Carrier positioning and control method and system based on machine vision

Abstract

The invention relates to the technical field of vehicle control, and discloses a vehicle positioning and control method and system based on machine vision, and the method comprises the steps: building a vehicle kinematics prediction model, and generating a reference pose sequence according to a driving instruction through the vehicle kinematics prediction model; recording a first reference pose during image acquisition; calculating a deviation vector between a positioning result and the first reference pose when positioning calculation is completed; determining a dynamic damping coefficient according to a direction included angle between the velocity vector and the deviation vector, generating a compensation gain sequence, and converting the compensation gain sequence into a compensation torque of a current loop given value; according to the method, the effective decoupling of the sensing time delay of the sensor and the dynamic response of the carrier is realized, the phase lag is reduced on the premise of ensuring the momentum continuity of the controlled object, the instantaneous control oscillation of positioning correction is eliminated, and the mechanical stress impact of an execution mechanism is inhibited.

Description

Technical Field

[0001] This invention relates to a vehicle positioning and control method and system based on machine vision, belonging to the field of vehicle control technology. Background Technology

[0002] Current autonomous vehicle navigation systems utilize machine vision sensors to extract environmental features, calculate vehicle pose, and provide absolute position references. Combined with odometers and kinematic models, they achieve attitude estimation and path tracking. However, in industrial warehousing conditions, ground friction fluctuations, wheel hub wear, and heavy-load starts cause nonlinear slippage in kinematic calculations, leading to error accumulation. While vision systems provide global pose correction, image acquisition and processing have processing delays, causing the control system to lag behind the real-time state in acquiring pose information.

[0003] Optimizing the shape of the vehicle rollers or improving the hardware of the actuator to enhance physical response efficiency is difficult to achieve high-precision closed-loop stable operation if the control algorithm logic cannot absorb the phase lag of the perceived data in the time domain. The Chinese invention patent with authorization announcement number CN118691667B discloses a machine vision positioning method that uses color scene-based vector analysis to process the transition between each pixel and the surrounding pixels in the vehicle image to achieve defect positioning. This method focuses on extracting static image features and processing pixel connection consistency, which belongs to static detection of discrete spatial features and does not consider the dynamic constraints of the continuous motion process of the controlled object. It uses kinematic model extrapolation to compensate for visual time delay, and defaults to quasi-instantaneous feedback absorption, ignoring the matching between the discrete pulse characteristics of the visual signal and the continuous energy process of the dynamic response. When the visual calculation time fluctuates, the compensation gain and the phase mismatch of the physical motion cause the controller output to jump, resulting in control oscillations when the vehicle is turning or braking.

[0004] Therefore, how to couple visual feedback with dynamic processes to reduce tracking smoothness, and how to achieve coupling between visual feedback and vehicle dynamic processes under conditions of limited computing resources and perceived disturbances, so as to ensure phase alignment and momentum smoothness of the control loop, has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A vehicle positioning and control method based on machine vision, comprising the following steps: Step 101: Establish a vehicle kinematics prediction model, and use the vehicle kinematics prediction model to generate a virtual reference pose of the vehicle based on the real-time driving commands of the vehicle drive unit, so as to construct a reference pose sequence synchronized with the vehicle motion. Step 102, at the time of image acquisition by the vision sensor. Record the corresponding first virtual reference pose; Step 103, at the current moment when the visual positioning result corresponding to the image acquisition is solved. The differential deviation vector between the visual positioning result and the first virtual reference pose is calculated. The differential deviation vector is used to characterize the vehicle's position at the acquisition time. Up to the current moment The resulting nonlinear motion deviation; Step 104, Obtain the vehicle's current position. The real-time velocity vector is obtained, the directional angle between the real-time velocity vector and the differential deviation vector is calculated, and the dynamic damping coefficient is determined according to the directional angle based on the preset mapping rules. Step 105: The differential deviation vector is scaled using the dynamic damping coefficient to generate a compensation gain sequence that decays exponentially with time, and the compensation gain sequence is converted into a virtual compensation torque acting on the current loop setpoint of the vehicle drive unit. Step 106: Adjust the output torque of the vehicle drive unit according to the compensation gain sequence to drive the physical pose of the vehicle to converge towards the target pose corrected by the differential deviation vector, so as to reduce the impact of the acquisition time while maintaining the continuity of vehicle dynamics. Up to the current moment The phase lag that occurs during this period.

[0006] Preferably, the process of scaling the differential deviation vector using the dynamic damping coefficient includes: obtaining the preset motion envelope constraint value of the vehicle; calculating the initial amplitude of the compensation gain sequence and the time constant used to limit the decay rate based on the dynamic damping coefficient and the preset motion envelope constraint value; and constructing a compensation gain sequence that decays exponentially with time using the initial amplitude and the time constant, so that the electromagnetic torque of the vehicle drive unit is smoothly adjusted within a preset time window.

[0007] Preferably, the differential deviation vector is used to describe the nonlinear displacement deviation of the vehicle during the image processing delay period caused by fluctuations in the friction coefficient of the road surface or changes in the vehicle load.

[0008] Preferred time of data collection The phase alignment of observed and predicted data in the time dimension is achieved by synchronizing the trigger pulse of the visual sensor with the system clock of the vehicle's main controller.

[0009] Preferably, before calculating the differential deviation vector, the method further includes: step 501, predicting the predicted motion features of image feature points in the pixel coordinate system based on the output of the vehicle kinematics prediction model; step 502, extracting the actual motion features from the visual positioning results and calculating the consistency deviation value between the actual motion features and the predicted motion features; step 503, determining the confidence level of the current visual positioning based on the consistency deviation value, and adjusting the weight coefficient of the compensation gain sequence in the driving adjustment process according to the confidence level.

[0010] Preferably, after calculating the differential deviation vector in step 103, the method further includes: step 601, extracting the radius of curvature information of the preset trajectory in front of the vehicle in real time; step 602, determining the centripetal motion deviation component of the vehicle within the image processing delay period based on the radius of curvature information; step 603, using the centripetal motion deviation component to perform second-order interpolation compensation on the differential deviation vector to correct the dynamic tracking deviation caused by the vehicle's turning.

[0011] Preferably, in the process of establishing the vehicle kinematic prediction model in step 101, the method further includes dynamically calibrating the vehicle kinematic prediction model, the steps of which include: step 701, applying a micro-pulse excitation torque of a preset frequency to the vehicle drive unit; step 702, acquiring the instantaneous speed feedback of the vehicle under the action of the micro-pulse excitation torque, and extracting the response features corresponding to the preset frequency; step 703, using the response features to calculate the slip parameters between the vehicle and the road surface. The calculation formula is: ,in, For the slip parameter, In response to the characteristic amplitude, The amplitude of the micropulse excitation torque. The instantaneous speed feedback is the phase lag angle relative to the micro-pulse excitation torque; step 704, based on the slip parameter Correct the driving gain coefficient of the vehicle kinematics prediction model.

[0012] Preferably, before calculating the differential deviation vector in step 103, the method further includes: step 801, calculating the real-time processing delay of the data acquired by the visual sensor from image acquisition to solution completion; step 802, reducing the weight coefficient when the real-time processing delay exceeds a preset delay reference threshold; and step 803, maintaining the motion control output of the vehicle using the reference pose sequence generated by the vehicle kinematic prediction model.

[0013] Preferably, in step 105, when injecting the compensation gain sequence into the current loop setpoint of the vehicle drive unit, the method further includes using a bandpass filtering algorithm to extract the feedback current characteristics of the drive motor of the vehicle drive unit, and using the feedback current characteristics to perform phase correction on the virtual compensation torque in order to suppress the mechanical impact of the actuator. The adjustment process of the output torque of the vehicle drive unit follows the second-order continuity constraint of the vehicle motion state to eliminate the control signal jump at the moment when the visual positioning result takes effect.

[0014] A machine vision-based vehicle positioning and control system includes: The kinematic prediction module is used to generate a virtual reference pose of the vehicle by using the vehicle kinematic prediction model and combining it with the real-time drive commands of the vehicle drive unit, so as to construct a reference pose sequence synchronized with the vehicle motion. The acquisition and recording module is used to record the acquisition time when the vision sensor performs image acquisition. Record the first virtual reference pose corresponding to the virtual reference pose. The differential calculation module is used to calculate the visual positioning results at the current moment. Calculate the difference deviation vector between the visual localization result and the first virtual reference pose; The dynamics analysis module is used to obtain the vehicle's current position. The real-time velocity vector is obtained, and the dynamic damping coefficient is determined based on the directional angle between the real-time velocity vector and the differential deviation vector. The compensation mapping module is used to scale the differential deviation vector using the dynamic damping coefficient, generate a compensation gain sequence that decays exponentially with time, and map the compensation gain sequence into a virtual compensation torque acting on the current loop setpoint of the vehicle drive unit. The drive control module adjusts the output torque of the vehicle drive unit according to the compensation gain sequence, driving the physical pose of the vehicle to converge towards the target pose corrected by the differential deviation vector, thereby reducing the impact of the acquisition time while maintaining the continuity of vehicle dynamics. Up to the current moment The phase lag that occurs during this period.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the vehicle positioning and control method, the kinematic prediction model generates a virtual reference pose, which forms a time phase correlation with the acquisition time of the visual sensor. When the visual positioning result is completed, the cross-time domain pose deviation is extracted to eliminate the influence of the sensor discrete sampling feedback delay on the closed-loop control loop. When the control system obtains the delayed pose feedback, it removes the nonlinear motion deviation caused by external physical environment disturbances, and avoids vehicle control oscillation caused by feedback adjustment phase mismatch.

[0016] 2. By utilizing the instantaneous speed feedback flow output by the motor encoder of the drive unit, the response characteristics of the micro-pulse excitation torque are analyzed, and the driving coefficients of the kinematic prediction model are calibrated in real time. The pose generation logic has the ability to adapt to fluctuations in ground friction coefficient and tire wear variables. The perception mechanism, combined with the differential deviation vector compensation logic, ensures that the positioning results used for control decision fusion under complex working conditions are anchored to the real physical motion reference, eliminating the cumulative control error caused by model parameter drift.

[0017] 3. The confidence level is determined based on the consistency deviation between the predicted motion features and the actual feature point extraction results. The compensation weight is adjusted to dynamically suppress visual jump noise. The feature processing mechanism is combined with the second-order interpolation compensation of trajectory curvature. When the vehicle enters the nonlinear motion range or encounters instantaneous visual interference, the inertial output of the virtual shadow model is used to maintain the smooth transition of control commands, ensure the momentum continuity of the drive actuator, reduce the instantaneous mechanical stress impact of the drive motor, and suppress the vibration noise of the transmission chain. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the vehicle control method integrating time delay compensation and dynamic regulation according to the present invention; Figure 2 This is a comparative experimental curve showing the convergence characteristics of path deviation under different sliding conditions of the present invention; Figure 3 This is an exploded view of the vehicle positioning and control system architecture of the multi-dimensional collaborative mechanism of the present invention. Detailed Implementation

[0019] The following disclosure is for illustrative purposes only and is not intended to limit the scope of protection of the present invention; unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0020] This invention provides a vehicle positioning and control method and system based on machine vision, comprising a kinematic prediction layer, a visual perception layer, a dynamics analysis layer, and a drive compensation layer. The kinematic prediction layer establishes a virtual reference pose sequence synchronized with the vehicle's physical motion, and the visual perception layer utilizes image acquisition data... and the time of result return The time-domain span between the two is used to extract the differential deviation vector; the dynamic analysis layer determines the dynamic damping characteristics based on the geometric relationship between the vehicle's real-time velocity direction and the deviation direction; the drive compensation layer converts the position deviation into a virtual compensation torque that decays exponentially with time and injects it into the current loop setpoint of the vehicle drive unit, achieving smooth decoupling between the sensor's discrete sensing delay and the vehicle's dynamic response process; when the vehicle operates under conditions such as industrial warehousing, nonlinear slippage caused by fluctuations in ground friction, load changes, or wheel hub wear will lead to attitude drift accumulated over time in dead reckoning; to solve this problem, this invention constructs a vehicle kinematic prediction model; it receives real-time drive commands issued by the vehicle's main controller, uses discretized motion equations to calculate the vehicle's virtual reference attitude, thereby constructing a continuous attitude sequence that is phase-synchronized with the vehicle's physical state; in specific execution, the system uses The feedback angular velocity of the drive motor is read during the sampling period. With steering angle Input to include wheel radius With wheelbase In the differential-driven model, the coordinate increment at the current moment is calculated and accumulated into the virtual reference pose at the previous moment; for example, when the vehicle... When traveling at a straight speed, the prediction model produces a displacement increment in each sampling period. The coordinates of these points form a continuous virtual trajectory, which serves as a reference for subsequent visual feedback correction, enabling the control loop to have state estimation capabilities during the perception data gap period.

[0021] When the vision sensor performs environmental feature extraction, there is a processing delay from the shutter triggering image acquisition to the positioning algorithm completing the calculation and outputting the result. This causes the pose information acquired by the control system to lag behind the real-time state of the vehicle. The system adopts the following phase locking and deviation extraction procedure: the system clock of the vehicle's main controller is triggered by the trigger pulse signal of the vision sensor to record the image acquisition moment of the vision sensor. Record the first virtual reference pose at that moment, generated by the kinematic prediction model and used as a spatiotemporal anchor point. The visual positioning result is expected at the current moment. When the calculation is completed and returned to the main controller, the corresponding visual positioning result is obtained. System calculation and The difference deviation vector between The difference deviation vector peeling by to The motion increment generated by the vehicle during this period represents the nonlinear displacement deviation caused by ground friction fluctuations or load changes, reducing the phase lag caused by sensor sensing delay in closed-loop feedback adjustment. If the deviation value at the coordinate level is directly compensated to the vehicle control quantity, discrete signal jumps will break the momentum continuity of the controlled object, inducing mechanical stress impact in the actuator and causing control oscillation. To solve this dynamic matching problem, this invention converts coordinate deviation elimination into a smooth torque adjustment process of the controlled object. The processor obtains the vehicle's current position... Real-time velocity vector And calculate the real-time velocity vector. With difference deviation vector The angle between the directions According to the preset mapping rules based on the direction angle Determine the dynamic damping coefficient Among them, the mapping rule follows cosine correlation, when the direction angle Approaching When the direction of deviation is consistent with the direction of motion, the dynamic damping coefficient is relatively small. To accelerate correction; when the direction angle Approaching When this occurs, it indicates a lateral deviation; in this case, the dynamic damping coefficient should be increased. To suppress yaw oscillations; dynamic damping coefficient Mapping rules are established based on the direction angle. For a continuous mapping function of the independent variable, set a lower damping limit. for And the upper limit of damping for When the direction angle In to In the interval, the dynamic damping coefficient Pick Maintain the sensitivity of the correction response; when the direction angle Exceed At that time, according to the formula Calculate the dynamic damping coefficient Used to adjust the compensation gain sequence The exponential decay rate increases the damping ratio to suppress the lateral oscillation of the actuator as the lateral deviation increases.

[0022] Determining the dynamic damping coefficient Then, the system utilizes the dynamic damping coefficient For the difference deviation vector Modulus scaling is performed to generate a compensated gain sequence that decays exponentially over time. In the specific calibration, the maximum compensation amplitude is limited by a preset motion envelope constraint value, and the calculation formula is as follows: ,in, For compensating for the gain sequence; For difference bias vector The initial amplitude is determined by the modulus; This refers to the dynamic damping coefficient; For time indexing; The system uses this compensation gain sequence as a time constant to limit the decay rate. This is converted into virtual compensation torque acting on the current loop setpoint of the vehicle drive unit. By adjusting the electromagnetic torque of the drive motor, the physical pose of the vehicle is driven to converge towards the target pose within a preset time window; when the lateral deviation is... And dynamic damping coefficient for At that time, the generated compensating torque is The torque gradually decreases smoothly from its peak value to zero, ensuring that the output torque of the drive unit follows the second-order continuity constraint of the vehicle's motion state, eliminating the control signal jump at the moment the visual feedback takes effect; virtual compensation torque. The current loop of the injected vehicle drive unit utilizes the cutoff frequency of the drive motor speed loop. Determine the time constant , time constant Set as reciprocal to The feedback current is sampled and extracted using a bandpass filter, and the result is compared with the micropulse excitation torque. Harmonic components of the same frequency current; adjust the initial amplitude according to the envelope variation characteristics of the harmonic components. The scaling factor ensures that the compensation vector acting on the current loop's given value follows a second-order continuity constraint in the time domain, eliminating the torque jump in the actuator caused by the positional step deviation at the moment the visual positioning result takes effect.

[0023] To determine the initial amplitude The mapping weights and the system establish the proportional gain coefficient. This coefficient is determined by the rated torque of the vehicle's actuator. With the preset maximum correction step The ratio is determined in this embodiment, where the initial amplitude is... Follow computational logic ;in, This is the initial amplitude, in units of , This is the proportional gain coefficient, in units of... , The magnitude of the difference deviation vector, in units of . The processor uses this proportional correlation to establish a quantization conversion benchmark for position deviation within the torque space of the driving actuator; changes in light and shadow or feature occlusion in industrial environments can cause random fluctuations in visual positioning results; to address this challenge, the system introduces a gain adaptive adjustment mechanism based on motion consistency determination; in calculating the differential deviation vector First, based on the output of the kinematic prediction model, the predicted motion features of image feature points in the pixel coordinate system are predicted; simultaneously, the actual motion features in the visual localization results are extracted, and the consistency deviation value between the actual motion features and the predicted motion features is calculated. Based on consistency deviation value Determine the confidence level of the current visual localization When using a consistency determination mechanism, the predicted motion features in the pixel coordinate system of the predicted image feature points output by the kinematic prediction model are used to extract the actual motion features from the visual localization results and calculate the consistency deviation value between the two. If the deviation value If the tolerance threshold is exceeded, the compensation weight coefficient is adjusted downwards; at this time, the reference pose sequence is used to maintain the vehicle's inertial output, ensuring the momentum continuity of the drive actuator. When the consistency deviation value... Exceeding the preset tolerance threshold At pixel level, the system lowers the compensation weight coefficient and uses the reference pose sequence generated by the kinematic prediction model to maintain the inertial output of the vehicle, thus maintaining the momentum continuity of the drive actuator when encountering instantaneous visual disturbances.

[0024] The accuracy of vehicle kinematics prediction models is affected by fluctuations in the friction coefficient of the road surface and tire wear; this invention employs an active excitation response mechanism to calibrate the model; the processor periodically applies a preset frequency to the vehicle drive unit. Micropulse excitation torque Simultaneously, the instantaneous speed feedback output by the encoder of the vehicle drive motor is acquired; the response characteristic amplitude corresponding to the preset frequency is extracted using a bandpass filtering algorithm. With phase lag angle The calculation formula is as follows: ,in, These are the slip parameters; For response characteristic amplitude; The amplitude of the micropulse excitation torque; The phase lag angle of instantaneous speed feedback relative to the micro-pulse excitation torque; slip parameter The modified vehicle kinematics prediction model will adjust the slip parameters. The pose estimation logic is introduced as a driving gain verification factor, and the sampling period is... Obtain the feedback angular velocity of the drive motor Using slip parameters Correct preset wheel radius To obtain the effective driving radius The virtual reference pose sequence is updated in conjunction with the state transition equation; the response characteristic amplitude generated by active excitation is used to update the virtual reference pose sequence. With phase lag angle The scaling factor is converted into a kinematic model scale to correct for pose extrapolation bias caused by road friction fluctuations or tire wear. This allows the kinematic prediction model to anchor the reference pose sequence during the perception window to a physical motion benchmark. The system is based on slip parameters. The driving gain coefficients of the kinematic prediction model are corrected to anchor the virtual reference pose to the physical motion reference. In engineering deployments, the processing time of visual positioning systems often fluctuates due to algorithm search complexity or fluctuations in computing power allocation. To address reliability issues under computationally limited conditions, the system monitors the real-time processing latency of visual data from image acquisition to solution completion. When real-time processing latency Exceeding the preset latency threshold At the same time, the system reduces the weight coefficient according to the degree of time delay deviation and simultaneously extends the decay period of the virtual compensation torque, so that the control loop adaptively absorbs the impact of load jitter. This time-based control parameter allocation method improves the operational stability and trajectory tracking accuracy of the vehicle in the industrial controller environment.

[0025] Example 1: In the case of a vehicle performing a precision handling task in a high-density automated warehouse aisle, the local coefficient of friction on the ground is affected by residual oil stains from the operation. to Fluctuations occur within the range, and link jitter in the industrial wireless network increases the processing delay of the vision sensor from image acquisition to the return of the positioning calculation result, which is affected by standard... Dynamically increase to This leads to a situation where, if the control system directly uses the discrete positioning result with a large time delay as the position setpoint for closed-loop feedback, the acquired pose information belongs to the vehicle's historical state. This can cause the actuator to receive erroneous step correction commands, resulting in mechanical stress impact on the vehicle during correction and inducing lateral path oscillations in the controlled object. When the system faces the above-mentioned superposition of nonlinear slip and variable time delay, the main controller uses a kinematic prediction model based on... The driving command stream continuously generates a virtual reference pose sequence; at the acquisition moment triggered by the vision system shutter. The main controller will generate the coordinates at this moment. First virtual reference pose Phase anchoring storage is performed; the vision processing unit undergoes... After the calculation delay, at the current time The solution results are returned, which include the vehicle's position at the time of data acquisition. Visual positioning results Its value is The system then calculates the visual positioning result. With the first virtual reference pose The difference deviation vector between get The deviation value; the difference deviation vector Accurately peel off the vehicle in The motion displacement generated during the delay period due to autonomous operation transforms the positioning error extraction from relying on absolute coordinate mapping to deviation extraction based on virtual and real phase alignment.

[0026] The system obtains the vehicle's current position. Real-time velocity vector for Calculate the real-time velocity vector With difference deviation vector The angle between the directions The dynamic damping coefficient determined according to the mapping rule Set as Using this dynamic damping coefficient For the difference deviation vector Perform dynamic elimination; generate a compensated gain sequence Follow the calculation formula below: ,in, For compensating for the gain sequence; For difference bias vector The initial amplitude is determined by the modulus; This refers to the dynamic damping coefficient; For time indexing; The system will use the preset time constant to calculate the compensation gain sequence. This is converted into virtual compensation torque acting on the current loop setpoint of the drive unit. The vehicle's physical pose converges to the target pose within a preset time window; when encountering combined disturbances of ground slippage and visual perception lag, the rate of curvature change of its trajectory remains within the stress threshold of the actuator, eliminating momentum jumps or mechanical vibrations caused by abrupt changes in position coordinates; a zero-delay benchmark is provided for visual feedback using a sequence constructed using a kinematic prediction model, and a damped torque distribution mechanism is combined to smoothly eliminate discrete perception deviations in the continuous dynamic process; the vehicle's pose deviation stabilizes after path correction. Within a certain range, and no high-frequency oscillation was observed in the output current of the drive unit.

[0027] Example 2: The test group operated on a differential drive wheeled vehicle platform, including a resolution of The photoelectric encoder with a pulse per revolution has a measurement accuracy of no less than The sampling frequency of the vision sensor and main controller is set to... The test environment was established at a length of A linear track, by controlling the thickness of the oil film on the ground to keep the adhesion coefficient within a certain range. to The range is designed to simulate slip conditions, and at the time of data acquisition... Introducing amplitude Gaussian white noise simulation of industrial electromagnetic interference; time constant The set value is based on the response bandwidth of the drive motor. When the response bandwidth When in the high-frequency range, in order to avoid the resonant frequency of the mechanical structure, the time constant... With response bandwidth The relationship is inversely proportional; under the experimental conditions, the time constant will be... Set as After the experiment was started, the main controller calculated the encoder feedback data and used the kinematic prediction model to produce a reference pose sequence. Table 1 shows the system's extracted differential deviation vector and its control effect under different slip intensity gradients. The control group used the traditional proportional-integral-derivative control method, while the experimental group used the method disclosed in this invention.

[0028] Table 1: Comparison of experimental results under different slip gradients See Table 1 for the difference deviation vector caused by slip. Length of the module is determined by Increase to The dynamic damping coefficient in the test group The mapping rule is adaptively adjusted upwards to maintain the maximum path amplitude at [value missing]. Within this range, the control group exhibited transverse oscillations under the same slip intensity, with a path amplitude reaching [value missing]. and The difference in this data confirms the difference bias vector. , with dynamic damping coefficient The effect of coupling in suppressing momentum jumps, and the compensated gain sequence generated by the system. Follow the formula below: ,in, For compensating for the gain sequence; For difference bias vector The initial amplitude determined by the modulus length, in units of ; This refers to the dynamic damping coefficient; This is a time index, in units of... ; The preset time constant is in units of The experimental data show a nonlinear effect and a performance inflection point; under the condition of exceeding the upper limit, when the differential deviation vector... achieve And dynamic damping coefficient Approaching the physical upper limit At that time, a sudden increase in the maximum path amplitude indicated that the system's absorption capacity reached saturation after the displacement deviation exceeded the linear adjustment range of the electromagnetic torque of the driving unit. Experimental results show that the reference pose sequence generated by the kinematic prediction model provides a phase reference for visual feedback. to Within the deviation range, the system converts the coordinate deviation into an exponentially decaying virtual compensation torque, reducing the phase lag caused by processing delay. After completing path correction, the vehicle's pose deviation stabilizes. Within a certain range, and with no high-frequency oscillation observed in the output current of the drive unit, the controlled object maintains operational stability even under non-ideal environmental conditions.

[0029] Example 3: This example combines Figures 1 to 3 This describes a machine vision-based vehicle positioning and control method and system, such as... Figure 1 As shown, the process receives real-time drive commands and proceeds to step 101, establishes a kinematic prediction model, and generates a virtual reference pose sequence based on the real-time drive commands. Step 102 records the first virtual reference pose and locks the predicted pose state at visual acquisition time T1. Simultaneously, the process receives visual positioning results and calculates the differential deviation vector in step 103, i.e., calculates the deviation between the visual result and the first virtual reference pose at time T2. Combined with the real-time velocity vector, the process proceeds to step 104, determines the dynamic damping coefficient, and maps the direction angle between the velocity vector and the deviation vector. The process then proceeds to step 105, generates a compensation gain sequence, scales the module length, and generates an exponentially decaying virtual compensation torque. Finally, in step 106, the output torque of the drive unit is adjusted to reduce phase lag and maintain the continuity of vehicle dynamics, thereby achieving physical pose convergence control.

[0030] like Figure 2 As shown, the horizontal axis represents time in seconds (s), and the vertical axis represents path deviation in millimeters (mm). The figure displays four curves: low slip in the experimental group, medium slip in the experimental group, high slip in the experimental group, and medium slip in the control group. The medium slip curve in the control group shows an upward trend and reaches a relatively high peak during the experiment. In contrast, the three curves in the experimental group, at low, medium, and high slip levels, all show significantly lower path deviation amplitudes than the control group, and after reaching a smaller peak, they exhibit a gentle convergence trend, indicating the effectiveness of different slip conditions in controlling deviation. Figure 3 As shown, the architecture mainly consists of four branches: the temporal phase anchoring branch includes recording the acquisition time T1, the solution completion time T2, and real-time processing delay monitoring; the core prediction algorithm branch covers kinematic prediction models, differential deviation vector calculation, and virtual reference pose generation; the dynamic control branch is responsible for exponential decay sequences, virtual compensation torque injection, and dynamic damping coefficient mapping; and the environmental disturbance resistance mechanism branch includes visual jump noise suppression, motion consistency confidence, and road slip parameter correction.

[0031] Example 4: In the case of a carrier performing wafer cassette transfer tasks in a precision semiconductor manufacturing workshop, due to airflow disturbances generated by the workshop's ventilation system and the micron-level sensitivity required for positioning accuracy, if the carrier generates more than [a certain amount of airflow disturbance] during path correction... Torque fluctuations cause vibrations at the end of the actuator, resulting in pose overshoot of the robotic arm at the docking station and triggering a safety shutdown.

[0032] The system adopts the following kinematic update and parameter calibration methods, with the main controller using... Frequency execution state estimation, in each sampling period The system receives real-time angular velocity feedback from the drive motor, updates the virtual reference pose using the state transition equation, and the calculation logic follows the formula below: ,in, for Virtual reference pose vector at any given moment; for Virtual reference pose vector at any given moment; The displacement increment is calculated based on encoder feedback, in units of... ; For the vehicle's heading angle; the system establishes the dynamic damping coefficient. The analytic function is specifically expressed as: ,in, This refers to the dynamic damping coefficient; To preset the lower limit of damping, it is set to [value] under this operating condition. ; The upper limit of damping is set to [value]. ; The angle between the vehicle's real-time velocity vector and the differential deviation vector; used to establish the consistency deviation value. The benchmark is determined by data collection by the processor in offline mode. The residuals of motion characteristics under the group of driving trajectories were calculated using statistical distribution to obtain the standard deviation of the residuals. Pixels, with tolerance thresholds set according to the three-standard-deviation criterion. Set as Pixel, when the consistency deviation value is measured in real time Greater than When the pixel count is low, the system determines that visual positioning has failed and performs weight reduction; the torque output derivative of the vehicle during the tracking process remains within the preset safety range, the current loop feedback waveform exhibits a smooth envelope characteristic, and no high-frequency tremors caused by position pulse compensation occur; by establishing a quantitative correlation between the state transition equation and the damping mapping function, the perception deviation is linearly eliminated at the dynamic response level, enabling the controlled object to maintain the second-order continuity of its trajectory under interference conditions; the vehicle's positioning at the wafer docking station is stable within... Within this scope, ensure the continuity of precision electronics manufacturing tasks.

[0033] Example 5: When the system faces the challenge of changing the vehicle model or being deployed in a condition with specific friction characteristics, the main controller initiates a pre-calibration program, controlling the vehicle drive unit to... On a straight path The system performs a reciprocating motion at a certain speed, and simultaneously collects the total displacement feedback from the visual sensor. The total number of pulses recorded by the encoder of the drive motor Using the formula Determine the wheel calibration radius ,in, The wheel's calibrated radius, in units of , The physical displacement measured by the vision sensor, in units of , This refers to the number of lines per revolution of the encoder. Total number of accumulated pulses; calibration procedure performed. By sampling and calculating the arithmetic mean, the observed numerical fluctuations are controlled within the baseline value. Within.

[0034] In scenarios where the load distribution changes, the processor executes the slip parameters. The field tests were conducted, and the vehicle received signals under both empty and fully loaded conditions. to The duty cycle step drive torque is monitored by the system as the drive motor accelerates from a standstill to... Instantaneous slip ratio during the process; calculated slip parameters exist to Within the range, the slip parameters exhibit a linear variation trend with the load weight, and the processor will measure the slip parameters. The initial gain factor of the slip correction stage in the kinematic prediction model is stored in the non-volatile memory area of ​​the main controller; the initial trajectory tracking error of the vehicle after completing the debugging procedure is within a certain range. Within a certain range, and when crossing ground junctions with different friction coefficients, the phase deviation residual between the virtual reference pose sequence and the visual feedback result remains within a certain range. scale.

[0035] Example 6: During the maiden voyage testing of the vehicle platform, due to the wheelbase... The physical structural parameters have manufacturing tolerances, the system executes dynamic characteristic characterization procedures, and the processor drives the vehicle to generate a frequency range of [missing information]. to Sweep frequency excitation torque The speed feedback signal is recorded synchronously, and the frequency response curve of the controlled object is generated using the Fast Fourier Transform algorithm; among which, This is the frequency sweep excitation torque, in units of , Wheelbase, in units of The system determines the dynamic damping coefficient by identifying the resonant peak frequency in the response curve. The reference range maps the damping ratio corresponding to the critical damping state to the upper limit of the dynamic damping coefficient. and take it Strength as the lower limit of the dynamic damping coefficient The offline system identifies the torque absorption boundary in the context of mechanical structures, enabling the compensation gain sequence to be optimized. The attenuation characteristics are physically matched with the second-order dynamic characteristics of the vehicle.

[0036] When the system faces processing latency caused by dynamic allocation of computing resources Under fluctuating operating conditions, the processor performs stability boundary stress tests in a controlled environment. The simulation delay of visual feedback is added to the gradient, and the convergence rate of the vehicle center path deviation is monitored; the system calculates the phase margin of the feedback loop. and in phase margin Descending to Record the total delay at the critical value, and use the formula Determine the time delay reference threshold ;in, This is the time delay baseline threshold, in units of , Phase margin, in degrees. The cutoff frequency of the control loop, in units of To establish the cutoff frequency of the control loop. The processor injects a wideband excitation torque while the vehicle is traveling at a constant speed under no-load conditions. It then identifies the amplitude gain drop by calculating the amplitude-frequency response of the closed-loop transfer function. The frequency characteristic point at that time, under the hardware architecture of this vehicle, is the cutoff frequency. Locked as The main controller will calibrate the obtained data. The data is stored in the storage unit as a logical trigger threshold for weight reduction, so that the compensation weight coefficient can compensate for the real-time processing latency. When the value is exceeded, linear decay is implemented, and the position error convergence process of the vehicle during operation exhibits a monotonically decreasing characteristic.

[0037] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A vehicle positioning and control method based on machine vision, characterized in that, Includes the following steps: Step 101: Establish a vehicle kinematics prediction model, and use the vehicle kinematics prediction model to generate a virtual reference pose of the vehicle based on the real-time driving instructions of the vehicle drive unit, so as to construct a reference pose sequence synchronized with the vehicle motion. Step 102, at the time of image acquisition by the vision sensor. Record the corresponding first virtual reference pose; Step 103, at the current moment when the visual positioning result corresponding to the image acquisition is solved. The differential deviation vector between the visual positioning result and the first virtual reference pose is calculated. The differential deviation vector is used to characterize the vehicle's position at the acquisition time. Up to the current moment The resulting nonlinear motion deviation; Step 104, Obtain the vehicle's current position. The real-time velocity vector is calculated, the directional angle between the real-time velocity vector and the differential deviation vector is calculated, and the dynamic damping coefficient is determined according to the directional angle based on the preset mapping rules. Step 105: The differential deviation vector is scaled using the dynamic damping coefficient to generate a compensation gain sequence that decays exponentially with time, and the compensation gain sequence is converted into a virtual compensation torque acting on the current loop setpoint of the vehicle drive unit. Step 106: Adjust the output torque of the vehicle drive unit according to the compensation gain sequence to drive the physical pose of the vehicle to converge towards the target pose corrected by the differential deviation vector, so as to reduce the impact of the acquisition time while maintaining the continuity of vehicle dynamics. Up to the current moment The phase lag that occurs during this period.

2. The vehicle positioning and control method based on machine vision according to claim 1, characterized in that, The process of scaling the differential deviation vector using the dynamic damping coefficient includes: obtaining the preset motion envelope constraint value of the vehicle; calculating the initial amplitude of the compensation gain sequence and the time constant used to limit the decay rate based on the dynamic damping coefficient and the preset motion envelope constraint value; and constructing a compensation gain sequence that decays exponentially with time using the initial amplitude and the time constant, so that the electromagnetic torque of the vehicle drive unit can be smoothly adjusted within a preset time window.

3. The vehicle positioning and control method based on machine vision according to claim 1, characterized in that, The differential deviation vector is used to describe the nonlinear displacement deviation of the vehicle during the image processing delay period caused by fluctuations in the friction coefficient of the road surface or changes in the vehicle load.

4. The vehicle positioning and control method based on machine vision according to claim 1, characterized in that, Collection time The phase alignment of observed and predicted data in the time dimension is achieved by synchronizing the trigger pulse of the visual sensor with the system clock of the vehicle's main controller.

5. The vehicle positioning and control method based on machine vision according to claim 1, characterized in that, Before calculating the differential deviation vector, the method further includes: step 501, predicting the predicted motion features of image feature points in the pixel coordinate system based on the output of the vehicle kinematics prediction model; step 502, extracting the actual motion features from the visual positioning results and calculating the consistency deviation value between the actual motion features and the predicted motion features; step 503, determining the confidence level of the current visual positioning based on the consistency deviation value, and adjusting the weight coefficients of the compensation gain sequence in the driving adjustment process according to the confidence level.

6. The vehicle positioning and control method based on machine vision according to claim 1, characterized in that, After calculating the differential deviation vector in step 103, the method further includes: step 601, extracting the radius of curvature information of the preset trajectory in front of the vehicle in real time; step 602, determining the centripetal motion deviation component of the vehicle in the image processing delay period based on the radius of curvature information; step 603, using the centripetal motion deviation component to perform second-order interpolation compensation on the differential deviation vector to correct the dynamic tracking deviation caused by the vehicle turning.

7. The vehicle positioning and control method based on machine vision according to claim 1, characterized in that, In the process of establishing the vehicle kinematic prediction model in step 101, the method also includes dynamic calibration of the vehicle kinematic prediction model, the steps of which include: step 701, applying a micro-pulse excitation torque of a preset frequency to the vehicle drive unit; step 702, acquiring the instantaneous speed feedback of the vehicle under the action of the micro-pulse excitation torque, and extracting the response features corresponding to the preset frequency; step 703, using the response features to calculate the slip parameters between the vehicle and the road surface. The calculation formula is: ,in, For the slip parameter, In response to the characteristic amplitude, The amplitude of the micropulse excitation torque. The instantaneous speed feedback is the phase lag angle relative to the micro-pulse excitation torque; step 704, based on the slip parameter Correct the driving gain coefficient of the vehicle kinematics prediction model.

8. The vehicle positioning and control method based on machine vision according to claim 5, characterized in that, Before calculating the differential deviation vector in step 103, the method further includes: step 801, calculating the real-time processing delay of the data acquired by the visual sensor from image acquisition to solution completion; step 802, reducing the weight coefficient when the real-time processing delay exceeds a preset delay reference threshold; and step 803, maintaining the motion control output of the vehicle using the reference pose sequence generated by the vehicle kinematic prediction model.

9. The vehicle positioning and control method based on machine vision according to claim 1, characterized in that, In step 105, when injecting the compensation gain sequence into the current loop setpoint of the vehicle drive unit, the method further includes using a bandpass filtering algorithm to extract the feedback current characteristics of the drive motor of the vehicle drive unit, and using the feedback current characteristics to perform phase correction on the virtual compensation torque in order to suppress the mechanical impact of the actuator. The adjustment process of the output torque of the vehicle drive unit follows the second-order continuity constraint of the vehicle motion state.

10. A machine vision-based vehicle positioning and control system, used to implement the method described in any one of claims 1-9, characterized in that, include: The kinematic prediction module is used to generate a virtual reference pose of the vehicle by using the vehicle kinematic prediction model and combining it with the real-time drive commands of the vehicle drive unit, so as to construct a reference pose sequence synchronized with the vehicle motion. The acquisition and recording module is used to record the acquisition time when the vision sensor performs image acquisition. Record the first virtual reference pose corresponding to the virtual reference pose. The differential calculation module is used to calculate the visual positioning results at the current moment. Calculate the difference deviation vector between the visual localization result and the first virtual reference pose; The dynamics analysis module is used to obtain the vehicle's current position. The real-time velocity vector is obtained, and the dynamic damping coefficient is determined based on the directional angle between the real-time velocity vector and the differential deviation vector. The compensation mapping module is used to scale the differential deviation vector using the dynamic damping coefficient, generate a compensation gain sequence that decays exponentially with time, and map the compensation gain sequence into a virtual compensation torque acting on the current loop setpoint of the vehicle drive unit. The drive control module adjusts the output torque of the vehicle drive unit according to the compensation gain sequence, driving the physical pose of the vehicle to converge towards the target pose corrected by the differential deviation vector, thereby reducing the impact of the acquisition time while maintaining the continuity of vehicle dynamics. Up to the current moment The phase lag that occurs during this period.

Images