of an outer shell, an inner lining and straps. They are designed to disperse and absorb the impact force through the rigid structure of the outer shell and the cushioning material of the inner lining in the event of a collision, thereby reducing damage to the head. However, this passive protection has obvious limitations. It can only work at the moment of an accident, and cannot warn of potential dangers in advance, nor can it send out a distress signal to the outside world in time after an accident. Moreover, traditional helmets have a single function and lack the ability to monitor and analyze the rider's movement status in real time, and cannot meet the rider's needs for safe and intelligent riding.

[0025] Therefore, in response to the above technical problems, the technical concept of the present application is to collect the user's head motion data in real time through the acceleration sensor and gyroscope built into the smart cycling helmet to obtain the time queue of the head motion data (acceleration and angular velocity), and transmit it to the microcontroller, in which the data processing and coding technology based on deep learning is used to sort and time-series encode the head data, so as to intelligently judge whether there is a riding abnormality based on the implicit key clue interaction between the acceleration time-related coding features and the angular velocity time-related coding features after the time-series coding, and when there is an abnormality, an alarm is issued and a location notification is sent. The present application can capture and analyze the timing changes in acceleration and angular velocity in real time, thereby improving the accuracy of detection. In addition, once a potential riding abnormality is detected, the system will not only issue a local alarm, but also send a notification containing location information to ensure that the outside world can be informed of the situation and provide help in a timely manner, thereby greatly enhancing the safety of riding and the efficiency of emergency response.

[0026] Figure 1 FIG. 1 is a flow chart of a method for detecting a state of a smart cycling helmet according to an embodiment of the present application. Figure 1 As shown, according to the embodiment of the present application, the state detection method based on the smart cycling helmet includes: S110, collecting the user's head movement data in real time through the acceleration sensor and gyroscope built into the smart cycling helmet to obtain a time queue of the head movement data; S120, transmitting the time queue of the head movement data to the microcontroller processor for riding state abnormality detection to obtain a detection result indicating whether there is a riding abnormality, wherein the microcontroller processor is used to perform key clue deep interactive association on the time queue of the head movement data to obtain the head movement data timing interaction coding feature and perform the riding state abnormality detection based on the head movement data timing interaction coding feature; S130, in response to the detection result that the riding abnormality exists, triggering an alarm mechanism and sending a notification containing location information to a preset emergency contact via Bluetooth.

[0027] In step S110, the user's head movement data is collected in real time by the acceleration sensor and gyroscope built into the smart cycling helmet to obtain the time queue of the head movement data. In particular, the head movement data includes acceleration and angular velocity. It should be understood that during riding, the movement state of the head is closely related to the riding state. When the rider rides normally, the acceleration and angular velocity of the head movement will show a certain regularity and range. Specifically, acceleration describes the speed and direction of the change of the head movement speed, such as the change when the head suddenly accelerates or decelerates; angular velocity describes the speed and direction of the head rotation, such as the speed of the head turning left and right, up and down, etc. When abnormal conditions occur, such as falling, collision, etc., the acceleration and angular velocity of the head will change significantly, and this change is significantly different from the situation during normal riding. Therefore, by obtaining head movement data, rich and critical information can be provided for judging the riding state.

[0028] In order to ensure that the accelerometer and gyroscope can accurately collect head movement data, they are carefully placed inside the smart cycling helmet. When designing, the structure of the helmet and the actual position relationship after wearing it were fully considered, and the position closest to the center of head movement that can minimize external interference was selected for installation. The choice of this position is very critical, as it enables the sensor to directly and sensitively sense various changes in head movement, avoiding inaccurate data collection due to the influence of other parts of the helmet.

[0029] Accelerometers are mainly used to measure the acceleration of head movement. Its working principle is based on Newton's second law, and acceleration information is obtained by detecting the stress or strain generated by internal sensitive components under the action of acceleration. In smart cycling helmets, the accelerometer uses micro-electromechanical system (MEMS) technology, which has the advantages of small size, low power consumption, and high sensitivity. When the rider performs various actions during riding, such as accelerating the head forward, the forward acceleration of the head suddenly decreases when braking, and the head generates lateral acceleration when turning, the mass block inside the accelerometer will be displaced due to the action of acceleration, and the parameters of the components such as capacitors, resistors or inductors connected to it will change accordingly. By measuring the changes in these parameters, the magnitude and direction of the acceleration can be accurately calculated. The accelerometer samples these acceleration changes at a very high frequency, for example, it may collect data hundreds or even thousands of times per second, and record the acceleration data collected each time in the order of the collection time, thus forming a time queue of acceleration changes over time.

[0030] The gyroscope focuses on measuring the angular velocity of the head, that is, the speed and direction of the head rotation. It uses the principle of conservation of angular momentum. The common MEMS gyroscope contains a vibrating object. When the head rotates, according to the action of the Coriolis force, the vibrating object will produce an additional vibration component perpendicular to the vibration direction. By detecting the change of this additional vibration component, the angular velocity of the head rotation can be calculated. During riding, when the rider turns his head to observe the vehicle behind, lowers his head to check the dashboard, or looks up at the traffic lights, the gyroscope can quickly sense these head rotation movements and convert them into corresponding angular velocity data. Similar to the accelerometer, the gyroscope also collects data at a high frequency, and arranges the collected angular velocity data in chronological order to form an angular velocity time queue.

[0031] The data collection process also involves a series of data processing and calibration links. Since the sensor itself may have certain errors, in order to ensure that the collected data is accurate and reliable, the sensor needs to be calibrated. During the helmet production stage, professional calibration equipment and methods will be used to calibrate the accelerometer and gyroscope to determine their zero offset, sensitivity and other parameters, and these calibration parameters will be stored in the helmet's control system. During actual use, the helmet will make real-time corrections to the collected data based on these calibration parameters. At the same time, in order to ensure the continuity and stability of the data, a filtering algorithm is also used to process the collected data to remove noise and interference signals, so that the final acceleration and angular velocity time queue data is smoother and more accurate.

[0032] In addition, in order to ensure that the sensors can work continuously and stably, the smart cycling helmet is also equipped with a special power management system. This system is responsible for providing a stable power supply for the acceleration sensor and gyroscope to ensure that they are always in normal working condition during the entire riding process. At the same time, the power management system also has an energy-saving function, which can automatically adjust the power consumption according to the working status of the sensor, extend the battery life of the helmet, and ensure that data collection will not be interrupted due to insufficient power during long-term riding.

[0033] In step S120, the time queue of the head motion data is transmitted to the microcontroller for riding state abnormality detection to obtain a detection result indicating whether there is riding abnormality, wherein the microcontroller is used to perform key clue deep interactive correlation on the time queue of the head motion data to obtain the head motion data temporal interactive coding feature and perform the riding state abnormality detection based on the head motion data temporal interactive coding feature. Accordingly, considering that the time queue of the head motion data contains key information such as acceleration and angular velocity, this information is crucial for judging the riding state. As the core processing unit of the smart cycling helmet, the microcontroller has the ability to analyze and process data, and can perform further calculations, analysis and judgments on these raw data to extract valuable information for riding state detection.

[0034] Figure 2 is a flow chart of step S120 in the state detection method based on the smart cycling helmet according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 2 As shown, the step S120, transmitting the time queue of the head motion data to the microcontroller processor for riding state abnormality detection to obtain a detection result for indicating whether there is a riding abnormality, includes: S121, data sorting and time series encoding of the time queue of the head motion data to obtain an acceleration time-related coding feature vector and an angular velocity time-related coding feature vector; S122, performing motion parameter time series depth implicit key clue interactive association on the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector to obtain a head motion data time series interactive coding feature vector as the head motion data time series interactive coding feature; S123, obtaining the detection result based on the head motion data time series interactive coding feature vector.

[0035] Specifically, the step S121, performs data sorting and time series encoding on the time queue of the head motion data to obtain the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector. Specifically, in an embodiment of the present application, the time queue of the head motion data is sorted and time series encoded to obtain the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector, including: sorting the time queue of the head motion data according to the sample dimension and the time dimension to obtain the acceleration time queue and the angular velocity time queue; using a DABiLSTM-based time series encoder to respectively encode the acceleration time queue and the angular velocity time queue to obtain the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector.

[0036] More specifically, in an embodiment of the present application, the time queue of the head motion data is sorted according to the sample dimension and the time dimension to obtain the time queue of acceleration and the time queue of angular velocity. It should be understood that, considering the different physical meanings represented by the acceleration and angular velocity data, their respective time variation laws and characteristics are also different. Acceleration reflects the change in speed, while angular velocity reflects the rotation. Therefore, in order to analyze the respective data characteristics more clearly, to analyze and understand the timing separately, in the technical solution of the present application, the time queue of the head motion data is sorted according to the sample dimension and the time dimension to sort the acceleration and angular velocity data into time queues respectively, to obtain the time queue of acceleration and the time queue of angular velocity. That is, by sorting the time queues of acceleration and angular velocity separately, the respective time-related coding features can be extracted more accurately. For example, when analyzing the acceleration time queue, you can focus on the characteristics of the acceleration change trend over time, peak value, fluctuation, etc.; when analyzing the angular velocity time queue, you can focus on the characteristics of the rotation frequency and angle change of the angular velocity. These accurately extracted features have important reference value for subsequent judgment of riding abnormality.

[0037] More specifically, in an embodiment of the present application, a DABiLSTM-based timing encoder is used to encode the time queue of acceleration and the time queue of angular velocity respectively to obtain the acceleration time-associated coding feature vector and the angular velocity time-associated coding feature vector. In particular, a DABiLSTM-based timing encoder is used to encode the time queue of acceleration and the time queue of angular velocity respectively to obtain the acceleration time-associated coding feature vector and the angular velocity time-associated coding feature vector, including: performing forward LSTM encoding based on a feature attention mechanism on the time queue of acceleration to obtain an acceleration time series forward feature vector, including: arranging the time queue of acceleration into an acceleration time series input vector and then performing weight conversion through a multi-layer perceptron to obtain an acceleration weight vector; weighting the acceleration weight vector Renormalization processing, multiplying the normalized acceleration weight vector and the acceleration time series input vector by position to obtain the acceleration time series weighted input vector; inputting the acceleration time series weighted input vector into the forward LSTM model to obtain the acceleration time series forward feature vector; performing backward LSTM encoding based on the time attention mechanism on the acceleration time series forward feature vector to obtain the acceleration time series backward feature vector; performing weighted summation of the acceleration time series forward feature vector and the acceleration time series backward feature vector to obtain the acceleration time association encoding feature vector. Accordingly, considering that the changes of acceleration and angular velocity over time during riding are regular and related, these changes contain important information about the riding state. Although the bidirectional long short-term memory network (BiLSTM) has certain advantages over the unidirectional LSTM in processing time series data, it is still difficult for BiLSTM to effectively transmit early information when facing particularly long time series. In addition, BiLSTM uniformly compresses all input information, making it difficult to learn the overall characteristics of the sequence and its differences in the time dimension. Based on this, in the technical solution of the present application, a DABiLSTM-based temporal encoder is used to encode the acceleration time queue and the angular velocity time queue respectively to obtain an acceleration time-related coding feature vector and an angular velocity time-related coding feature vector. Specifically, the DABiLSTM model can adaptively assign attention weights to input information, thereby highlighting the importance of key header data information, and deeply extracting the overall features of the sequence and mining the temporal correlation between them.

[0038] Specifically, first, the acceleration time queue is encoded by forward LSTM based on the feature attention mechanism to obtain the acceleration time series forward feature vector. Specifically, the acceleration time queue is arranged into an acceleration time series input vector and then weighted by a multi-layer perceptron to obtain an acceleration weight vector. That is to say, the importance of acceleration data at different times to the judgment of riding status may be different. The multi-layer perceptron is a powerful nonlinear model that can learn the complex mapping relationship between the acceleration time series input vector and the weight. The acceleration time series input vector is processed by the multi-layer perceptron, and a corresponding weight is assigned to each input element to form an acceleration weight vector. This enables the model to adaptively adjust the importance of each element according to the characteristics of the data to highlight those data features that are more critical to the judgment of the riding status. Then, the acceleration weight vector is weight normalized, and the normalized acceleration weight vector is multiplied by the acceleration time series input vector by position to obtain the acceleration time series weighted input vector. In this way, more attention can be paid to key data that are given higher weights, thereby improving the sensitivity to important features. Afterwards, the acceleration time series weighted input vector is input into the forward LSTM model to analyze the development trend of the acceleration data from front to back, extract the time series features related to the riding state, and obtain the acceleration time series forward feature vector.

[0039] Next, the forward feature vector of the acceleration time series is encoded using the backward LSTM encoding based on the time attention mechanism to obtain the backward feature vector of the acceleration time series. That is, the backward LSTM model is the opposite of the forward LSTM model, and it processes data from the end of the sequence forward. Based on the time attention mechanism, the model assigns different attention weights to the forward feature vectors at different times, highlighting those parts that have an important impact on the overall features in the time dimension. Through the backward LSTM encoding, the backward feature vector of the acceleration time series is obtained, which contains the time series information from back to front, so as to mine more time correlations hidden in the data.

[0040] Finally, the forward feature vector of the acceleration time series and the backward feature vector of the acceleration time series are weighted and summed to more accurately describe the change of the acceleration data over time, and obtain the acceleration time correlation coding feature vector. In this way, DABiSTM adds a feature attention mechanism before the forward LSTM and a time attention mechanism before the backward LSTM. This dual attention mechanism can deeply extract the overall characteristics of the sequence, explore the time correlation between the acceleration and angular velocity time queues, and provide richer and more valuable information for subsequent accurate judgment of the riding status.

[0041] Specifically, in step S122, the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector are interactively correlated with the motion parameter temporal depth implicit key clues to obtain the head motion data temporal interactive coding feature vector as the head motion data temporal interactive coding feature. Furthermore, it is considered that the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector describe the motion state of the head from different angles respectively. The angular velocity time-related coding feature vector reflects the change in the head motion speed, while the angular velocity time-related coding feature vector reflects the situation of head rotation. In the actual riding process, there is an implicit correlation and mutual influence between the acceleration and angular velocity changes of the head. For example, when a rider makes a sharp turn, the acceleration and angular velocity will change at the same time, and the change relationship between them contains important clues about the riding action and state. Based on this, in order to integrate these two different types of motion information, make full use of the complementarity between the data, dig out the information hidden behind the data, and discover the intrinsic connection between the changes in acceleration and angular velocity, so as to fully understand the movement of the head, in the technical solution of the present application, the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector are subjected to deep implicit key clue interaction of motion parameters to obtain the head motion data time-series interaction coding feature vector as the head motion data time-series interaction coding feature. In particular, the deep implicit key clue interaction of the acceleration and angular velocity time-related coding feature vectors can understand the feature relationship between the two at a deeper level, avoid misjudgment based on surface features alone, significantly improve the accuracy of feature matching, and make the head motion data time-series interaction coding feature vector more accurately reflect the motion state, helping to make riding abnormality judgment more reliable.

[0042] Figure 3 is a flow chart of step S122 in the state detection method based on the smart cycling helmet according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 3As shown, step S122, performs motion parameter temporal depth implicit key clue interactive association on the acceleration time-associated coding feature vector and the angular velocity time-associated coding feature vector to obtain a head motion data temporal interactive coding feature vector, including: S1221, performs feature principal component analysis on the acceleration time-associated coding feature vector to obtain a set of acceleration time-associated principal component coding vectors; S1222, performs implicit key clue anchoring on the set of acceleration time-associated principal component coding vectors and the angular velocity time-associated coding feature vector to obtain a feature pair of {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector}; S1223, based on the {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector} feature pair, performs head motion data feature fine-grained interaction on the acceleration time-associated coding feature vector and the angular velocity time-associated coding feature vector to obtain the head motion data temporal interactive coding feature vector.

[0043] Specifically, in step S1221, characteristic principal component analysis is performed on the acceleration time-correlated coding feature vector to obtain a set of acceleration time-correlated principal component coding vectors. This process can be expressed by the following formula:

[0044]

[0045] in, is the acceleration time-correlated encoding feature vector, is the characteristic principal component analysis operation, It is through The calculated acceleration time-correlated sample covariance matrix, is the acceleration time-correlated principal component orthogonal matrix, , , and are the first, second, and third principal component encoding vectors of the acceleration time correlation. and acceleration time-correlated principal component encoding vector, is the acceleration time-dependent diagonal matrix, , They are and The corresponding eigenvalues ​​are for The transposed matrix of .

[0046] It should be understood that, considering that the acceleration-time correlation coding feature vector contains a lot of complex information, has a high dimension and may be redundant, this will increase the difficulty and amount of calculation for subsequent processing. Therefore, the acceleration-time correlation coding feature vector is subjected to feature principal component analysis to obtain a set of acceleration-time correlation principal component coding vectors. That is, principal component analysis can convert the original features into a set of linearly independent principal components through linear transformation. These principal components are sorted according to the variance contribution size, retaining the main features of the data and achieving data dimensionality reduction. In this way, several principal components that best represent data changes can be extracted from the original acceleration-time correlation coding feature vector, reducing data dimensions, improving computational efficiency, and making subsequent processing pay more attention to key information.

[0047] More specifically, in the embodiment of the present application, the step S1222, anchoring the set of the acceleration time-associated principal component coding vectors and the angular velocity time-associated coding feature vectors with implicit key clues to obtain a feature pair of {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector}, includes:

[0048] Point convolution implicit coding feature extraction is performed on each acceleration time-correlated principal component coding vector in the set of acceleration time-correlated principal component coding vectors to obtain a set of acceleration time-correlated principal component implicit coding vectors. The process can be expressed by the following formula:

[0049]

[0050] in, is the first in the set of acceleration time-correlated principal component encoding vectors acceleration time-correlated principal component encoding vector, is the point convolutional coding, yes function, , , and are the first, second, and third implicit encoding vectors of the acceleration time-correlated principal components. and The acceleration time-correlated principal component implicit encoding vector, is a set of implicit encoding vectors of the acceleration time-correlated principal components;

[0051] The point convolution implicit coding feature extraction is performed on the angular velocity time-related coding feature vector to obtain an angular velocity time-related implicit feature coding vector. The process can be expressed by the following formula:

[0052]

[0053] in, is the point convolutional coding, yes function, is the angular velocity time-dependent encoding feature vector, is the implicit feature encoding vector of angular velocity time correlation;

[0054] The implicit key clue anchoring is performed on the set of the angular velocity time-related implicit feature coding vector and the acceleration time-related principal component implicit coding vector to obtain the {angular velocity time-related implicit feature coding vector, anchored acceleration time-related principal component implicit coding vector} feature pair, and the process can be expressed by the following formula:

[0055]

[0056]

[0057] in, is the first in the set of implicit encoding vectors of the acceleration time correlation principal component The acceleration time-correlated principal component implicit encoding vector, is the angular velocity time-correlated implicit feature encoding vector, represents the inner product, To calculate the Euclidean norm of a vector, is the modulation coefficient, To return the maximum value value, is to find the location of the maximum approximate matching value in the set of implicit encoding vectors of the acceleration time correlation principal component, is the anchored acceleration time-correlated principal component implicit encoding vector, It is a feature pair of {angular velocity time-related implicit feature encoding vector, anchored acceleration time-related principal component implicit encoding vector}.

[0058] Accordingly, considering that the principal component coding vector obtained through principal component analysis has simplified the data, there may still be some complex features that are difficult to be directly used by the model. Point convolution is an effective local feature extraction method, which can mine the local information in each principal component coding vector and extract the key features hidden in the vector. Therefore, point convolution implicit coding feature extraction is performed on each acceleration time-related principal component coding vector in the set of acceleration time-related principal component coding vectors to capture the subtle change patterns and local features of the acceleration time-related principal component coding vectors at different time points, and obtain a set of acceleration time-related principal component implicit coding vectors. These implicit features may be crucial for accurately describing the head movement state.

[0059] Similarly, the point convolution implicit coding feature extraction is performed on the angular velocity time-related coding feature vector to obtain an angular velocity time-related implicit feature coding vector. In this way, the local change pattern and implicit key features of the angular velocity in the time series can be extracted to obtain a feature vector that can more accurately reflect the change of the angular velocity, so as to effectively fuse and analyze it with the acceleration-related features, thereby more comprehensively understanding the head movement state.

[0060] It should be understood that in actual head movement, acceleration and angular velocity are closely related, but their feature vectors come from different sources and have different semantics. Therefore, the implicit key clues are anchored on the set of the angular velocity time-related implicit feature coding vector and the acceleration time-related principal component implicit coding vector to obtain the feature pair of {angular velocity time-related implicit feature coding vector, anchored acceleration time-related principal component implicit coding vector}. That is, the core task of implicit key clue anchoring is to achieve semantic alignment of features from different sources, establish connections between multidimensional feature spaces, and obtain a series of feature pairs with clear semantic associations, which accurately associate the corresponding relationship between acceleration and angular velocity at the implicit feature level.

[0061] More specifically, in the embodiment of the present application, the step S1223, based on the {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector} feature pair, performs head motion data feature fine-grained interaction on the acceleration time-associated coding feature vector and the angular velocity time-associated coding feature vector to obtain the head motion data time series interaction coding feature vector, and the process can be expressed by the following formula:

[0062]

[0063] in, is the acceleration time-correlated encoding feature vector, is the angular velocity time-dependent encoding feature vector, is the angular velocity time-correlated implicit feature encoding vector, is the anchored acceleration time-correlated principal component implicit encoding vector, yes The transposed vector of is the activation function, yes Length, is matrix multiplication, and is the weighted hyperparameter, It is the temporal interaction encoding feature vector of the head motion data.

[0064] Finally, based on the {angular velocity time-related implicit feature coding vector, anchored acceleration time-related principal component implicit coding vector} feature pair, the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector are subjected to fine-grained interaction of head motion data features to obtain the head motion data time series interaction coding feature vector. It should be understood that the multi-head self-attention mechanism can capture long-distance dependencies and multi-dimensional semantic interactions, giving feature fusion higher flexibility, decomposing the granularity differences of feature interactions through parallel streams, and dynamically adjusting the interaction weights to adapt to contextual information, so that the obtained vector integrates the key information and coordinated change characteristics of acceleration and angular velocity in the time series, which can more accurately describe the head motion state and provide a rich and accurate basis for intelligent judgment of riding abnormalities.

[0065] Preferably, in another example of the present application, the step S1223, based on the {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector} feature pair, performs fine-grained interaction of head motion data features on the acceleration time-associated coding feature vector and the angular velocity time-associated coding feature vector to obtain the head motion data time series interaction coding feature vector, including: performing a weak virtualization convergence constraint based on the eigenvalue on the {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector} feature pair to obtain a corrected {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector} feature pair; based on the corrected {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector} feature pair, performs fine-grained interaction of head motion data features on the acceleration time-associated coding feature vector and the angular velocity time-associated coding feature vector to obtain the head motion data time series interaction coding feature vector. The process is expressed by the formula:

[0066]

[0067]

[0068]

[0069] in, yes Middle The eigenvalues ​​at the positions, yes Middle The eigenvalues ​​at the positions, yes The corrected eigenvalues, yes The corrected eigenvalues, is the modified angular velocity time-correlated implicit feature encoding vector, is the implicit encoding vector of the time-correlated principal component of the acceleration after correction and anchoring, It is the temporal interaction encoding feature vector of the head motion data.

[0070] That is, in order to address the alignment ambiguity between high-dimensional time series features from different sources of acceleration and angular velocity caused by source uncertainty, a weakening and virtualization mechanism is introduced to perform weak virtualization power-rate expansion. That is, the 3 / 8 exponent is used as a precursor prior extension to perform power-prior responsive fuzzification convergence on the feature parameter boundary alignment conditions, and the 1 / 4 exponent is used as the main alignment extension to strictly constrain the alignment attenuation relaxation of the feature power-law distribution. In this way, when the alignment boundary condition constraints within the effective range of the association are unclear, the value relevance mechanism is used to avoid the a priori ambiguity of the system behavior under a single mechanism, so as to correct the semantic distribution consistency fuzzification within the alignment interval and enhance the intuitiveness of mining implicit fine-grained interactive associations of anchor features.

[0071] Specifically, the step S123 obtains the detection result based on the head motion data temporal interaction coding feature vector. Specifically, in an embodiment of the present application, the detection result is obtained based on the head motion data temporal interaction coding feature vector, including: inputting the head motion data temporal interaction coding feature vector into a riding state detector based on a classifier to obtain the detection result. That is, the head motion data temporal interaction coding feature vector obtained by the key clue interaction using the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector is classified and processed to intelligently determine whether there is riding abnormality. It should be understood that the classifier can divide the input data into different categories according to the feature pattern of the input data. In the riding state judgment, there are differences in the head motion feature patterns corresponding to normal riding and abnormal riding, and the classifier can distinguish different riding states by learning these differences. The riding state detector based on the classifier can use its powerful classification ability to analyze and judge the head motion data temporal interaction coding feature vector, so as to determine the current riding state.

[0072] In step S130, in response to the detection result that there is a riding abnormality, the alarm mechanism is triggered and a notification containing location information is sent to the preset emergency contact via Bluetooth. That is to say, when a riding abnormality is detected, the rider may be unable to send a distress signal by himself due to injury or other reasons. Triggering the alarm mechanism and sending a notification can actively convey the dangerous situation and let the outside world know that the rider has encountered a problem. As a common short-range wireless communication technology, Bluetooth can communicate with nearby devices (such as mobile phones, etc.) relatively stably and quickly in the application scenario of smart riding helmets to ensure that the notification can be sent in time. Location information is crucial for rescue. After the riding abnormality occurs, it is not enough to just know that the rider is in danger, but also to know its specific location in order to launch an effective rescue operation. By including location information in the notification, the emergency contact can accurately understand the location of the rider so as to take corresponding rescue measures in time. In this way, the outside world can obtain its dangerous situation and location information as soon as possible, so that rescue personnel or emergency contacts can quickly rush to the scene and implement rescue. Timely rescue can greatly increase the survival rate of riders in accidents, reduce the degree of injury, and ensure the life safety of riders.

[0073] Specifically, in response to the detection result that there is riding abnormality, triggering an alarm mechanism and sending a notification containing location information to a preset emergency contact via Bluetooth can be achieved in the following ways:

[0074] First, after detecting riding abnormalities, the system will immediately activate the alarm mechanism. The implementation of this mechanism depends on the hardware devices integrated inside the helmet and the corresponding software programs. The helmet is equipped with a sound-generating device, such as a small speaker or buzzer. When the system receives the detection result of riding abnormalities, it will send instructions to the sound-generating device to control it to emit a sound signal of a specific frequency and rhythm. In order to ensure that the sound can attract the attention of people around, the alarm sound is usually designed to be louder and more recognizable, such as a sharp whistle or a continuous alarm sound. At the same time, in order to meet the needs and usage scenarios of different users, the volume, frequency and other parameters of the alarm sound may allow users to personalize within a certain range.

[0075] In addition to sound alarms, some smart cycling helmets may also be equipped with light alarm devices, such as LED lights. When an abnormality is detected, the system controls the LED light to flash in a specific pattern, such as a fast flashing red light. This visual warning can more effectively attract the attention of others in noisy environments or low-light conditions, increasing the rider's chances of getting help. The mode and color selection of the light alarm are also carefully designed to ensure that it is clearly visible in a variety of scenarios and will not be confused with other conventional light signals.

[0076] The next step is to send a notification containing location information to the preset emergency contact via Bluetooth. In this process, Bluetooth technology plays a key role as a wireless communication bridge. The smart cycling helmet has a built-in Bluetooth module, which has the function of pairing and data transmission with other Bluetooth devices. When the user uses the helmet for the first time or sets up an emergency contact, the helmet's Bluetooth module needs to be paired and connected with the user's mobile phone or other device with communication function. The pairing process usually follows the Bluetooth standard pairing process. The user searches and selects the helmet's Bluetooth device on the mobile phone, enters the preset pairing password (if necessary), and completes the pairing operation.

[0077] After pairing is completed, the system will store the information of the preset emergency contact in the storage unit of the helmet, which includes the contact's phone number, social media account (if notifications are supported through the corresponding platform) or other contact information. When an abnormal riding situation is detected, the microcontroller in the helmet will call the stored emergency contact information and start the notification sending process.

[0078] The acquisition of location information depends on the helmet's built-in global positioning system (GPS) module or other positioning technologies (such as base station-based positioning technology, which can be used as a supplement if the GPS signal is poor). The GPS module receives signals from satellites to calculate the precise geographic location of the helmet, including longitude, latitude, and altitude. After obtaining the location information, the system will encode it and convert it into a data format suitable for transmission via Bluetooth.

[0079] The helmet's Bluetooth module then sends notification data containing location information to the paired phone or other device. On the phone side, a special application needs to be installed to receive and process notifications from the helmet. When the phone receives the notification data transmitted by Bluetooth, the application will parse the data and extract the location information and alarm content. Then, the application will send a notification to the preset emergency contact according to the preset notification method.

[0080] Notifications can be sent in a variety of ways. If the emergency contact is a phone number, the application may call the SMS function of the mobile phone to send a text message containing location information to the contact. The content of the text message usually clearly indicates that the rider has encountered an abnormal situation and provides accurate location information, such as "Your friend encountered an abnormality while riding at [specific location information]. Please check and provide help as soon as possible." If the emergency contact supports receiving notifications through a social media platform, the application may use the API interface of the corresponding platform to send the notification content to the contact in the form of a message. In addition, some applications may also support voice calling functions, which automatically dial the emergency contact's phone while sending text messages or messages, and inform the other party of the rider's abnormal situation and location information through voice prompts.

[0081] To ensure that notifications can be sent successfully, the system also sets up a series of error handling and retry mechanisms. If there is a signal interruption or other failure during Bluetooth transmission, the Bluetooth module will try to reconnect and resend the notification data. If a network problem occurs when sending a text message or message, causing the sending to fail, the application will record the failure and automatically retry sending the notification after the network returns to normal, to maximize the guarantee that emergency contacts can receive notifications in time.

[0082] In summary, the state detection method based on the smart cycling helmet according to the embodiment of the present application is explained, which collects the user's head motion data in real time through the acceleration sensor and gyroscope built into the smart cycling helmet to obtain the time queue of the head motion data (acceleration and angular velocity), and transmits it to the microcontroller, in which the data processing and coding technology based on deep learning is used to sort and time-series encode the head data, so as to intelligently judge whether there is a riding abnormality according to the implicit key clue interaction between the acceleration time-related coding features and the angular velocity time-related coding features after the time-series coding, and when there is an abnormality, an alarm is issued and a location notification is sent at the same time. The present application can capture and analyze the timing changes in acceleration and angular velocity in real time, thereby improving the accuracy of detection. In addition, once a potential riding abnormality is detected, the system will not only issue a local alarm, but also send a notification containing location information to ensure that the outside world can be informed of the situation and provide help in a timely manner, thereby greatly enhancing the safety of riding and the efficiency of emergency response.

[0083] Figure 4 FIG. 1 is a block diagram of a state detection system based on a smart cycling helmet according to an embodiment of the present application. Figure 4 As shown, according to the embodiment of the present application, the state detection system 100 based on the smart cycling helmet includes: a head motion data acquisition module 110, which is used to collect the user's head motion data in real time through the acceleration sensor and the gyroscope built into the smart cycling helmet to obtain a time queue of the head motion data; a riding abnormality detection module 120, which is used to transmit the time queue of the head motion data to the microcontroller processor for riding state abnormality detection to obtain a detection result indicating whether there is a riding abnormality, wherein the microcontroller processor is used to perform key clue deep interactive association on the time queue of the head motion data to obtain the head motion data time series interactive coding feature and perform the riding state abnormality detection based on the head motion data time series interactive coding feature; an abnormal result response module 130, which is used to trigger an alarm mechanism in response to the detection result that there is a riding abnormality and send a notification containing location information to a preset emergency contact via Bluetooth.

[0084] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned state detection system based on the smart cycling helmet have been referred to above. Figures 1 to 3 The description of the state detection method based on the smart cycling helmet has been introduced in detail, and therefore, its repeated description will be omitted.

[0085] As described above, the state detection system 100 based on the smart cycling helmet according to the embodiment of the present disclosure can be implemented in various wireless terminals, such as a server with a state detection algorithm based on the smart cycling helmet. In a possible implementation, the state detection system 100 based on the smart cycling helmet according to the embodiment of the present disclosure can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the state detection system 100 based on the smart cycling 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 state detection system 100 based on the smart cycling helmet can also be one of the many hardware modules of the wireless terminal.

[0086] Alternatively, in another example, the smart cycling helmet-based status detection system 100 and the wireless terminal may also be separate devices, and the smart cycling helmet-based status detection 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.

[0087] In summary, it is intended that the above detailed description is considered to be illustrative rather than restrictive, and it should be understood that the above embodiments should be understood to be only used to illustrate the present invention and not to limit the scope of protection of the present invention. After reading the contents of the present invention, the technician can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A state detection method based on a smart cycling helmet, characterized in that: include: The user's head movement data is collected in real time through the acceleration sensor and gyroscope built into the smart cycling helmet to obtain the time queue of the head movement data; The time queue of the head motion data is transmitted to a microcontroller for riding state abnormality detection to obtain a detection result indicating whether riding abnormality exists, wherein the microcontroller is used to perform key clue deep interactive correlation on the time queue of the head motion data to obtain a temporal interactive coding feature of the head motion data and perform the riding state abnormality detection based on the temporal interactive coding feature of the head motion data; In response to the detection result that there is riding abnormality, an alarm mechanism is triggered and a notification containing location information is sent to a preset emergency contact via Bluetooth.

2. The state detection method based on the smart cycling helmet according to claim 1 is characterized in that: The head movement data includes acceleration and angular velocity.

3. The state detection method based on the smart cycling helmet according to claim 2 is characterized in that: The time queue of the head motion data is transmitted to a microcontroller to perform riding state abnormality detection to obtain a detection result indicating whether there is riding abnormality, including: Performing data sorting and time-series coding on the time queue of the head motion data to obtain an acceleration time-related coding feature vector and an angular velocity time-related coding feature vector; Performing motion parameter temporal depth implicit key clue interactive correlation on the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector to obtain a head motion data temporal interactive coding feature vector as the head motion data temporal interactive coding feature; The detection result is obtained based on the temporal interactive encoding feature vector of the head motion data.

4. The state detection method based on the smart cycling helmet according to claim 3 is characterized in that: The time queue of the head motion data is sorted and time-series encoded to obtain an acceleration time-related encoding feature vector and an angular velocity time-related encoding feature vector, including: Arrange the time queue of the head motion data according to the sample dimension and the time dimension to obtain the time queue of acceleration and the time queue of angular velocity; The acceleration time queue and the angular velocity time queue are respectively encoded using a DABiLSTM-based temporal encoder to obtain the acceleration time-associated encoding feature vector and the angular velocity time-associated encoding feature vector.

5. The state detection method based on the smart cycling helmet according to claim 4 is characterized in that: Using a DABiLSTM-based temporal encoder to encode the acceleration time queue and the angular velocity time queue respectively to obtain the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector, including: The acceleration time queue is subjected to forward LSTM encoding based on a feature attention mechanism to obtain an acceleration time series forward feature vector, including: arranging the acceleration time queue into an acceleration time series input vector and then performing weight conversion through a multi-layer perceptron to obtain an acceleration weight vector; performing weight normalization processing on the acceleration weight vector, multiplying the normalized acceleration weight vector by the acceleration time series input vector by position to obtain an acceleration time series weighted input vector; inputting the acceleration time series weighted input vector into a forward LSTM model to obtain the acceleration time series forward feature vector; Performing backward LSTM encoding based on a temporal attention mechanism on the acceleration time series forward feature vector to obtain an acceleration time series backward feature vector; The acceleration time-related coding feature vector is obtained by performing a weighted summation on the acceleration time series forward feature vector and the acceleration time series backward feature vector.

6. The state detection method based on the smart cycling helmet according to claim 5 is characterized in that: Performing motion parameter temporal depth implicit key clue interactive correlation on the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector to obtain a head motion data temporal interactive coding feature vector, including: Performing characteristic principal component analysis on the acceleration time-correlated coding characteristic vector to obtain a set of acceleration time-correlated principal component coding vectors; Anchoring the set of acceleration time-correlated principal component coding vectors and the angular velocity time-correlated coding feature vectors with implicit key clues to obtain a feature pair of {angular velocity time-correlated implicit feature coding vector, anchored acceleration time-correlated principal component implicit coding vector}; Based on the feature pair of {angular velocity time-related implicit feature coding vector, anchored acceleration time-related principal component implicit coding vector}, the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector are subjected to fine-grained interaction of head motion data features to obtain the head motion data temporal interaction coding feature vector.

7. The state detection method based on the smart cycling helmet according to claim 6 is characterized in that: Anchoring the set of acceleration time-correlated principal component coding vectors and the angular velocity time-correlated coding feature vectors with implicit key clues to obtain a feature pair of {angular velocity time-correlated implicit feature coding vector, anchored acceleration time-correlated principal component implicit coding vector}, including: Performing point convolution implicit coding feature extraction on each acceleration time-correlated principal component coding vector in the set of acceleration time-correlated principal component coding vectors to obtain a set of acceleration time-correlated principal component implicit coding vectors; Performing the point convolution implicit coding feature extraction on the angular velocity time-related coding feature vector to obtain an angular velocity time-related implicit feature coding vector; Anchoring of implicit key clues is performed on the set of the angular velocity time-correlated implicit feature coding vector and the acceleration time-correlated principal component implicit coding vector to obtain the feature pair of {angular velocity time-correlated implicit feature coding vector, anchored acceleration time-correlated principal component implicit coding vector}.

8. The state detection method based on the smart cycling helmet according to claim 7 is characterized in that: Based on the {angular velocity time-related implicit feature coding vector, anchored acceleration time-related principal component implicit coding vector} feature pair, performing head motion data feature fine-grained interaction on the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector to obtain the head motion data time series interaction coding feature vector, including: Performing a weak virtualization convergence constraint based on eigenvalues ​​on the {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector} feature pair to obtain a modified {angular velocity time-associated implicit feature coding vector, anchored acceleration time-associated principal component implicit coding vector} feature pair; Based on the corrected {angular velocity time-related implicit feature coding vector, anchored acceleration time-related principal component implicit coding vector} feature pair, the acceleration time-related coding feature vector and the angular velocity time-related coding feature vector are subjected to fine-grained interaction of head motion data features to obtain the head motion data temporal interaction coding feature vector.

9. The state detection method based on the smart cycling helmet according to claim 8 is characterized in that: The detection result is obtained based on the temporal interaction encoding feature vector of the head motion data, including: inputting the temporal interaction encoding feature vector of the head motion data into a riding state detector based on a classifier to obtain the detection result.

10. A state detection system based on a smart cycling helmet, characterized in that: include: A head motion data acquisition module is used to acquire the user's head motion data in real time through an acceleration sensor and a gyroscope built into the smart cycling helmet to obtain a time queue of the head motion data; A riding anomaly detection module, used for transmitting the time queue of the head motion data to the microcontroller to perform riding state anomaly detection to obtain a detection result indicating whether there is riding anomaly, wherein the microcontroller is used for performing key clue deep interactive correlation on the time queue of the head motion data to obtain a temporal interactive coding feature of the head motion data and performing the riding state anomaly detection based on the temporal interactive coding feature of the head motion data; The abnormal result response module is used to trigger an alarm mechanism and send a notification containing location information to a preset emergency contact via Bluetooth in response to the detection result that there is a riding abnormality.

Images