5G intelligent riding camera anti-shake method and system based on micro holder

Abstract

The invention discloses a 5G intelligent riding camera anti-shake method and system based on a micro holder, and belongs to the technical field of camera shooting. According to the method, the micro-holder and the data sensor are integrated in the riding camera, a specific structure jitter detection model (including an RCNN and a Bi-LSTM network) is constructed through calibration and data fusion to obtain the jitter level and direction, the posture of the micro-holder is adaptively adjusted according to the jitter level and direction to achieve jitter compensation, and an image intelligent algorithm can be used for recognizing a riding scene and automatically adjusting shooting parameters. According to the invention, the anti-shake precision and stability of the riding camera can be obviously improved, the method adapts to a complex riding environment, shooting parameters are automatically adjusted to improve the shooting quality and convenience, and better experience is provided for riding shooting.

Description

Technical Field

[0001] The present invention relates to the field of camera technology, and more specifically to a 5G smart cycling camera anti-shake method and system based on a micro-pan head. Background Art

[0002] Currently, with the rapid development of 5G technology and the popularization of smart devices, cycling enthusiasts are increasingly demanding smart cycling cameras that can record high-quality, stable videos. Such cameras not only need to have high-definition shooting capabilities, but also should be able to effectively reduce image blur and shaking caused by vibration during cycling, especially on bumpy roads, so as to provide a smooth and clear video recording experience.

[0003] However, traditional anti-shake technologies, such as optical image stabilization or electronic image stabilization, often fail to achieve the desired anti-shake effect when faced with high-frequency vibrations under extreme riding conditions. The optical image stabilization structure is relatively complex and costly, and its anti-shake compensation range is limited, making it difficult to effectively eliminate severe shaking during riding; although electronic image stabilization can correct the image to a certain extent through algorithms, it will reduce the image resolution and detail performance to a certain extent, and cannot meet users' pursuit of high-definition image quality.

[0004] Therefore, how to propose a 5G smart cycling camera anti-shake method and system based on a micro-gimbal to achieve stable and high-definition shooting picture output is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention provides a 5G smart cycling camera anti-shake method and system based on a micro-pan head, which can keep the video picture stable and clear even in high-speed movement and complex environments.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] On the one hand, the present invention proposes a 5G smart cycling camera anti-shake method based on a micro-pan head, comprising the following steps:

[0008] Install the micro gimbal between the lens and image sensor in the cycling camera and integrate the data sensor;

[0009] Calibrate the data sensor, use the calibrated data sensor to acquire sensor data and perform data fusion to obtain fused data;

[0010] Constructing a jitter detection model, inputting the fused data into the jitter detection model, and obtaining a jitter detection result;

[0011] The posture of the micro gimbal is adaptively adjusted according to the jitter detection result, thereby achieving jitter compensation for the riding camera.

[0012] Preferably, calibrating the data sensor includes:

[0013] Collecting the measured values ​​of the data sensor and the corresponding true values;

[0014] Calculate error characteristic data of the data sensor based on the measured value and the true value;

[0015] Based on the error signature data, the data sensor is calibrated using a machine learning algorithm.

[0016] Preferably, using a calibrated data sensor to acquire sensor data and perform data fusion to obtain fused data includes:

[0017] Preprocessing the sensor data to obtain data to be fused;

[0018] Extracting characteristic values ​​of the data to be fused, and performing comprehensive quality assessment based on the characteristic values;

[0019] Dynamically adjust the Kalman filter according to the comprehensive quality evaluation result to optimally estimate the data to be fused;

[0020] The weight coefficient of the data sensor is adjusted through the Kalman filter residual to perform data fusion and obtain fused data.

[0021] Preferably, the jitter detection model includes an RCNN network and a Bi-LSTM network connected in sequence;

[0022] The RCNN network is composed of a layer of feedforward convolutional network and a multi-layer RCL network; the multi-layer RCL network is a cyclic network composed of multiple iterative convolutional networks; the multi-layer RCL network traverses the path lengths of all convolutional networks and outputs the initial features of the fused data;

[0023] Input the initial features into the Bi-LSTM network, obtain the time series information of the fused data, process the time series information using the self-attention mechanism, and output the jitter detection result;

[0024] The jitter detection result includes different jitter levels and jitter directions.

[0025] Preferably, the posture of the micro gimbal is adaptively adjusted according to the jitter detection result, thereby realizing jitter compensation of the riding camera, including:

[0026] Determining the adjustment angle of the micro gimbal according to the mapping relationship between the jitter level and the adjustment amplitude;

[0027] Determining the adjustment direction of the micro gimbal according to the shaking direction;

[0028] The driving mechanism of the micro gimbal adjusts the posture of the micro gimbal according to the adjustment angle and the adjustment direction.

[0029] Preferably, a 5G smart cycling camera anti-shake method based on a micro-pan platform also includes:

[0030] Use image intelligent algorithms to identify cycling scenes and automatically adjust camera shooting parameters based on the recognition results.

[0031] On the other hand, the present invention also proposes a 5G smart cycling camera anti-shake system based on a micro-pan head, which is used to implement the above-mentioned 5G smart cycling camera anti-shake method based on a micro-pan head, comprising: a micro-pan head, a cycling camera and a control module, wherein the micro-pan head is installed between the lens and the image sensor in the cycling camera and integrated with a data sensor; the control module is communicatively connected with the micro-pan head and the cycling camera;

[0032] The control module comprises:

[0033] A data processing unit, used to calibrate the data sensor, obtain sensor data using the calibrated data sensor, and perform data fusion to obtain fused data;

[0034] A jitter detection unit, used for constructing a jitter detection model, inputting the fused data into the jitter detection model, and obtaining a jitter detection result;

[0035] The shake compensation unit is used to adaptively adjust the posture of the micro gimbal according to the shake detection result, thereby realizing shake compensation of the riding camera.

[0036] Preferably, the control module further includes:

[0037] The scene recognition unit is used to use image intelligent algorithms to recognize riding scenes and automatically adjust camera shooting parameters according to the riding scene recognition results.

[0038] It can be seen from the above technical solutions that compared with the prior art, the present invention discloses a 5G intelligent cycling camera anti-shake method and system based on a micro-pan head, which integrates a micro-pan head and a data sensor inside the cycling camera, obtains the jitter level and direction through calibration, data fusion, and construction of a specific structure jitter detection model (including RCNN and Bi-LSTM network), and adaptively adjusts the micro-pan head posture to achieve jitter compensation, and can also use image intelligent algorithms to identify cycling scenes and automatically adjust shooting parameters. The present invention can significantly improve the anti-shake accuracy and stability of cycling cameras, adapt to complex cycling environments, automatically adjust shooting parameters to improve shooting quality and convenience, and provide a better experience for cycling shooting. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0040] Figure 1 A flow chart of the method provided by the present invention;

[0041] Figure 2 An exploded view of the micro gimbal structure of the present invention;

[0042] Figure 3 It is a structural diagram of the jitter detection model of the present invention;

[0043] Figure 4 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0045] The embodiment of the present invention discloses a 5G smart cycling camera anti-shake method based on a micro-pan head, such as Figure 1 As shown, the method comprises the following steps:

[0046] S1. Install the micro gimbal between the lens and image sensor in the cycling camera and integrate the data sensor.

[0047] In cycling cameras, the micro gimbal integrates double L flexible circuit boards, tilt springs, micro gimbal voice coil motors and other components (reference Figure 2 ) work between the lens and the image sensor. When the camera is shaken, these components work together to compensate for the shake and ensure a stable picture.

[0048] The double-L flexible circuit board plays a key role in signal transmission and control in the whole process, transmitting the control signal to the micro-pan-tilt voice coil motor, which drives the micro-pan-tilt according to the control signal. The tilt spring is used to assist the smooth movement of the micro-pan-tilt and buffer vibration. When the micro-pan-tilt voice coil motor adjusts the angle according to the received command, the tilt spring will use its own elastic characteristics to reduce the impact force caused by rapid adjustment.

[0049] Choose a suitable micro gimbal model, the size of which should match the internal space of the cycling camera, ensuring that it can be firmly installed between the lens and the image sensor without affecting the normal operation of other camera components.

[0050] Design a special mounting bracket or fixing structure to ensure that the micro gimbal is accurately positioned and securely fixed in the camera. You can use screws, buckles or other fixing methods to tightly connect the micro gimbal to the camera body, and at the same time consider easy installation and removal so that it can be easily operated when the micro gimbal needs to be repaired or replaced. During installation, ensure that the optical axis of the micro gimbal is strictly consistent with the optical axis of the camera lens to avoid image offset or deformation due to installation deviation.

[0051] High-precision data sensors such as gyroscope sensors and acceleration sensors are selected. The gyroscope sensor is installed near the center of gravity of the camera to obtain more accurate angular velocity information; the acceleration sensor is arranged in different directions as needed to comprehensively monitor the acceleration changes of the camera in all directions.

[0052] S2. Calibrate the data sensor, use the calibrated data sensor to obtain sensor data and perform data fusion to obtain fused data, including:

[0053] S21. Collect the measured values ​​of the data sensor and the corresponding true values.

[0054] In a laboratory environment, high-precision measuring equipment is used as a reference standard to collect measurement values ​​of data sensors. For acceleration sensors, the camera is fixed on a test platform that accurately controls acceleration. The platform can generate acceleration of known magnitude and direction, and record the acceleration value measured by the sensor. For gyroscope sensors, a high-precision rotating table is used to rotate the camera at a known angular velocity to obtain the angular velocity data measured by the sensor.

[0055] Perform multiple measurements under different environmental conditions (such as temperature, humidity, etc.) and motion states (such as static, uniform motion, accelerated motion, etc.) to obtain comprehensive measurement data. At the same time, ensure that the collected measurement values ​​are accurately matched with the corresponding true values ​​to establish a detailed measurement data set.

[0056] S22. Calculate error characteristic data of the data sensor based on the measured value and the true value.

[0057] Statistical analysis methods are used to calculate the error characteristic data of the sensor. The mean, variance, standard deviation and other statistics of the deviation (error) between the measured value and the true value are calculated, and the systematic error and random error characteristics of the sensor are reflected based on these statistics.

[0058] Analyze the variation of errors with time, temperature, motion state and other factors, and establish an error model. Specifically, it is found that the error of some sensors increases when the temperature rises, and a linear or nonlinear relationship model between temperature and error can be established so that the sensor can be compensated according to the actual ambient temperature during calibration.

[0059] S23. Calibrate the data sensor using a machine learning algorithm based on the error characteristic data.

[0060] Select a machine learning algorithm such as a neural network or a support vector machine, use the error feature data as input, and train the calibration model. In this embodiment, a neural network is used as an example to construct a neural network structure including an input layer (error feature data), a hidden layer (several neurons are used to learn the complex relationship between the error and the calibration parameters), and an output layer (calibrated sensor data or calibration parameters).

[0061] The calibration model is trained using the collected measurement data set, and the model parameters (such as the weights and biases of the neural network) are adjusted to enable the model to accurately predict the calibrated sensor data based on the input error feature data. During the training process, the mean square error loss function and stochastic gradient descent method are used to optimize the model performance until the preset calibration accuracy is achieved.

[0062] S24. Preprocess the sensor data to obtain data to be fused.

[0063] The collected sensor data is filtered to remove noise interference. For gyroscope sensor data, since it mainly reflects the angular velocity changes of the camera, there may be high-frequency noise. Low-pass filtering can be used to remove high-frequency interference. For acceleration sensor data, there may be abnormal values ​​due to sudden interference. Median filtering can effectively remove these abnormal values ​​while retaining the change trend of the data.

[0064] Perform data interpolation and time synchronization to ensure that the data from different sensors are consistent in time and value. Since the sampling frequencies of different sensors may be different, the data are aligned on the time axis through interpolation algorithms (such as linear interpolation, spline interpolation, etc.) so that they can be fused and analyzed at the same time point. At the same time, the data is normalized and mapped to a specific interval (such as [-1,1] or [0,1]) to facilitate subsequent calculations and comparisons.

[0065] S25. Extract eigenvalues ​​of the data to be fused, and perform comprehensive quality assessment based on the eigenvalues.

[0066] The jitter-related feature values ​​are extracted from the preprocessed sensor data, including the maximum value, minimum value, mean value, variance, and rate of change of the acceleration data, and the angular velocity mean value, variance, and integral value (used to calculate the angle change) of the gyroscope data, reflecting the camera's motion state and jitter from different angles.

[0067] Fuzzy logic, weighted average and other methods are used to assign different weights according to the importance of the eigenvalues ​​to jitter detection, and a comprehensive quality evaluation index is calculated. The quality evaluation index is used to judge the reliability and validity of the sensor data, providing a basis for the subsequent Kalman filter adjustment.

[0068] S26. Dynamically adjust the Kalman filter based on the comprehensive quality evaluation results to make the best estimate of the fused data.

[0069] If the comprehensive quality evaluation results show that the sensor data quality is good, the measurement noise covariance matrix of the Kalman filter is appropriately reduced to improve the trust in the measurement data; conversely, if the data quality is poor, the measurement noise covariance matrix is ​​increased to reduce the impact of bad data on the estimation results.

[0070] The adjusted Kalman filter is used to optimally estimate the fused data to obtain a more accurate state estimate. The Kalman filter uses two steps, prediction and update, to combine the state estimate of the previous moment with the measurement value of the current moment to optimally estimate the camera's motion state (such as position, speed, angle, etc.).

[0071] S27. The weight coefficient of the data sensor is adjusted by the Kalman filter residual (the difference between the measured value and the estimated value). The weight of the sensor data with a larger residual is reduced, and the weight of the sensor data with a smaller residual is increased, thereby achieving data fusion. Finally, the fused data is obtained, which integrates the information of multiple sensors and more accurately reflects the motion state and jitter of the camera.

[0072] S3. Build a jitter detection model, input the fused data into the jitter detection model, and obtain the jitter detection result.

[0073] The jitter detection model includes an RCNN network and a Bi-LSTM network connected in sequence, such as Figure 3 As shown:

[0074] The RCNN network consists of a layer of feedforward convolutional network and a multi-layer RCL network.

[0075] The feedforward convolutional network uses multiple convolution kernels of different sizes to perform convolution operations on the input fused data (in the form of feature vectors or matrices after preprocessing and feature extraction) to extract local features. Smaller convolution kernels can capture detailed features in the data, such as the changing trend of local jitter; larger convolution kernels can obtain more macroscopic feature information, such as the frequency range of overall jitter. Each convolution kernel slides on the data and obtains a series of feature maps through convolution operations. These feature maps can highlight jitter-related features at different scales.

[0076] The multi-layer RCL network is a recurrent network composed of multiple iterative convolutional networks; the multi-layer RCL network traverses the path lengths of all convolutional networks and outputs the initial features of the fused data.

[0077] In each iterative convolutional network, the feature map output by the feedforward convolutional network is further convolved and pooled to gradually reduce the dimension of the data while extracting more abstract and representative features. The loop structure enables the network to process data in the time dimension, taking into account the correlation between the data at the previous and next moments. When processing continuous sensor fusion data, the continuity and change pattern of jitter features in the time series can be captured. By traversing the multi-layer RCL network, the initial features of the fused data are output. These initial features contain rich jitter-related information, which provides a basis for subsequent Bi-LSTM network processing.

[0078] The initial features are input into the Bi-LSTM network to obtain the time series information of the fused data, and then the self-attention mechanism is used to process the time series information and output the jitter detection results; the jitter detection results include different jitter levels and jitter directions.

[0079] The LSTM network consists of three special structures: input gate, output gate and forget gate, as well as internal state and candidate state.

[0080] The formulas and states involved in the LSTM network are as follows:

[0081] it=σ(Wi[ht-1,xt]+bi)

[0082] ft=σ (W f[ht-1,xt ]+b r)

[0083] ot=σ(Wo[ht-1,xt]+bo)

[0084] ct=ft·ct-1+it·ct

[0085] ht=ot·tanh(ct)

[0086] Where: i t , ot and f t Respectively represent the data corresponding to the input gate, output gate and forget gate; c t is the neuron state; h t is the input value of the unit at time t; W i , W o , W f and b i , b o , b f They represent the weight and bias values ​​of the input gate, output gate, and forget gate respectively.

[0087] The Bi-LSTM network is built based on LSTM, which can be referenced Figure 3 .

[0088] The initial features output by the RCNN network are input into the Bi-LSTM network. The forward and reverse LSTM units of the Bi-LSTM network process the feature sequence respectively. The forward LSTM unit starts from the beginning of the sequence and gradually learns the changing trend of the features over time to capture the positive development process of the jitter features. The reverse LSTM unit starts from the end of the sequence and reversely learns the changes of the features to obtain the impact of the jitter features in the past on the current state. Through bidirectional processing, the Bi-LSTM network can make full use of the timing information and more comprehensively understand the dynamic change process of the jitter.

[0089] After acquiring the timing information, the self-attention mechanism is used to process it. The self-attention mechanism calculates the attention weight of the features at each moment and assigns different weight values ​​according to the importance of the features. For key features closely related to jitter (such as the features corresponding to the moment when the jitter amplitude suddenly increases), a higher attention weight is given so that it plays a greater role in subsequent processing; while for relatively unimportant features, their weights are reduced. Through the processing of the self-attention mechanism, the key timing features can be highlighted, and the accuracy of the jitter detection results can be further improved. The final output includes the detection results of different jitter levels and jitter directions.

[0090] S4. Adaptively adjust the posture of the micro gimbal according to the jitter detection result, thereby realizing jitter compensation of the riding camera, including:

[0091] S41. Determine the adjustment angle of the micro gimbal according to the mapping relationship between the jitter level and the adjustment amplitude.

[0092] A mapping relationship table between the jitter level and the micro-pan adjustment amplitude is established in advance. Through a large number of experimental tests and data analysis, the micro-pan adjustment angle range corresponding to different jitter levels is determined. In this embodiment, the jitter level is divided into three levels: low (level 1-3), medium (level 4-6), and high (level 7-9). For low jitter levels, the adjustment angle range of the micro-pan is within ±0.1°; medium jitter levels correspond to an adjustment angle of ±0.1°-±0.5°; high jitter levels require an adjustment amplitude of more than ±0.5°.

[0093] According to the jitter level output by the jitter detection model, the corresponding adjustment angle range is searched in the mapping relationship table. Then, the adjustment angle is further optimized by combining the current camera shooting parameters (such as focal length, shooting mode, etc.) and the user's personalized settings (such as the user's preference for stability).

[0094] In specific application scenarios, when taking telephoto photos, the camera is more sensitive to shake. Even with a lower shake level, the micro-gimbal may need to make relatively large adjustments to ensure the clarity of the picture. When shooting videos, in order to maintain the smoothness of the picture, the angle changes may be relatively smooth to avoid sudden large adjustments.

[0095] S42. Determine the adjustment direction of the micro gimbal according to the shaking direction.

[0096] Directly determine the adjustment direction of the micro-gimbal according to the shaking direction information output by the shaking detection model. The shaking direction can be represented by a vector. Based on the camera coordinate system, the horizontal shaking (1,0,0) is represented by, the vertical shaking is represented by (0,1,0), and the shaking in the tilt direction can be represented by the corresponding three-dimensional vector.

[0097] Considering that there may be compound shaking in multiple directions during actual riding, the shaking direction is decomposed into two main components: horizontal and vertical (for tilt shaking, it is decomposed according to its angle with the horizontal and vertical directions). The micro gimbal adjusts the horizontal and vertical axes according to the decomposed direction components to achieve accurate compensation for compound shaking. If the shaking direction is detected to be 45° upward, it is decomposed into equal components in the horizontal and vertical directions. The micro gimbal makes corresponding adjustments on the horizontal and vertical axes at the same time to ensure that the camera can remain stable in multiple directions.

[0098] S43. The driving mechanism of the micro gimbal (a micro gimbal voice coil motor in this embodiment) adjusts the posture of the micro gimbal according to the adjustment angle and adjustment direction.

[0099] In terms of determining the adjustment angle according to the jitter level, when a low jitter level is detected, the micro gimbal voice coil motor drives the micro gimbal to make precise small angle adjustments based on the pre-set adjustment angle range within ±0.1°, with the signal transmission and circuit support of the double L flexible circuit board. Since the adjustment range is small at low jitter levels, the tilt spring can use its own elastic characteristics to assist the micro gimbal in smooth rotation, ensuring accurate compensation without affecting the shooting stability, and avoiding the adverse effects of excessive adjustment on the picture.

[0100] For medium jitter levels (4-6), corresponding to the adjustment angle of ±0.1°-±0.5°, the micro-pan head voice coil motor increases the driving force after receiving the control signal. At this time, the double L flexible circuit board ensures the stable transmission of the motor control signal. During the adjustment process of a larger angle, the tilt spring effectively buffers the impact force caused by the rapid adjustment through its tilt structure and elastic deformation, so that the micro-pan head can quickly and stably rotate to the specified angle, keeping the camera shooting picture stable.

[0101] When high jitter levels (levels 7-9) require an adjustment range of more than ±0.5°, the micro gimbal voice coil motor operates at full capacity to achieve large-angle compensation (±5°). The double-L flexible circuit board ensures high-intensity power supply and signal transmission, and the tilted spring fully exerts its elastic energy storage and release functions, cooperating with the voice coil motor to cope with large-scale posture adjustments, ensuring that the micro gimbal can still effectively compensate under extreme jitter conditions and maintain the recognizability of the captured image.

[0102] In terms of determining the adjustment direction according to the shaking direction, this embodiment takes the case where the shaking direction is detected to be 45° upward as an example. The micro gimbal first uses its internal sensor data processing mechanism, combined with the signal transmission and processing capabilities of the double L flexible circuit board, to decompose the shaking vector in this oblique direction into equal components in the horizontal and vertical directions. The voice coil motor of the micro gimbal is driven on the horizontal and vertical axes respectively according to the decomposed direction components, and the inclined spring provides elastic support and buffering in two directions. In the horizontal direction, the voice coil motor pushes the micro gimbal to move along the horizontal axis, and the spring resists the inertial force and impact force in the horizontal direction; the same is true in the vertical direction. Through this synergistic effect, the micro gimbal can accurately adjust the posture in a complex compound shaking environment to ensure that the camera remains stable in all directions, thereby effectively compensating for various shakes during riding and improving the quality of the captured image.

[0103] S5. Use image intelligent algorithm to identify riding scenes and automatically adjust camera shooting parameters according to the riding scene recognition results.

[0104] In this embodiment, a convolutional neural network (CNN) algorithm is used to train a scene recognition model. A large amount of image data containing different cycling scenes (such as city streets, mountain roads, country roads, park cycling paths, etc.) is collected as a training set to train the CNN model.

[0105] The CNN model can identify the current riding scene by extracting and classifying the features of the image. During the feature extraction process, the network's convolutional layer automatically learns the texture, color, shape and other features in the image, the pooling layer reduces the data dimension, and the fully connected layer makes classification decisions. By continuously adjusting the model's parameters, the accuracy of scene recognition can be improved.

[0106] Urban street scenes have obvious features such as buildings and traffic signs; mountain road scenes contain natural landscape features such as mountains and forests; park cycling paths have features such as lawns and flower beds. The scene recognition model classifies images based on these feature differences and outputs recognition results.

[0107] According to the cycling scene recognition results, the camera's shooting parameters are automatically adjusted. Take the following scenes as an example:

[0108] In urban street riding scenes, due to the complex environment and large changes in light, the camera automatically adjusts parameters such as sensitivity (ISO), shutter speed and white balance to adapt to rapidly changing light conditions and ensure the clarity and color accuracy of the captured image. Specifically, the ISO value is increased to increase the camera's sensitivity to light, and the shutter speed is adjusted according to the light intensity and movement speed to avoid blurry images due to fast movement of the subject or insufficient light.

[0109] In the mountain road riding scene, due to the open scenery and sufficient light, the camera automatically adjusts the focus and depth of field to create a better visual effect. The focal length is appropriately increased to make the distant mountains clearer in the picture, and the depth of field is adjusted to make the foreground and background have a certain blur effect, enhancing the layering of the picture.

[0110] In the park cycling scene, in order to capture more natural landscape details, the camera will lower the ISO value to reduce noise, and adjust the color saturation to make the picture more vivid. In addition, according to the user's possible shooting needs in different scenes, the shooting mode (such as landscape mode, sports mode, etc.) can be automatically adjusted to provide a shooting effect that is more in line with the characteristics of the scene and enhance the user's shooting experience.

[0111] On the other hand, the present invention also proposes a 5G smart cycling camera anti-shake system based on a micro-pan platform, which is used to implement the above-mentioned 5G smart cycling camera anti-shake method based on a micro-pan platform. The system architecture is as follows: Figure 4As shown, it includes: a micro gimbal, a cycling camera and a control module. The micro gimbal is installed between the lens and the image sensor in the cycling camera, and a data sensor is integrated; the control module is connected to the micro gimbal and the cycling camera for communication;

[0112] The control module includes:

[0113] A data processing unit is used to calibrate the data sensor, obtain sensor data using the calibrated data sensor, and perform data fusion to obtain fused data;

[0114] A jitter detection unit, used for constructing a jitter detection model, inputting the fused data into the jitter detection model, and obtaining a jitter detection result;

[0115] The shake compensation unit is used to adaptively adjust the posture of the micro gimbal according to the shake detection result, thereby realizing the shake compensation of the riding camera.

[0116] Preferably, the control module further includes:

[0117] The scene recognition unit is used to use image intelligent algorithms to recognize riding scenes and automatically adjust camera shooting parameters according to the riding scene recognition results.

[0118] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0119] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A 5G smart cycling camera anti-shake method based on a micro-pan head, characterized in that: The following steps are involved: Install the micro gimbal between the lens and image sensor in the cycling camera and integrate the data sensor; Calibrate the data sensor, use the calibrated data sensor to acquire sensor data and perform data fusion to obtain fused data; Constructing a jitter detection model, inputting the fused data into the jitter detection model, and obtaining a jitter detection result; The posture of the micro gimbal is adaptively adjusted according to the jitter detection result, thereby achieving jitter compensation for the riding camera.

2. According to the micro-pan-tilt-based 5G smart riding camera anti-shake method of claim 1, it is characterized in that: Calibrating the data sensor includes: Collecting the measured values ​​of the data sensor and the corresponding true values; Calculate error characteristic data of the data sensor based on the measured value and the true value; Based on the error signature data, the data sensor is calibrated using a machine learning algorithm.

3. According to the 5G smart cycling camera anti-shake method based on micro-pan head according to claim 2, it is characterized in that: The calibrated data sensor is used to obtain sensor data and perform data fusion to obtain fused data, including: Preprocessing the sensor data to obtain data to be fused; Extracting characteristic values ​​of the data to be fused, and performing comprehensive quality assessment based on the characteristic values; Dynamically adjust the Kalman filter according to the comprehensive quality evaluation result to optimally estimate the data to be fused; The weight coefficient of the data sensor is adjusted through the Kalman filter residual to perform data fusion and obtain fused data.

4. According to the micro-pan-tilt-based 5G smart riding camera anti-shake method of claim 1, it is characterized in that: The jitter detection model includes an RCNN network and a Bi-LSTM network connected in sequence; The RCNN network is composed of a layer of feedforward convolutional network and a multi-layer RCL network; the multi-layer RCL network is a cyclic network composed of multiple iterative convolutional networks; the multi-layer RCL network traverses the path lengths of all convolutional networks and outputs the initial features of the fused data; Input the initial features into the Bi-LSTM network, obtain the time series information of the fused data, process the time series information using the self-attention mechanism, and output the jitter detection result; The jitter detection result includes different jitter levels and jitter directions.

5. According to the micro-pan-tilt-based 5G smart riding camera anti-shake method of claim 1, it is characterized in that: Adaptively adjusting the posture of the micro gimbal according to the jitter detection result, thereby realizing jitter compensation of the riding camera, including: Determining the adjustment angle of the micro gimbal according to the mapping relationship between the jitter level and the adjustment amplitude; Determining the adjustment direction of the micro gimbal according to the shaking direction; The driving mechanism of the micro gimbal adjusts the posture of the micro gimbal according to the adjustment angle and the adjustment direction.

6. According to the micro-pan-tilt-based 5G smart riding camera anti-shake method of claim 1, it is characterized in that: Also includes: Use image intelligent algorithms to identify cycling scenes and automatically adjust camera shooting parameters based on the recognition results.

7. A 5G smart cycling camera anti-shake system based on a micro-pan head, characterized in that: include: Micro gimbal, cycling camera and control module, the micro gimbal is installed between the lens and image sensor in the cycling camera, and the data sensor is integrated; The control module is connected to the micro gimbal and the riding camera for communication; The control module comprises: A data processing unit, used to calibrate the data sensor, obtain sensor data using the calibrated data sensor, and perform data fusion to obtain fused data; A jitter detection unit, used for constructing a jitter detection model, inputting the fused data into the jitter detection model, and obtaining a jitter detection result; The shake compensation unit is used to adaptively adjust the posture of the micro gimbal according to the shake detection result, thereby realizing shake compensation of the riding camera.

8. The 5G smart cycling camera anti-shake system based on micro-pan head according to claim 7 is characterized in that: The control module also includes: The scene recognition unit is used to use image intelligent algorithms to recognize riding scenes and automatically adjust camera shooting parameters according to the riding scene recognition results.

Images