Electric scooter with holder support and intelligent imaging method thereof

Abstract

The invention discloses an electric scooter with a holder support and an intelligent imaging method thereof, and the method comprises the steps: synchronously collecting multi-modal motion data through a plurality of sensor groups, and generating a motion characteristic parameter set; an active control intention and a random vibration component are analyzed through a motion separation model, a pan-tilt compensation parameter and a control instruction set are generated, a visual semantic analysis module identifies a scene key target and calculates a motion vector of the scene key target, a pan-tilt view field coverage range is dynamically planned in combination with a reinforcement learning strategy, and a view field optimization parameter is output; a layered control framework is adopted to drive a holder motor to execute target attitude adjustment, and a calibrated control signal is output; based on a control signal and a scene semantic analysis result, a multi-frame fusion algorithm is used for eliminating motion blur and generating a high-resolution image, random vibration interference in dynamic shooting of the scooter is remarkably reduced, the imaging stability is improved, and meanwhile, layered anti-disturbance control and scene sensing imaging are fused, so that the stability of the scooter is improved. The mobility of the scooter is guaranteed, and meanwhile the high-quality shooting effect is achieved.

Description

Technical Field

[0001] The present invention relates to the field of imaging technology, and in particular to an electric scooter with a pan / tilt bracket and an intelligent imaging method thereof. Background Art

[0002] With the rapid development of short video creation and outdoor sports photography, electric scooters, thanks to their portability, maneuverability, and ability to carry peripherals, have gradually become a new vehicle for mobile photography. In scenarios such as travel photography, extreme sports documentation, and urban exploration, users tend to use scooters as a mounted camera for first-person perspective dynamic shots, while also using them as a mobile tripod for static composition. However, the multi-dimensional vibrations generated by scooters during driving (such as steering column yaw and pedal bumps), as well as the coupling effect between user control intentions and road disturbances, make it difficult for traditional camera equipment to achieve stable images, severely limiting the practical application value of scooters as mobile photography platforms.

[0003] In existing technologies, imaging stabilization solutions for moving vehicles primarily rely on gimbal active stabilization algorithms or electronic image stabilization technologies based on image processing. For example, gyroscope data feedback is used to control the gimbal motor to compensate for changes in the vehicle's posture. However, this only suppresses low-frequency vibrations in a single dimension and cannot effectively separate the user's active steering / acceleration control components in scooter scenarios from random high-frequency vibrations on the road surface. This can lead to gimbal overcompensation or delayed response, and problems such as image tilt and target loss of focus can still occur during high-speed cornering or bumpy roads. In addition, traditional methods do not combine scene semantics and geographic information to dynamically optimize imaging parameters, making it difficult to adapt to the shooting requirements of complex lighting and fast-moving targets. The final image quality is significantly different from that of professional shooting equipment.

[0004] In view of this, it is necessary to improve the existing technology to solve the technical problems of blurred images and unstable composition caused by multi-dimensional shaking of the vehicle body under high-speed movement and complex road conditions. Summary of the Invention

[0005] The purpose of the present invention is to provide an electric scooter with a pan-tilt bracket and an intelligent imaging method thereof to solve the above technical problems.

[0006] To achieve this object, the present invention adopts the following technical solutions: An intelligent imaging method for an electric scooter with a pan-tilt bracket, comprising: Collect multimodal motion data in real time through a sensor group to generate an aligned set of motion feature parameters; Inputting the motion feature parameter set into a motion separation model, decomposing the active control component and the random vibration component in the scooter motion, generating gimbal compensation parameters and a predicted trajectory baseline, and constructing a gimbal control instruction set; Based on the gimbal control instruction set and the real-time position data of the scooter, the visual semantic analysis module identifies key scene targets and calculates target motion vectors. The reinforcement learning strategy is combined to dynamically plan the gimbal field of view coverage and generate a field of view optimization parameter set. Based on the field of view optimization parameter set, a hierarchical control architecture is used to generate motion trajectory planning signals for the gimbal motor, drive the gimbal to perform target posture adjustment, and simultaneously use a piezoelectric micro-motion mechanism to compensate for high-frequency residual disturbances and output a calibrated gimbal control signal. Based on the gimbal control signal and the identified key scene targets, the captured image is dynamically deblurred using a multi-frame fusion algorithm, and the adaptive imaging optimization model is loaded in combination with the geographic location data to generate an imaging result containing motion trajectory metadata.

[0007] Optionally, the electric scooter includes a pedal and a steering column, and the gimbal bracket is arranged at the top of the steering column; Wherein, a shock-absorbing platform is provided between the steering column and the gimbal bracket.

[0008] Optionally, the sensor group includes: A multi-axis inertial sensor disposed on the steering column generates a motion trajectory of the scooter through the multi-axis inertial sensor and a wheel speed encoder; A pressure sensor provided on the pedal, for obtaining pedal pressure data of changes in the scooter load; A posture encoder installed on the rotary joint of the pan / tilt bracket, used to obtain the current posture angle of the camera; The deformation sensor embedded in the shock-absorbing platform is used to measure the deformation data of the shock-absorbing component under vibration environment; The sensor group operates to obtain multimodal motion data including the motion trajectory of the scooter, pedal pressure data, the attitude angle of the camera and the deformation data of the shock absorbing platform.

[0009] Optionally, the motion feature parameter set is input into a motion separation model to decompose the active control component and the random vibration component in the scooter motion, generate gimbal compensation parameters and a predicted trajectory baseline, and construct a gimbal control instruction set, specifically including: Preprocessing the motion feature parameter set in the time-frequency domain, separating the high-frequency vibration component and the low-frequency motion component in the acceleration data through a filter, and extracting the time series characteristics of the attitude angle change rate to generate a standardized motion input matrix; The motion input matrix is ​​input in parallel to the motion separation model composed of the Kalman filter and the long short-term memory network: The Kalman filter estimates the state of the high-frequency vibration component based on the rigid body dynamics equation of the scooter, and outputs the random vibration compensation vector V^ and the high-frequency acceleration and attitude angle change rate ; The long short-term memory network learns the characteristics of the user's control mode and outputs a predicted trajectory of the active control intention by analyzing the correlation between the steering angular velocity and the pedal pressure data.

[0010] Optionally, the motion input matrix is ​​input in parallel to a motion separation model formed by fusing a Kalman filter and a long short-term memory network, and then the following steps are further included: Construct the scooter-gimbal coupling dynamic equation and calculate the compensation torque τc required by the gimbal: τc=J⋅α+D⋅ω+K⋅δ−τL Where J is the gimbal moment of inertia, D is the damping coefficient, K is the stiffness coefficient, τL is the control torque output by the long short-term memory network, and δ is the deformation data of the shock-absorbing platform; Based on the random vibration compensation vector and the compensation torque τc, the mapping coordinates g(V^, τc) of the compensation parameter matrix C_ of the gimbal three-axis attitude and the target trajectory baseline G_{base}=h(τL) are generated, where h() is a trajectory smoothing function; The compensation parameter matrix C_ and the target trajectory baseline G_{base} are encoded into a motor control protocol to construct a gimbal control instruction set including timestamp synchronization.

[0011] Optionally, based on the gimbal control instruction set and the real-time position data of the scooter, a visual semantic analysis module is used to identify key scene targets and calculate target motion vectors, and a reinforcement learning strategy is combined to dynamically plan the gimbal field of view coverage to generate a field of view optimization parameter set, specifically including the following steps: Receive the pan / tilt control instruction set and the real-time positioning data of the scooter, synchronously start the front camera to capture the environment video stream, and extract the feature point set in the continuous frame image; Input the feature point set into a lightweight semantic recognition network to detect all key target objects in the scene, annotate their category attributes and position bounding boxes, and generate a structured target attribute list; Combined with the real-time gimbal rotation angle data, the motion direction and speed of each target object in the gimbal coordinate system are calculated, and the motion vector deviation is corrected by superimposing the gimbal's own rotation component; According to the importance of target category, motion vector strength and geographical location association rules, the visual weight coefficient of each target object is dynamically assigned to generate a weighted priority target list.

[0012] Optionally, the method further includes dynamically allocating visual weight coefficients of target objects according to target category importance, motion vector strength, and geographic location association rules to generate a weighted priority target list, and then further includes: A field of view decision model is built based on a reinforcement learning strategy, with the optimization goal of maximizing coverage of high-weight targets while minimizing gimbal jitter. The model then outputs preliminary planning values ​​for the gimbal's pitch, horizontal, and zoom parameters. The motion smoothness of the preliminary planning values ​​is optimized, and the field of view optimization parameter set is generated after eliminating the mutation instructions.

[0013] Optionally, based on the field of view optimization parameter set, a hierarchical control architecture is used to generate a motion trajectory planning signal for a gimbal motor, drive the gimbal to perform target posture adjustment, and simultaneously utilize a piezoelectric micro-motion mechanism to compensate for high-frequency residual disturbances, and output a calibrated gimbal control signal, specifically comprising the following steps: Receiving the field of view optimization parameter set, parsing the target pitch angle, horizontal angle and zoom parameters through a hierarchical control decision module, and generating a global motion trajectory planning instruction for the pan / tilt motor; Based on the global motion trajectory instruction, a spline interpolation algorithm is used to calculate the angle-time continuous curve of each axis of the pan / tilt motor, and an acceleration constraint is added to generate a smooth motor drive signal; The controller converts the motor drive signal into a modulation command, drives the gimbal actuator to complete the target posture adjustment, and simultaneously activates the piezoelectric micro-motion mechanism to compensate for high-frequency mechanical vibrations; The attitude encoder of the gimbal is used to collect attitude deviation data in real time, and the motor control quantity is dynamically corrected through the feedback calibration module, and the calibrated gimbal control signal set is output.

[0014] The present invention also provides an electric scooter with a pan-tilt bracket, and the intelligent imaging method of the electric scooter with a pan-tilt bracket as described above is applied. The electric scooter specifically includes: A motion sensor group, integrated into the vehicle body, is used to collect multimodal motion data; An intelligent PTZ system includes a PTZ bracket and a control platform. The PTZ bracket includes a drive bracket and a piezoelectric micro-motion mechanism, and a detachable mounting box provided at the front end of the drive bracket. The control platform includes a built-in motion separation calculation unit, a visual semantic analysis engine and a reinforcement learning strategy controller, as well as an integrated real-time image processor.

[0015] Compared with the existing technology, the present invention has the following advantages: first, a multi-sensor group synchronously collects multimodal motion data to generate an aligned set of motion feature parameters; then, a motion separation model is used to parse the active control intention and random vibration components from the composite motion, generating gimbal compensation parameters and a control instruction set; based on the instruction set and real-time position data, a visual semantic analysis module identifies key scene targets and calculates their motion vectors. Combined with a reinforcement learning strategy, it dynamically plans the gimbal field of view coverage and outputs field of view optimization parameters; then, a hierarchical control architecture is used to drive the gimbal motor to perform target posture adjustment, while a piezoelectric micro-motion mechanism offsets high-frequency residual disturbances and outputs a calibrated control signal; based on the control signal and scene semantic analysis results, a multi-frame fusion algorithm is used to eliminate motion blur, and an adaptive optimization model is loaded with geographic location data to generate high-resolution images containing motion trajectory metadata; through multi-sensor data fusion and motion intention separation mechanisms, this method significantly reduces random vibration interference in dynamic scooter photography and improves imaging stability. Simultaneously, the hierarchical anti-disturbance control and scene-aware imaging are integrated to achieve high-quality photography while ensuring the scooter's maneuverability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] The structures, proportions, sizes, etc. depicted in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not intended to limit the conditions under which the present invention can be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportional relationships, or adjustments in size should still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and objectives that can be achieved by the present invention.

[0018] Figure 1 This is a flow chart of the intelligent imaging method for an electric scooter with a pan-tilt bracket according to the first embodiment of the present invention; Figure 2 This is a second flow chart of the intelligent imaging method for the electric scooter with a pan-tilt bracket according to the first embodiment of the present invention; Figure 3 This is a schematic diagram of the overall structure of the electric scooter with a pan / tilt bracket according to the second embodiment; Figure 4 Schematic diagram of the structure of the pan-tilt bracket of the electric scooter with a pan-tilt bracket according to the second embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0020] In the description of the present invention, it should be understood that the terms "upper," "lower," "top," "bottom," "inner," "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate the description of the present invention and simplify the description. They are not intended to indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. It should be noted that when a component is considered to be "connected" to another component, it may be directly connected to the other component or there may be a centrally located component.

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and through specific implementation methods.

[0022] Example 1: An embodiment of the present invention provides an intelligent imaging method for an electric scooter with a pan / tilt bracket, comprising: S1 uses a sensor array to collect multimodal motion data in real time and generate an aligned set of motion feature parameters. The multimodal sensor array integrated into the scooter simultaneously collects the scooter's triaxial acceleration / angular velocity, displacement velocity, and geographic location. Using timestamp alignment and sensor fusion algorithms, a consistent set of motion feature parameters is generated, providing a highly consistent input source for motion decomposition.

[0023] S2, inputs the motion feature parameter set into the motion separation model, decomposes the active control component and random vibration component in the scooter motion, generates the gimbal compensation parameters and predicted trajectory baseline, and constructs the gimbal control instruction set.

[0024] The feature set is input into the motion separation model (based on a pre-trained LSTM network or state-space model). By analyzing the time-frequency characteristics of the motion signal and the correlation with the control intention, two components are decoupled: Active control components (generated by user steering / acceleration operations) are used to generate the predicted trajectory baseline. Random vibration components (caused by external disturbances such as road roughness) are used to generate gimbal compensation parameters.

[0025] Based on the above decomposition results, a gimbal control instruction set including angle compensation and motion prediction path is constructed.

[0026] S3, based on the gimbal control instruction set and the scooter's real-time position data, uses a visual semantic analysis module to identify key scene targets and calculate target motion vectors. It then combines a reinforcement learning strategy to dynamically plan the gimbal's field of view coverage and generate a field of view optimization parameter set.

[0027] Combining the gimbal control instruction set with real-time position data, the visual semantic analysis module (using lightweight target detection networks such as YOLO) identifies key dynamic targets (such as pedestrians and vehicles) in the captured image and calculates their motion vector (speed / direction) relative to the scooter.

[0028] A reinforcement learning strategy (with field of view coverage as the reward function) dynamically adjusts the gimbal field of view parameters (focal length / viewing angle) according to the target motion vector, generating a set of optimized field of view parameters that balances target tracking stability and scene coverage integrity.

[0029] S4, based on the field of view optimization parameter set, adopts a hierarchical control architecture to generate the motion trajectory planning signal of the gimbal motor, drives the gimbal to perform target posture adjustment, and uses the piezoelectric micro-motion mechanism to compensate for high-frequency residual disturbances and output a calibrated gimbal control signal.

[0030] Based on the field of view optimization parameter set, the hierarchical control architecture performs: Low-frequency trajectory layer: Generates motion trajectory planning signals (PID or model predictive control) for the gimbal motor to achieve macroscopic attitude adjustment of pitch / yaw angles; High-frequency compensation layer: Injects reverse micro-displacement signals through the piezoelectric micro-motion mechanism to offset the residual high-frequency vibration of the gimbal.

[0031] The dual channels collaboratively output calibrated gimbal control signals to ensure the stability of the imaging device under complex movements.

[0032] S5, based on the gimbal control signal and the identified key targets of the scene, dynamically deblurs the captured images through a multi-frame fusion algorithm, loads an adaptive imaging optimization model with geographic location data, and generates imaging results containing motion trajectory metadata.

[0033] The image sensor uses gimbal control signals to compensate for motion, and multi-frame fusion algorithms (such as Wiener filtering or deep learning deblurring) are used to eliminate motion blur. An adaptive imaging optimization model associated with geolocation data is simultaneously loaded (adjusting color and contrast parameters for different scenes, such as urban and rural areas). Ultimately, the resulting image contains metadata, including a motion baseline, timestamp, and positioning information, providing structured data for subsequent motion analysis.

[0034] The working principle of the present invention is as follows: first, a multi-sensor group synchronously collects multimodal motion data to generate an aligned set of motion feature parameters. Then, a motion separation model is used to parse the active control intention and random vibration components from the composite motion, generating gimbal compensation parameters and a control instruction set. Based on the instruction set and real-time position data, the visual semantic analysis module identifies key scene targets and calculates their motion vectors. In combination with a reinforcement learning strategy, the gimbal field of view coverage is dynamically planned and outputs optimized field of view parameters. A hierarchical control architecture is then used to drive the gimbal motor to perform target posture adjustment, while a piezoelectric micro-motion mechanism offsets high-frequency residual disturbances and outputs a calibrated control signal. Based on the control signal and the results of scene semantic analysis, a multi-frame fusion algorithm is used to eliminate motion blur. An adaptive optimization model is loaded with geographic location data to generate a high-resolution image containing motion trajectory metadata. Through multi-sensor data fusion and motion intention separation mechanisms, this method significantly reduces random vibration interference in dynamic scooter photography and improves imaging stability. Simultaneously, the hierarchical anti-disturbance control is integrated with scene-aware imaging to achieve high-quality photography while ensuring the scooter's maneuverability.

[0035] In this embodiment, it is specifically described that step S2 specifically includes: S21, preprocessing the motion feature parameter set in the time-frequency domain, separating the high-frequency vibration component and the low-frequency motion component in the acceleration data through a filter, and extracting the time series characteristics of the attitude angle change rate to generate a standardized motion input matrix.

[0036] Perform time-frequency domain preprocessing on the raw acceleration data collected by the sensor: A bandpass filter is used to separate high-frequency vibration components (>20Hz, reflecting random disturbances on the road surface) from low-frequency motion components (<5Hz, reflecting macroscopic motion of the vehicle body). Synchronously extract the timing features of the attitude angle change rate (gyroscope data) and construct a standardized motion input matrix containing low-frequency acceleration, angular velocity, and timestamp to ensure uniform data dimensions.

[0037] S22, the motion input matrix is ​​input in parallel to the motion separation model composed of the Kalman filter and the long short-term memory network: The Kalman filter estimates the state of the high-frequency vibration component based on the scooter's rigid-body dynamics equations and outputs a random vibration compensation vector V^, as well as high-frequency acceleration α and attitude angle change rate ω. The Kalman filter also estimates the state of the high-frequency vibration component based on the scooter's rigid-body dynamics equations (state-space model) and outputs a random vibration compensation vector V^ and high-frequency motion parameters (high-frequency acceleration / attitude angle change rate).

[0038] The long short-term memory network learns the characteristics of user control patterns and outputs the predicted trajectory of active control intention by analyzing the correlation between steering angular velocity and pedal pressure data.

[0039] S23, construct the scooter-gimbal coupling dynamic equation and calculate the compensation torque required by the gimbal : τc=J⋅α+D⋅ω+K⋅δ−τL Where J is the gimbal moment of inertia, D is the damping coefficient, K is the stiffness coefficient, τL is the control torque output by the long short-term memory network, and δ is the deformation data of the shock-absorbing platform. The gimbal compensation torque that offsets the external disturbance is obtained by solving the equation.

[0040] S24, based on the random vibration compensation vector and the compensation torque τc, generate the mapping coordinates g(V^, τc) of the compensation parameter matrix C_ of the gimbal three-axis attitude and the target trajectory baseline G_{base}=h(τL), where h() is the trajectory smoothing function; Compensation parameter matrix: The random vibration compensation vector V^ and the compensation torque Mapped into three-axis attitude compensation; target trajectory baseline: apply the smoothing function h() to the original predicted trajectory output by LSTM to generate the anti-shake G_{base}.

[0041] S25, encode the compensation parameter matrix C_ and the target trajectory baseline G_{base} into a motor control protocol and construct a gimbal control instruction set including timestamp synchronization.

[0042] In this embodiment, it is specifically explained that step S3 specifically includes the following steps: S31 receives the pan / tilt control instruction set and the real-time positioning data of the scooter, synchronously starts the front camera to capture the environment video stream, and extracts the feature point set in the continuous frame image.

[0043] The system receives the generated gimbal control command set and the scooter's real-time positioning data from the GNSS / IMU, and synchronously triggers the front camera to capture the surrounding video stream. It uses a feature point extraction algorithm (such as ORB or SIFT) to obtain a set of key points in consecutive frames and establish a basis for inter-frame motion correlation.

[0044] S32, inputting the feature point set into a lightweight semantic recognition network, detecting all key target objects in the scene, annotating their category attributes and position bounding boxes, and generating a structured target attribute list; Input the feature point set into a lightweight semantic recognition network (such as the MobileNet-SSD architecture): Detect dynamic / static key targets in the scene (vehicles, pedestrians, traffic signs, etc.) The output is a structured list containing the object category attributes, location bounding box, and confidence score.

[0045] S33, combining the real-time gimbal rotation angle data, calculating the motion direction and speed of each target object in the gimbal coordinate system, and correcting the motion vector deviation by superimposing the gimbal's own rotation component; Basic vector calculation: Calculate the original motion direction and speed in the image coordinate system based on the displacement of the target bounding box between consecutive frames.

[0046] Gimbal motion compensation: This system integrates the real-time gimbal rotation angle data (pitch / yaw angular velocity) and converts the original vectors to the earth coordinate system through a coordinate transformation matrix, eliminating observation errors introduced by the gimbal's own motion.

[0047] S34, dynamically assigning visual weight coefficients to target objects based on target category importance, motion vector strength, and geographic location association rules, and generating a weighted priority target list.

[0048] Constructing priority decision rules: Category importance: traffic participants (e.g., cars > pedestrians) > static objects; Motion threat: targets approaching the scooter are given a higher weight; Geographic association rules: Call the regional traffic rules library based on positioning data (such as increasing the weight of pedestrians in school areas).

[0049] Output a target list with visual weight coefficients to ensure that key targets are covered first.

[0050] S35 builds a field of view decision model based on reinforcement learning strategy, with the optimization goal of maximizing coverage of high-weight targets while minimizing gimbal jitter, and outputs preliminary planning values ​​for the gimbal pitch angle, horizontal angle, and zoom parameters.

[0051] S36, performing motion smoothness optimization processing on the preliminary planning values, eliminating mutation instructions and generating a field of view optimization parameter set.

[0052] The output parameters are smoothed by applying a sliding mean filter to eliminate angle mutations, and a rate limit is imposed on the zoom parameters to generate a field of view optimization parameter set that meets the mechanical response constraints.

[0053] In this embodiment, it is specifically explained that step S4 specifically includes the following steps: S41, receiving the field of view optimization parameter set, parsing the target pitch angle, horizontal angle and zoom parameters through the hierarchical control decision module, and generating the global motion trajectory planning instructions of the gimbal motor.

[0054] This step uses a hierarchical control decision module to analyze the field of view optimization parameter set (target horizontal angle, pitch angle, and zoom parameters), verify that the parameters meet the physical limits of the gimbal (such as ±180° for horizontal rotation and -30° to +90° for pitch). It also automatically limits the pitch angle velocity for high-magnification zoom scenarios and generates global motion trajectory planning instructions to ensure that the target posture can be safely executed.

[0055] S42, based on the global motion trajectory instruction, uses a spline interpolation algorithm to calculate the angle-time continuous curve of each axis of the gimbal motor, and adds acceleration constraints to generate a smooth motor drive signal.

[0056] Based on the global motion trajectory instructions, the cubic spline interpolation algorithm is used to convert discrete angle points into a continuous and smooth angle-time curve. At the same time, angular acceleration constraints and gimbal rotational inertia compensation factors are added to generate motor drive signals, significantly reducing the mechanical impact caused by motion mutations and improving the smoothness of gimbal motion.

[0057] S43, converting the motor drive signal into a modulation command through the controller, driving the gimbal actuator to complete the target posture adjustment, and synchronously activating the piezoelectric micro-motion mechanism to compensate for high-frequency mechanical vibration; The sliding mode variable structure controller converts the smooth motor drive signal into a high-precision pulse width modulation instruction to drive the gimbal motor to perform target attitude adjustment; the piezoelectric micro-motion mechanism is synchronously activated, and its millisecond-level response characteristics are used to compensate for high-frequency mechanical vibrations that the motor cannot suppress (such as >30Hz disturbances transmitted by road bumps), forming a dual-mode anti-disturbance mechanism with macro-positioning of the main motor and nano-level adjustment of the piezoelectric micro-motion.

[0058] S44, using the attitude encoder of the gimbal to collect attitude deviation data in real time, dynamically correcting the motor control amount through the feedback calibration module, and outputting a calibrated gimbal control signal set.

[0059] The gimbal attitude encoder is used to provide real-time feedback of actual angle data and calculate the attitude deviation between the target value and the actual value. When the deviation exceeds the set threshold (horizontal > 1.5° or pitch > 1.0°), the motor control variable is dynamically corrected through the proportional-integral-differential algorithm, and a calibrated gimbal control signal set is output to resolve the problem of control accuracy degradation caused by load changes or external interference, ensuring that the final execution attitude is consistent with the planned instructions.

[0060] Example 2: The present invention also provides an electric scooter with a pan-tilt bracket, and the intelligent imaging method of the electric scooter with a pan-tilt bracket as in the first embodiment is applied. The electric scooter specifically includes: The motion sensor group is integrated into the vehicle body and is used to collect multimodal motion data; the intelligent pan-tilt system includes a pan-tilt bracket 10 and a control platform 20. The pan-tilt bracket 10 includes a drive bracket 11 and a piezoelectric micro-motion mechanism 12, as well as a detachable mounting box arranged at the front end of the drive bracket 11 for detachably installing the camera, which can adopt a magnetic or snap-on positioning method.

[0061] The control platform 20 includes a built-in motion separation calculation unit, a visual semantic analysis engine and a reinforcement learning strategy controller, as well as an integrated real-time image processor.

[0062] In this embodiment, it is specifically described that the electric scooter 100 includes a pedal 101 and a steering column 102, and the gimbal bracket 10 is arranged at the top of the steering column; wherein a shock-absorbing platform is provided between the steering column and the gimbal bracket 10.

[0063] In this embodiment, the core mechanical structure of the electric scooter 100 includes a pedal 101 and a steering column 102, wherein a gimbal bracket 10 is fixedly mounted on the top of the steering column 102, forming a supporting platform for the filming equipment. A shock-absorbing platform is provided between the steering column and the gimbal bracket 10. This platform adopts a rubber-spring composite buffer structure, which can effectively absorb the multi-directional vibration (especially high-frequency bumps) transmitted by the steering column during the scooter's operation, reduce the vibration amplitude of the gimbal bracket 10, and provide a stable foundation for dynamic filming.

[0064] To further specify, the sensor group includes: a multi-axis inertial sensor set on the steering column, which generates the motion trajectory of the scooter through the multi-axis inertial sensor and wheel speed encoder; a pressure sensor set on the pedal 101, which is used to obtain pedal pressure data of the scooter load change; an attitude encoder installed on the rotary joint of the gimbal bracket 10, which is used to obtain the current attitude angle of the camera; and a deformation sensor embedded in the shock-absorbing platform, which is used to measure the deformation data of the shock-absorbing component in a vibration environment.

[0065] The sensor group operates to obtain multimodal motion data including the motion trajectory of the scooter, pedal pressure data, the attitude angle of the camera, and the deformation data of the shock absorbing platform.

[0066] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent imaging method for an electric scooter with a pan-tilt bracket, characterized in that: include: Collect multimodal motion data in real time through a sensor group to generate an aligned set of motion feature parameters; Inputting the motion feature parameter set into a motion separation model, decomposing the active control component and the random vibration component in the scooter motion, generating gimbal compensation parameters and a predicted trajectory baseline, and constructing a gimbal control instruction set; Based on the gimbal control instruction set and the real-time position data of the scooter, the visual semantic analysis module identifies key targets in the scene and calculates the target motion vector. The reinforcement learning strategy is combined to dynamically plan the gimbal field of view coverage and generate a field of view optimization parameter set. Based on the field of view optimization parameter set, a hierarchical control architecture is used to generate motion trajectory planning signals for the gimbal motor, drive the gimbal to perform target posture adjustment, and simultaneously use a piezoelectric micro-motion mechanism to compensate for high-frequency residual disturbances and output a calibrated gimbal control signal. Based on the gimbal control signal and the identified key scene targets, the captured image is dynamically deblurred using a multi-frame fusion algorithm, and the adaptive imaging optimization model is loaded in combination with the geographic location data to generate an imaging result containing motion trajectory metadata.

2. The intelligent imaging method for an electric scooter with a pan-tilt bracket according to claim 1, characterized in that: The electric scooter includes a pedal and a steering column, and the pan / tilt bracket is arranged on the top end of the steering column; Wherein, a shock-absorbing platform is provided between the steering column and the gimbal bracket.

3. The intelligent imaging method for an electric scooter with a pan / tilt bracket according to claim 2, characterized in that: The sensor group includes: A multi-axis inertial sensor disposed on the steering column generates a motion trajectory of the scooter through the multi-axis inertial sensor and a wheel speed encoder; A pressure sensor provided on the pedal, for obtaining pedal pressure data of changes in the scooter load; A posture encoder installed on the rotary joint of the pan / tilt bracket, used to obtain the current posture angle of the camera; The deformation sensor embedded in the shock-absorbing platform is used to measure the deformation data of the shock-absorbing component under vibration environment; The sensor group operates to obtain multimodal motion data including the motion trajectory of the scooter, pedal pressure data, the attitude angle of the camera and the deformation data of the shock absorbing platform.

4. The intelligent imaging method for an electric scooter with a pan / tilt bracket according to claim 1, characterized in that: The motion feature parameter set is input into the motion separation model to decompose the active control component and random vibration component in the scooter motion, generate gimbal compensation parameters and predicted trajectory baseline, and construct a gimbal control instruction set, specifically including: Preprocessing the motion feature parameter set in the time-frequency domain, separating the high-frequency vibration component and the low-frequency motion component in the acceleration data through a filter, and extracting the time series characteristics of the attitude angle change rate to generate a standardized motion input matrix; The motion input matrix is ​​input in parallel to the motion separation model composed of the Kalman filter and the long short-term memory network: The Kalman filter estimates the state of the high-frequency vibration component based on the rigid body dynamics equation of the scooter, and outputs the random vibration compensation vector V^ and the high-frequency acceleration and attitude angle change rate ; The long short-term memory network learns the characteristics of the user's control mode and outputs a predicted trajectory of the active control intention by analyzing the correlation between the steering angular velocity and the pedal pressure data.

5. The intelligent imaging method for an electric scooter with a pan / tilt bracket according to claim 4, characterized in that: The motion input matrix is ​​input in parallel to a motion separation model formed by the fusion of a Kalman filter and a long short-term memory network, and then further includes: Construct the scooter-gimbal coupling dynamic equation and calculate the compensation torque τc required by the gimbal: τc=J⋅α+D⋅ω+K⋅δ−τL Where J is the gimbal moment of inertia, D is the damping coefficient, K is the stiffness coefficient, τL is the control torque output by the long short-term memory network, and δ is the deformation data of the shock-absorbing platform; Based on the random vibration compensation vector and the compensation torque τc, the mapping coordinates g(V^, τc) of the compensation parameter matrix C_ of the gimbal three-axis attitude and the target trajectory baseline G_{base}=h(τL) are generated, where h() is a trajectory smoothing function; The compensation parameter matrix C_ and the target trajectory baseline G_{base} are encoded into a motor control protocol to construct a gimbal control instruction set including timestamp synchronization.

6. The intelligent imaging method for an electric scooter with a pan-tilt bracket according to claim 1, characterized in that: Based on the gimbal control instruction set and the real-time position data of the scooter, the visual semantic analysis module identifies key scene targets and calculates target motion vectors. Combined with the reinforcement learning strategy, the gimbal field of view coverage is dynamically planned to generate a field of view optimization parameter set. Specifically, the following steps are included: Receive the pan / tilt control instruction set and the real-time positioning data of the scooter, synchronously start the front camera to capture the environment video stream, and extract the feature point set in the continuous frame image; Input the feature point set into a lightweight semantic recognition network to detect all key target objects in the scene, annotate their category attributes and position bounding boxes, and generate a structured target attribute list; Combined with the real-time gimbal rotation angle data, the motion direction and speed of each target object in the gimbal coordinate system are calculated, and the motion vector deviation is corrected by superimposing the gimbal's own rotation component; According to the importance of target category, motion vector strength and geographical location association rules, the visual weight coefficient of each target object is dynamically assigned to generate a weighted priority target list.

7. The intelligent imaging method for an electric scooter with a pan-tilt bracket according to claim 1, characterized in that: The method dynamically allocates visual weight coefficients to target objects based on target category importance, motion vector strength, and geographic location association rules to generate a weighted priority target list, and then further includes: A field of view decision model is built based on a reinforcement learning strategy, with the optimization goal of maximizing coverage of high-weight targets while minimizing gimbal jitter. The model then outputs preliminary planning values ​​for the gimbal's pitch, horizontal, and zoom parameters. The motion smoothness of the preliminary planning values ​​is optimized, and the field of view optimization parameter set is generated after eliminating the mutation instructions.

8. The intelligent imaging method for an electric scooter with a pan-tilt bracket according to claim 1, characterized in that: Based on the field of view optimization parameter set, a hierarchical control architecture is used to generate motion trajectory planning signals for the gimbal motor, drive the gimbal to perform target posture adjustment, and simultaneously use a piezoelectric micro-motion mechanism to compensate for high-frequency residual disturbances, and output a calibrated gimbal control signal. Specifically, the following steps are included: Receiving the field of view optimization parameter set, parsing the target pitch angle, horizontal angle and zoom parameters through a hierarchical control decision module, and generating a global motion trajectory planning instruction for the pan / tilt motor; Based on the global motion trajectory instruction, a spline interpolation algorithm is used to calculate the angle-time continuous curve of each axis of the pan / tilt motor, and an acceleration constraint is added to generate a smooth motor drive signal; The controller converts the motor drive signal into a modulation command, drives the gimbal actuator to complete the target posture adjustment, and simultaneously activates the piezoelectric micro-motion mechanism to compensate for high-frequency mechanical vibrations; The attitude encoder of the gimbal is used to collect attitude deviation data in real time, and the motor control quantity is dynamically corrected through the feedback calibration module, and the calibrated gimbal control signal set is output.

9. An electric scooter with a pan / tilt bracket, characterized in that: The intelligent imaging method of an electric scooter with a pan-tilt bracket according to any one of claims 1 to 8 is applied, wherein the electric scooter specifically comprises: A motion sensor group, integrated into the vehicle body, is used to collect multimodal motion data; An intelligent PTZ system includes a PTZ bracket and a control platform. The PTZ bracket includes a drive bracket and a piezoelectric micro-motion mechanism, and a detachable mounting box provided at the front end of the drive bracket. The control platform includes a built-in motion separation calculation unit, a visual semantic analysis engine and a reinforcement learning strategy controller, as well as an integrated real-time image processor.

Images