A wrist-worn sensing device, detection system and method for driver behaviour recognition

Abstract

This invention discloses a wrist-worn sensing device, detection system, and method for driver behavior recognition, belonging to the fields of driver behavior monitoring and intelligent traffic safety technology. The electromagnetic power generation module converts the low-frequency random movements of the driver's wrist into electrical energy, reducing the system's dependence on external power sources or batteries. The triboelectric sensing module utilizes the contact friction between a dielectric rolling element and a conductive electrode to generate a time-series electrical signal, effectively reflecting the characteristics of the driver's wrist movements. The electromagnetic power generation module and the triboelectric sensing module share an eccentric oscillating component, resulting in a compact structure suitable for wearing as a watch or wristband. The device is miniaturized and lightweight, without affecting the driver's normal operation, making it suitable for driving scenarios such as intelligent vehicles, public transportation, and logistics transportation. The rolling friction structure reduces long-term friction and wear, improving the device's durability and output stability. Through a multi-scale convolutional recurrent neural network model, it can effectively identify multiple types of driver behaviors.

Description

Technical Field

[0001] This invention relates to the fields of driver behavior monitoring and intelligent transportation safety technology, and in particular to a wrist-worn sensing device, detection system and method for driver behavior recognition. Background Technology

[0002] With the development of intelligent transportation systems and vehicle-to-everything (V2X) technology, road traffic safety places higher demands on driver status monitoring and driving behavior recognition. When drivers engage in behaviors such as making or receiving phone calls, drinking water, operating the center console, prolonged periods of inactivity, or making abnormal turns while driving, it can easily lead to distraction and delayed reaction time, thereby increasing the risk of traffic accidents. Therefore, real-time, accurate, low-interference, and long-term stable monitoring of driver behavior has significant engineering value and application implications.

[0003] Currently, driver behavior monitoring technologies mainly include in-vehicle visual monitoring, inertial sensor monitoring, steering wheel or seat integrated sensor monitoring, and wearable device monitoring. Among these, in-vehicle visual monitoring is easily affected by factors such as changes in lighting, camera installation angle, obstruction, changes in driver posture, and privacy protection. Traditional inertial sensors or wearable devices usually rely on battery power, which has problems such as limited battery life, the need for frequent charging or battery replacement, and high maintenance costs for long-term wear. Some fixed in-vehicle monitoring devices have drawbacks such as complex wiring, poor adaptability, and high installation costs.

[0004] During driving, the human wrist generates low-frequency, random, and behaviorally characteristic mechanical movements. Electromagnetic power generation technology can convert this low-frequency mechanical energy into electrical energy, while triboelectric nanogenerators or triboelectric sensing technology can convert human movement into electrical signals using contact electrification and electrostatic induction effects. If an electromagnetic energy harvesting structure and a triboelectric sensing structure can be integrated into a wrist-worn device, it is hoped that driver behavior monitoring equipment can achieve self-powering and self-sensing capabilities.

[0005] However, existing technologies still have the following shortcomings: Most existing wearable monitoring devices rely on batteries or external power sources, making it difficult to achieve long-term, low-maintenance operation. Some self-powered devices only have energy harvesting capabilities and lack the ability to highly sense human motion characteristics. Existing energy harvesting modules and sensing modules are often set up separately, resulting in low structural integration and large device size, which is not conducive to wrist-worn use. Traditional driving behavior recognition methods are insufficient in extracting local and global temporal features of non-stationary and nonlinear sensor signals. Existing driving behavior monitoring solutions struggle to simultaneously achieve miniaturization, comfortable wear, self-powered operation, self-sensing, and high-accuracy behavior recognition.

[0006] Therefore, there is an urgent need to propose a wrist-worn sensing device, detection system, and method that is compact, suitable for wrist-worn use, can simultaneously achieve electromagnetic energy harvesting and triboelectric self-sensing, and can be combined with deep learning to achieve high-accuracy driver behavior recognition. Summary of the Invention

[0007] The purpose of this invention is to overcome the technical problems of insufficient battery life, inconvenience of wearing, low integration, insufficient sensing sensitivity, and insufficient accuracy of behavior recognition in the prior art, and to provide a wrist-worn sensing device, detection system, and method for driver behavior recognition.

[0008] To address the aforementioned technical problems, the present invention provides the following technical solution: On the one hand, a wrist-worn sensing device for driver behavior recognition is provided, including: a wrist-worn housing, an eccentric swinging component, an electromagnetic power generation module, and a triboelectric sensing module; The eccentric swinging component is rotatably or swingably disposed within the wrist-worn housing and can rotate or swing under the action of the driver's wrist movement; The electromagnetic power generation module includes at least one magnet disposed on the eccentric swing member and multiple coils disposed in the wrist-worn housing. When the eccentric swing member moves, it drives the magnet to move relative to the coils, causing a change in the magnetic flux in the coils, thereby generating induced electrical energy. The triboelectric sensing module includes a plurality of dielectric rolling elements disposed within or on the eccentric oscillating member, and at least two spaced conductive electrodes disposed on the inner wall of the wrist-worn housing. When the eccentric oscillating member moves, it drives the dielectric rolling elements to contact, roll, or rub against the conductive electrodes to generate triboelectric sensing signals for characterizing the driver's wrist movement.

[0009] As a preferred embodiment of the present invention, the wrist-worn housing is a watch-style or wristband-style structure.

[0010] As a preferred embodiment of the present invention, the eccentric oscillating component is a semi-circular eccentric balance wheel structure, which can rotate or oscillate around the central axis under the combined action of wrist movement, gravity and inertial force.

[0011] As a preferred embodiment of the present invention, the magnet is three permanent magnets, which are arranged alternately and orderly within the eccentric oscillating member.

[0012] As a preferred embodiment of the present invention, there are five coils, which are evenly arranged around the bottom center of the wrist-worn housing.

[0013] As a preferred embodiment of the present invention, the dielectric rolling element is a polytetrafluoroethylene ball, the quantity of which is ten, and the radius is 2mm.

[0014] As a preferred embodiment of the present invention, the conductive electrode is two copper rings, which are symmetrically arranged on the inner wall of the wrist-worn housing.

[0015] As a preferred embodiment of the present invention, it further includes a power management module and a low-power electronic unit. The power management module includes at least one of a rectifier unit, an energy storage unit, a boost unit, and a voltage regulator unit, and is used to process the electrical energy output by the electromagnetic power generation module and supply power to the low-power electronic unit.

[0016] On the other hand, a detection system for driver behavior recognition is provided, including the wrist-worn sensing device, signal acquisition unit, data processing unit and behavior recognition unit described in any of the above claims; The signal acquisition unit is used to acquire the triboelectric sensing signal output by the triboelectric sensing module; The data processing unit is used to filter, denoise, normalize amplitude, slice data and extract features from the triboelectric sensing signal. The behavior recognition unit is used to input the processed signal into a pre-trained multi-scale convolutional recurrent neural network model to output the driver behavior category. The driver behavior categories include at least two of the following: turning left, turning right, making a phone call, not operating the device, drinking water, and operating the center console.

[0017] On the other hand, a method for driver behavior recognition is also provided, comprising the following steps: S1: The driver wears a wrist-worn sensing device as described in any of the above items; S2: The driver's wrist movement drives the eccentric swinging component to rotate or swing, and the eccentric swinging component drives the magnet to move relative to the coil, causing the magnetic flux in the coil to change and generating induced electrical energy; S3: The eccentric oscillating component drives the dielectric rolling body to contact, roll, or rub against the conductive electrode, generating a triboelectric sensing signal. S4: Acquire the triboelectric sensing signal and preprocess the triboelectric sensing signal; S5: Extract the time-frequency features of the preprocessed triboelectric sensing signal, and input the time-frequency features into a multi-scale convolutional recurrent neural network model to output the driver behavior category; The multi-scale convolutional recurrent neural network model includes: Multi-scale convolution module: At least three different sizes of convolution kernels are deployed in parallel, namely 1×1, 3×3 and 5×5 convolution kernels, with a filter ratio of 1:2:1; each convolution kernel is followed by a batch normalization layer and a ReLU activation function layer; the 1×1 convolution kernel is used to capture fine-grained local features, the 3×3 convolution kernel is used to extract medium-scale transition features and balance computational efficiency, and the 5×5 convolution kernel is used to obtain global features with a larger receptive field; Feature fusion layer: The output feature maps of each branch in the multi-scale convolution module are spliced ​​and fused along the channel dimension to form multi-scale fused features; Recurrent Neural Network Layer: A long short-term memory network layer is used to perform temporal encoding on multi-scale fused features and extract temporally dependent feature vectors; Classification output layer: The temporal dependent feature vectors are sequentially input into the fully connected layer and the SoftMax layer to output the probability distribution of each driving behavior category.

[0018] As a preferred embodiment of the present invention, the preprocessing in step S4 includes: S41: Filtering and noise reduction: Bandpass filtering is performed on the triboelectric sensing signal to remove high-frequency noise and low-frequency drift; S42: Amplitude normalization: Normalizes the amplitude of the filtered signal to the range of [0,1] or [-1,1] to eliminate the signal amplitude deviation caused by the difference in the amplitude of different drivers' actions. S43: Sliding window slicing: A fixed-length time sliding window is used to slice the continuous signal to generate several signal samples. The length of a single sample is preferably 1200 data points, and the sliding step size is preferably 100 data points, corresponding to a sampling interval of 0.005 seconds.

[0019] As a preferred embodiment of the present invention, the time-frequency features mentioned in step S5 include instantaneous frequency and power spectral entropy, and the extraction steps are as follows: S51: Apply a short-time Fourier transform to each sliced ​​signal sample to obtain the time spectrum of the signal; S52: Within each time window, identify the frequency corresponding to the maximum power spectrum value, which is taken as the instantaneous frequency at that moment. The instantaneous frequencies at all moments constitute the instantaneous frequency feature vector of the sample. S53: Calculate the power spectral entropy based on the Shannon entropy definition within the same time window; S54: Connect the instantaneous frequencies of all time windows with the power spectral entropy to form a two-dimensional time-frequency feature matrix for the sample.

[0020] Compared with the prior art, the advantages of the present invention are as follows: The electromagnetic power generation module of this invention can convert the low-frequency random movements of the driver's wrist into electrical energy, reducing the system's dependence on external power sources or batteries. The triboelectric sensing module utilizes the contact friction between a dielectric rolling element and a conductive electrode to generate a time-series electrical signal, which can effectively reflect the characteristics of the driver's wrist movements. The electromagnetic power generation module and the triboelectric sensing module share an eccentric oscillating component, resulting in a compact structure suitable for wearing as a watch or wristband. The device is miniaturized and lightweight, without affecting the driver's normal operation, making it suitable for driving scenarios such as intelligent vehicles, public transportation, and logistics transportation. The rolling friction structure can reduce long-term friction and wear, improving the device's durability and output stability. Through a multi-scale convolutional recurrent neural network model, it can effectively identify various types of driver behaviors. It can be extended to fields such as intelligent transportation, vehicle networking, motion monitoring, rehabilitation training, and human-computer interaction. Attached Figure Description

[0021] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a schematic diagram of the structure of a wrist-worn sensing device for driver behavior recognition according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the electromagnetic power generation module structure of a wrist-worn sensing device for driver behavior recognition as described in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the triboelectric sensing module structure of a wrist-worn sensing device for driver behavior recognition according to Embodiment 1 of the present invention; Figure 4 This is a schematic diagram illustrating the working principle of the electromagnetic power generation module of a wrist-worn sensing device for driver behavior recognition as described in Embodiment 1 of the present invention. Figure 5 This is a schematic diagram illustrating the working principle of the triboelectric sensing module of a wrist-worn sensing device for driver behavior recognition as described in Embodiment 1 of the present invention. Figure 6 This is a system block diagram of a detection system for driver behavior recognition according to Embodiment 2 of the present invention; Figure 7 This is a flowchart of a method for driver behavior recognition according to Embodiment 3 of the present invention; Figure 8 This is a flowchart of the signal preprocessing process for a driver behavior recognition method according to Embodiment 3 of the present invention. Figure 9 This is a diagram of the multi-scale convolutional recurrent neural network model structure for a driver behavior recognition method according to Embodiment 3 of the present invention. Reference numerals: 1-Wrist-worn housing, 2-Magnet, 3-Coil, 4-Eccentric oscillating element, 5-Dielectric rolling element, 6-Copper ring, 7-Central shaft. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0023] Example 1 A wrist-worn sensing device for driver behavior recognition, such as Figure 1 As shown, it includes: a wrist-worn housing 1, an eccentric swinging component 4, an electromagnetic power generation module, and a triboelectric sensing module; The eccentric swinging member 4 is rotatably or swingably disposed within the wrist-worn housing 1, and can rotate or swing under the action of the driver's wrist movement; like Figure 2 As shown, the electromagnetic power generation module includes at least one magnet 2 disposed on the eccentric oscillating member 4, and multiple coils 3 disposed within the wrist-worn housing 1, such as... Figure 4 When the eccentric oscillating member 4 is in motion, it drives the magnet 2 to move relative to the coil 3, causing a change in the magnetic flux in the coil 3, thereby generating induced electrical energy. like Figure 3 As shown, the triboelectric sensing module includes a plurality of dielectric rolling elements 5 disposed within or on the eccentric oscillating member 4, and at least two spaced conductive electrodes disposed on the inner wall of the wrist-worn housing 1, such as... Figure 5 As shown, when the eccentric oscillating member 4 moves, it drives the dielectric rolling body 5 to contact, roll, or rub against the conductive electrode to generate a triboelectric sensing signal for characterizing the driver's wrist movement.

[0024] Preferably, the wrist-worn housing 1 is a watch-style or wristband-style structure with an overall size of Φ48mm×14mm.

[0025] Preferably, the eccentric oscillating element 4 is a semi-circular eccentric balance wheel structure, which can rotate or oscillate around the central axis 7 under the combined action of wrist movement, gravity and inertia.

[0026] Preferably, the magnet 2 consists of three permanent magnets with a size of Φ10×6mm, and the three permanent magnets are arranged alternately and orderly within the eccentric oscillating member 3.

[0027] Preferably, there are five coils 3, which are evenly arranged around the bottom center of the wrist-worn housing 1. Each coil has a size of Φ10×4mm and 690 turns.

[0028] Specifically, the wrist-worn housing 1 and the eccentric oscillating component 4 are made of 3D-printed acrylic material. The eccentric oscillating component 4 is a semi-circular eccentric balance wheel structure that can rotate around the central axis of the housing. Because the center of gravity of the eccentric oscillating component 4 is off-axis, when the driver's wrist moves, the eccentric oscillating component 4 will rotate or oscillate under the action of gravity and inertia.

[0029] The electromagnetic power generation module includes three permanent magnets and five coils 3. The three permanent magnets are located inside the eccentric oscillating component 4 and are arranged in an orderly manner with alternating magnetic pole directions. The five coils 3 are located around the center of the bottom of the housing and are evenly arranged circumferentially. When the eccentric oscillating component 4 rotates, the magnets 2 move relative to the coils 3, causing the magnetic flux in the coils 3 to change periodically, thereby generating an induced current.

[0030] Preferably, the dielectric rolling element 5 is a polytetrafluoroethylene ball, numbering ten, with a radius of 2mm.

[0031] Specifically, the triboelectric sensing module includes ten PTFE balls and two copper rings 6. The ten PTFE balls are disposed within an eccentric oscillating element 4, and the two copper rings 6 are symmetrically disposed on the inner wall of the housing. When the eccentric oscillating element 4 rotates, the PTFE balls roll along with it and come into contact with the copper rings 6 through friction. Due to the different triboelectric sequences between the PTFE and the copper rings 6, charge transfer occurs upon contact. When the PTFE balls move between the two copper rings 6, a potential difference is formed between the two copper rings 6, generating an alternating electrical signal. This electrical signal can be used to characterize the driver's wrist movements.

[0032] Preferably, the conductive electrode is two copper rings 6, which are symmetrically arranged on the inner wall of the wrist-worn housing.

[0033] Preferably, it further includes a power management module and a low-power electronic unit. The power management module includes at least one of a rectifier unit, an energy storage unit, a boost unit, and a voltage regulator unit, for processing the electrical energy output by the electromagnetic power generation module and supplying power to the low-power electronic unit.

[0034] Specifically, in this embodiment, the AC power output by the electromagnetic power generation module is converted into DC power by the rectifier unit, stored in the energy storage unit, and outputs a stable voltage through the boost unit and the voltage regulator unit.

[0035] The power management module can supply power to the low-power signal acquisition unit, microcontroller, accelerometer, wireless communication module, or early warning module.

[0036] Under external excitation conditions of 4Hz and 3cm, the open-circuit root mean square voltage of the electromagnetic power generation module can reach about 507mV, the maximum output power is about 0.678mW, and the power density is about 26.78W / m³.

[0037] Example 2 A detection system for driver behavior recognition, such as Figure 6 As shown, it includes the wrist-worn sensing device, signal acquisition unit, data processing unit, and behavior recognition unit described in any one of Embodiment 1; The signal acquisition unit is used to acquire the triboelectric sensing signal output by the triboelectric sensing module; The data processing unit is used to filter, denoise, normalize amplitude, slice data and extract features from the triboelectric sensing signal. The behavior recognition unit is used to input the processed signal into a pre-trained multi-scale convolutional recurrent neural network model to output the driver behavior category. The driver behavior categories include at least two of the following: turning left, turning right, making a phone call, not operating the device, drinking water, and operating the center console.

[0038] Specifically, the driver wears the wrist-worn sensing device on their wrist. During driving, when the driver performs actions such as turning left or right, making a phone call, not operating the device, drinking water, or operating the center console, the wrist movement drives the eccentric oscillating component to move, and the triboelectric sensing module outputs the corresponding triboelectric timing signal.

[0039] The signal acquisition unit acquires the triboelectric timing signal. The data processing unit performs filtering, noise reduction, amplitude normalization, and data slicing on the signal. Subsequently, the instantaneous frequency is extracted through short-time Fourier transform, and the power spectral entropy is calculated based on Shannon entropy to construct time-frequency feature samples.

[0040] The behavior recognition unit inputs the time-frequency feature samples into a pre-trained multi-scale convolutional recurrent neural network model. This model extracts local detail features and global behavioral features through convolutional kernels of different scales, captures temporal dependencies through recurrent neural network layers, and finally outputs the driver behavior category through fully connected layers and SoftMax classification layers.

[0041] Example 3 A method for driver behavior recognition, such as Figure 7 As shown, it includes the following steps: S1: The driver wears a wrist-worn sensing device as described in any of the above items; S2: The driver's wrist movement drives the eccentric swinging component to rotate or swing, and the eccentric swinging component drives the magnet to move relative to the coil, causing the magnetic flux in the coil to change and generating induced electrical energy; S3: The eccentric oscillating component drives the dielectric rolling body to contact, roll, or rub against the conductive electrode, generating a triboelectric sensing signal. S4: Acquire the triboelectric sensing signal and preprocess the triboelectric sensing signal; S5: Extract the time-frequency features of the preprocessed triboelectric sensing signal, and input the time-frequency features into a multi-scale convolutional recurrent neural network model to output the driver behavior category; The improvement of the basic structure of the multi-scale convolutional recurrent neural network model lies in the introduction of a multi-scale convolution module at the front end of the recurrent neural network to simultaneously capture the local detail features and global behavioral features of the signal.

[0042] Specifically, the multi-scale convolutional recurrent neural network model includes a multi-scale convolutional module and a recurrent neural network module; the multi-scale convolutional module is used to extract features at different scales; the recurrent neural network module is used to extract temporally dependent features, such as... Figure 9 As shown, it specifically includes: Multi-scale convolution module: At least three different sizes of convolution kernels are deployed in parallel, namely 1×1, 3×3 and 5×5 convolution kernels, with a filter ratio of 1:2:1; each convolution kernel is followed by a batch normalization layer and a ReLU activation function layer; the 1×1 convolution kernel is used to capture fine-grained local features, the 3×3 convolution kernel is used to extract medium-scale transition features and balance computational efficiency, and the 5×5 convolution kernel is used to obtain global features with a larger receptive field; Feature fusion layer: The output feature maps of each branch in the multi-scale convolution module are spliced ​​and fused along the channel dimension to form multi-scale fused features; Recurrent Neural Network Layer: A long short-term memory network layer is used to perform temporal encoding on multi-scale fused features and extract temporally dependent feature vectors; Classification output layer: The temporal dependent feature vectors are sequentially input into the fully connected layer and the SoftMax layer to output the probability distribution of each driving behavior category.

[0043] Preferably, the preprocessing procedure described in step S4 is as follows: Figure 8 As shown, it includes: S41: Filtering and noise reduction: Bandpass filtering is performed on the triboelectric sensing signal to remove high-frequency noise and low-frequency drift; S42: Amplitude normalization: Normalizes the amplitude of the filtered signal to the range of [0,1] or [-1,1] to eliminate the signal amplitude deviation caused by the difference in the amplitude of different drivers' actions. S43: Sliding window slicing: A fixed-length time sliding window is used to slice the continuous signal to generate several signal samples. The length of a single sample is preferably 1200 data points, and the sliding step size is preferably 100 data points, corresponding to a sampling interval of 0.005 seconds.

[0044] Preferably, the time-frequency features mentioned in step S5 include instantaneous frequency and power spectral entropy, and the extraction steps are as follows: S51: Apply a short-time Fourier transform to each sliced ​​signal sample to obtain the time spectrum of the signal; S52: Within each time window, identify the frequency corresponding to the maximum power spectrum value, which is taken as the instantaneous frequency at that moment. The instantaneous frequencies at all moments constitute the instantaneous frequency feature vector of the sample. S53: Calculate the power spectral entropy based on the Shannon entropy definition within the same time window; S54: Connect the instantaneous frequencies of all time windows with the power spectral entropy to form a two-dimensional time-frequency feature matrix for the sample.

[0045] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0046] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0047] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A wrist-worn sensing device for driver behavior recognition, characterized in that, include: Wrist-worn housing, eccentric oscillating component, electromagnetic power generation module, and triboelectric sensing module; The eccentric swinging component is rotatably or swingably disposed within the wrist-worn housing and can rotate or swing under the action of the driver's wrist movement; The electromagnetic power generation module includes at least one magnet disposed on the eccentric swing member and multiple coils disposed in the wrist-worn housing. When the eccentric swing member moves, it drives the magnet to move relative to the coils, causing a change in the magnetic flux in the coils, thereby generating induced electrical energy. The triboelectric sensing module includes a plurality of dielectric rolling elements disposed within or on the eccentric oscillating member, and at least two spaced conductive electrodes disposed on the inner wall of the wrist-worn housing. When the eccentric oscillating member moves, it drives the dielectric rolling elements to contact, roll, or rub against the conductive electrodes to generate triboelectric sensing signals for characterizing the driver's wrist movement.

2. The wrist-worn sensing device for driver behavior recognition according to claim 1, characterized in that, The wrist-worn housing is either a watch-style or wristband-style structure.

3. The wrist-worn sensing device for driver behavior recognition according to claim 1, characterized in that, The eccentric oscillating component is a semi-circular eccentric balance wheel structure, which can rotate or oscillate around the central axis under the combined action of wrist movement, gravity and inertia.

4. The wrist-worn sensing device for driver behavior recognition according to claim 1, characterized in that, The magnet consists of three permanent magnets, which are arranged alternately and in an orderly manner within the eccentric oscillating component.

5. A wrist-worn sensing device for driver behavior recognition according to claim 1, characterized in that, Five coils are evenly arranged around the bottom center of the wrist-worn housing.

6. A wrist-worn sensing device for driver behavior recognition according to claim 1, characterized in that, The dielectric rolling element is a polytetrafluoroethylene ball, numbering ten, with a radius of 2mm.

7. A wrist-worn sensing device for driver behavior recognition according to claim 1, characterized in that, The conductive electrodes are two copper rings, which are symmetrically arranged on the inner wall of the wrist-worn housing.

8. A wrist-worn sensing device for driver behavior recognition according to claim 1, characterized in that, It also includes a power management module and a low-power electronic unit. The power management module includes at least one of a rectifier unit, an energy storage unit, a boost unit, and a voltage regulator unit, which is used to process the electrical energy output by the electromagnetic power generation module and supply power to the low-power electronic unit.

9. A detection system for driver behavior recognition, characterized in that, Includes the wrist-worn sensing device, signal acquisition unit, data processing unit, and behavior recognition unit as described in any one of claims 1-8; The signal acquisition unit is used to acquire the triboelectric sensing signal output by the triboelectric sensing module; The data processing unit is used to filter, denoise, normalize amplitude, slice data and extract features from the triboelectric sensing signal. The behavior recognition unit is used to input the processed signal into a pre-trained multi-scale convolutional recurrent neural network model to output the driver behavior category. The driver behavior categories include at least two of the following: turning left, turning right, making a phone call, not operating the device, drinking water, and operating the center console.

10. A method for driver behavior recognition, characterized in that, Includes the following steps: S1: The driver wears a wrist-worn sensing device as described in any one of claims 1-8; S2: The driver's wrist movement drives the eccentric swinging component to rotate or swing, and the eccentric swinging component drives the magnet to move relative to the coil, causing the magnetic flux in the coil to change and generating induced electrical energy; S3: The eccentric oscillating component drives the dielectric rolling body to contact, roll, or rub against the conductive electrode, generating a triboelectric sensing signal. S4: Acquire the triboelectric sensing signal and preprocess the triboelectric sensing signal; S5: Extract the time-frequency features of the preprocessed triboelectric sensing signal, and input the time-frequency features into a multi-scale convolutional recurrent neural network model to output the driver behavior category; The multi-scale convolutional recurrent neural network model includes: Multi-scale convolution module: At least three different sizes of convolution kernels are deployed in parallel, namely 1×1, 3×3 and 5×5 convolution kernels, with a filter ratio of 1:2:1; each convolution kernel is followed by a batch normalization layer and a ReLU activation function layer; the 1×1 convolution kernel is used to capture fine-grained local features, the 3×3 convolution kernel is used to extract medium-scale transition features and balance computational efficiency, and the 5×5 convolution kernel is used to obtain global features with a larger receptive field; Feature fusion layer: The output feature maps of each branch in the multi-scale convolution module are spliced ​​and fused along the channel dimension to form multi-scale fused features; Recurrent Neural Network Layer: A long short-term memory network layer is used to perform temporal encoding on multi-scale fused features and extract temporally dependent feature vectors; Classification output layer: The temporal dependent feature vectors are sequentially input into the fully connected layer and the SoftMax layer to output the probability distribution of each driving behavior category.

11. A method for driver behavior recognition according to claim 10, characterized in that, The preprocessing described in step S4 includes: S41: Filtering and noise reduction: Bandpass filtering is performed on the triboelectric sensing signal to remove high-frequency noise and low-frequency drift; S42: Amplitude normalization: Normalizes the amplitude of the filtered signal to the range of [0,1] or [-1,1] to eliminate the signal amplitude deviation caused by the difference in the amplitude of different drivers' actions. S43: Sliding window slicing: A fixed-length time sliding window is used to slice the continuous signal to generate several signal samples. The length of a single sample is preferably 1200 data points, and the sliding step size is preferably 100 data points, corresponding to a sampling interval of 0.005 seconds.

12. The method for driver behavior recognition according to claim 11, characterized in that, The time-frequency features mentioned in step S5 include instantaneous frequency and power spectral entropy, and the extraction steps are as follows: S51: Apply a short-time Fourier transform to each signal sample after slicing to obtain the time spectrum of the signal; S52: Within each time window, identify the frequency corresponding to the maximum power spectrum value, which is taken as the instantaneous frequency at that moment. The instantaneous frequencies at all moments constitute the instantaneous frequency feature vector of the sample. S53: Calculate the power spectral entropy based on the Shannon entropy definition within the same time window; S54: Connect the instantaneous frequencies of all time windows with the power spectral entropy to form a two-dimensional time-frequency feature matrix for the sample.

Images