Bionic binocular vision perception device and method based on baseline and pitch angle control

Abstract

This invention relates to the field of visual perception and navigation technology for intelligent devices, specifically a biomimetic binocular visual perception device and method based on baseline and pitch angle control. The device includes a shared pitch platform and left and right cameras mounted thereon. The overall pitch angle is controlled by a rotation actuator, and the position of the right camera is adjusted by a linear actuator to change the baseline. A control unit executes perception, adjustment, calibration, and decision-making steps. This invention resolves the contradiction between near-field blind spots and long-range accuracy by extracting the distance to the nearest obstacle within the collision cone and actively adjusting the baseline. It actively adjusts the pitch angle based on the target's projected coordinates to lock onto the target, performs calibration using static feature point bundle adjustment, and finally fuses the target heading and safe obstacle avoidance heading based on the dynamic weights of the Sigmoid function. This improves the environmental adaptability, measurement accuracy, and flight safety of UAVs in complex environments.

Description

Technical Field

[0001] This invention relates to the field of visual perception and navigation technology for intelligent devices, specifically to a biomimetic binocular visual perception device and method based on baseline and pitch angle control. Background Technology

[0002] With the widespread application of intelligent devices such as drones and robots in fields such as industrial inspection, logistics and transportation, and public safety, enabling these intelligent devices to safely and efficiently replace human labor in more complex environments and perform tasks requires them to possess good environmental perception and navigation capabilities as a prerequisite for achieving these goals.

[0003] Among numerous sensing technologies, binocular vision has become one of the mainstream sensing technologies due to its ability to acquire rich 3D depth information of the environment, relatively low hardware cost, and low power consumption. However, most current binocular vision systems have a static structure, meaning that the spacing between the cameras and the viewing angle of the cameras are fixed. This non-adjustable configuration has limitations when dealing with complex scenes, mainly in the following two aspects: First, according to the physical principle of binocular ranging, depth measurement error... Distance to the object being measured It is directly proportional to the square of the baseline and inversely proportional to the baseline length b (i.e., A fixed baseline cannot simultaneously meet the needs of long-range detection and short-range obstacle avoidance. A smaller baseline results in higher measurement accuracy for nearby objects but a lower minimum measurable distance, which is beneficial for obstacle avoidance in confined spaces. However, it lacks sufficient detection capability for distant obstacles and offers insufficient early warning capabilities. Conversely, a larger baseline, while improving long-range detection capabilities, also increases the near-field blind spot and may lead to failure in recognizing nearby objects due to excessive parallax. Secondly, a fixed forward-facing perspective means that the device's observation direction is consistent with its movement direction, which limits the device's flexibility and environmental awareness. For example, in bridge inspection or power line inspection tasks, drones need to raise their viewing angle while flying horizontally to observe the structure under the bridge or on top of the tower, which is difficult for fixed-angle devices to achieve.

[0004] Relevant patent documents retrieved:

[0005] The document, published in China (CN113259589B) on May 2, 2023, discloses a method and apparatus for intelligent perception using a binocular camera with adaptive baseline adjustment. The technical solution involves identifying a specific target object from an image, calculating its measurement error, comparing the error with a preset expected value, and if the error is too large, calculating a suitable expected baseline length and driving a motor to adjust the baseline to that expected value. This ensures that the depth error for the specific target remains within a small range.

[0006] The application, published in China (CN110442145A) on November 12, 2019, discloses a binocular vision-based gimbal obstacle avoidance system and method for multi-rotor UAVs. The technical solution involves mounting a binocular camera on a three-axis gimbal. The gimbal primarily utilizes the real-time pitch and roll angles obtained from the UAV's attitude module to compensate for motion in these angles, thereby maintaining the binocular camera in a horizontal position and eliminating the problem of measurement reference changes caused by fuselage tilt.

[0007] The prior art represented by the aforementioned documents has at least the following unresolved technical problems or defects: Regarding the aforementioned patent document CN113259589B, although the technical solution of this application achieves baseline adjustment, its technical purpose is limited. The technical problem it aims to solve is "how to measure a single target more accurately." Its goal is "accurate measurement," not "good flight." This technical solution does not provide any instruction on how to use precise depth information to plan a path or avoid obstacles, nor does it have the ability to actively adjust the pitch angle to adapt to mission requirements.

[0008] Regarding the aforementioned patent document CN110442145A, although the technical solution utilizes a gimbal to adjust the viewing angle, its control logic passively compensates for the camera angle. The gimbal's function is to maintain horizontality to eliminate the influence of the drone's flight attitude. However, this device cannot perform tasks requiring upward observation of the bridge's underside. Furthermore, this solution uses a fixed baseline, failing to dynamically optimize its depth perception capabilities based on environmental distance.

[0009] Therefore, there remains an urgent need in the field for a technical solution that can overcome the aforementioned limitations. This solution should be able to collaboratively control the baseline length and pitch angle of the binocular vision module, enabling the device to be equally applicable in complex environments. Summary of the Invention

[0010] In order to solve at least one of the above-mentioned technical problems existing in the prior art, the present invention provides a biomimetic binocular vision perception device and method based on baseline and pitch angle control.

[0011] To achieve the above objectives, the technical solution of the present invention is as follows: In a first aspect, the present invention provides a biomimetic binocular vision perception device based on baseline and pitch angle control, comprising: a biomimetic binocular vision module, the biomimetic binocular vision module including a base; a shared pitch platform, the shared pitch platform being pivotally connected to the base via a pitch axis; a left camera, the left camera being fixedly mounted on one end of the shared pitch platform; a linear guide rail, the linear guide rail being fixedly mounted on the shared pitch platform in a horizontal direction; a slider, the slider being slidably connected to the linear guide rail; a right camera, the right camera being fixedly mounted on the slider; a rotary actuator, the rotary actuator being used to drive the shared pitch platform to rotate about the pitch axis; a linear actuator, the linear actuator being connected to the shared pitch platform and the slider to drive the slider to move along the linear guide rail; and a control unit, the control unit being communicatively connected to the left camera, the right camera, the rotary actuator and the linear actuator, the control unit being configured to perform the following operations: The system controls the left and right cameras to acquire images and calculates a real-time depth map; based on the depth map and a preset collision cone region, it extracts the distance to the nearest obstacle. Based on the target point and the current pose of the device, calculate the projected coordinates of the target point in the image coordinate system. ; and calculate the safe course based on the passable area in the depth map. ; Based on the nearest obstacle distance The desired baseline length is calculated using the baseline control law. And drive the linear actuator to adjust the distance between the left and right cameras; simultaneously, based on the projection coordinates The vertical deviation from the image center is used to calculate the pitch angle adjustment amount through the pitch control law. And drive the rotary actuator to adjust the angle of the shared pitch platform; After adjusting the baseline length or pitch angle, static sparse feature points in the left and right images are extracted and matched, and the external parameter matrix of the binocular system is updated by minimizing the reprojection error. Based on the nearest obstacle distance Calculate dynamic weights Utilizing the dynamic weights Desired heading toward the global target and safe course By performing weighted fusion, the final target heading is obtained. .

[0012] Furthermore, the linear actuator includes a stepper motor fixed to the shared pitch platform and a lead screw and nut mechanism connected to the slider.

[0013] Furthermore, the control unit employs a hysteresis control strategy when driving the linear actuator: According to the formula Calculate the expected baseline length, where The preset environmental gain coefficient, This is the physical minimum baseline of the bionic binocular vision module. This is the physical maximum baseline of the bionic binocular vision module; Real-time monitoring of the current actual baseline length Only if the condition is met Only then are drive commands sent, among which This is the preset hysteresis threshold.

[0014] Furthermore, the pitch control law in the control unit is configured to use proportional control, with the following formula: in, This is the pitch control proportional coefficient. The vertical center coordinates of the image. The vertical coordinate is the projected coordinate of the target point; the rotation actuator adjusts the pitch angle independently of the baseline length adjustment.

[0015] Furthermore, the control unit is specifically configured to update the external parameter matrix of the binocular system as follows: Detect sparse feature points in the left and right images of the current frame; IMU data is used to remove feature points belonging to dynamic objects, retaining only feature points of the static environment; Using epipolar constraints, stereo matching is performed on the preserved static environmental feature points; An optimization equation is constructed with the goal of minimizing the reprojection error. Using the matched feature point pairs, the relative translation vector T and rotation correction matrix R between the left and right cameras are solved and updated.

[0016] Furthermore, the calculation formula for the dynamic weight w configured in the control unit is as follows: ; in, To adjust the sensitivity coefficient, The preset safe distance threshold; The formula for calculating the target heading is: ; when Much larger hour, The flight decision of the device tends towards 0. ; when Close to or less than hour, Approaching 1, the device's flight decision focuses on .

[0017] Furthermore, the control unit extracts the distance to the nearest obstacle. The specific method is as follows: Based on the device’s current flight speed and physical size, a forward collision cone region is defined in the image space, which represents the danger zone in which the device is most likely to collide in the short term. The depth values ​​within the collision cone region are analyzed, and noise points caused by matching errors are removed using median filtering. The minimum valid depth value within this region is then taken as the minimum depth value. .

[0018] Secondly, the present invention also provides a biomimetic binocular vision perception method based on baseline and pitch angle control, applied to the aforementioned biomimetic binocular vision perception device based on baseline and pitch angle control, comprising the following steps: Step S1: Control the left and right cameras to acquire images, and calculate a real-time depth map using SGM or deep learning algorithms; based on the depth map, define a forward collision cone region in the image space, combined with the device's current flight speed and physical size, and extract the distance to the nearest obstacle from it. Based on the target point and the current pose of the device, calculate the projected coordinates of the target point in the image coordinate system. ; and calculate the safe course based on the passable area in the depth map. ; Step S2: Based on the distance to the nearest obstacle The desired baseline length is calculated using the baseline control law. And drive the linear actuator to adjust the distance between the left and right cameras; simultaneously, based on the projection coordinates The vertical deviation from the image center is used to calculate the pitch angle adjustment amount through the pitch control law. And drive the rotary actuator to adjust the angle of the shared pitch platform; Step S3: After adjusting the baseline length or pitch angle, extract and match sparse feature points in the left and right images, remove feature points belonging to dynamic objects by combining IMU data, perform stereo matching on the retained static feature points using epipolar constraints, and update the external parameter matrix of the binocular system by minimizing the reprojection error. Step S4: Based on the distance to the nearest obstacle Calculate the dynamic weight w, and use the dynamic weight w to determine the desired heading towards the global target. and the aforementioned safe course By performing weighted fusion, the final target heading is obtained. And fly based on the target heading control device.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention provides a biomimetic binocular vision perception device based on baseline and pitch angle control. It employs a shared pitch platform and a linear variable baseline design, where the left and right cameras achieve continuous baseline adjustment within a preset physical range via linear actuators. The device can also perform a wide pitch range from -90° to +90° with the platform. This design breaks through the limitations of traditional fixed binocular viewing angles, utilizing a large baseline to ensure depth accuracy for long-distance measurements, and a small baseline to eliminate parallax blind spots in near-distance perception. Combined with independent pitch control, it achieves omnidirectional target locking of targets above (e.g., under a bridge) or below, improving the device's environmental adaptability and detection coverage in complex three-dimensional spatial tasks such as bridge inspection.

[0020] 2. This invention provides a biomimetic binocular vision perception method based on baseline and pitch angle control. Through the synergistic application of collision cone risk extraction and hysteresis control strategies, it effectively solves the stability problem of variable baseline systems in dynamic environments. The system defines a forward collision cone and only extracts collisions that pose a substantial threat to flight. This avoids interference from irrelevant obstacles on the side; at the same time, a hysteresis threshold is introduced. The motor is only driven when the expected baseline change exceeds the threshold, which not only enables proactive adaptation to environmental risks, but also prevents frequent motor fine-tuning and screen oscillation caused by measurement noise, effectively extending the service life of the mechanism.

[0021] 3. This invention provides a biomimetic binocular vision perception method based on baseline and pitch angle control. Combining adaptive calibration with a Sigmoid dynamic weight fusion algorithm, it effectively solves the problems of inaccurate extrinsic parameters caused by motion and abrupt navigation mode switching. By combining IMU to remove dynamic feature points and updating binocular extrinsic parameters in real time based on static environmental features, it effectively compensates for minor errors caused by mechanical gaps, ensuring high accuracy in depth calculation. Furthermore, by using the Sigmoid function to nonlinearly map obstacle distances into smooth weights, it enables smooth flight and operation of the UAV in complex obstacle environments. Attached Figure Description

[0022] Figure 1 This is an overall flowchart of the control method for the bionic binocular vision module in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of dynamic weighted heading fusion of the bionic binocular vision module in an embodiment of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0025] It should be noted that, unless otherwise specifically stated, the relative arrangement and numerical expressions of the components and steps described in these embodiments should not be construed as limiting the scope of the invention.

[0026] The following description of exemplary embodiments is merely illustrative and is not intended to limit the invention or its application or use in any way. Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but where applicable, such techniques, methods, and apparatus should be considered part of this specification.

[0027] Example 1 This embodiment provides a biomimetic binocular vision module. The module uses a base as its supporting structure, which has an interface for connecting to the fuselage of intelligent devices such as drones. The base is pivotally connected to a shared pitch platform via a pitch hinge. The shared pitch platform is the core component for achieving pitch angle adjustment.

[0028] The shared pitch platform is also connected via a gear or linkage mechanism rotary actuator, which drives the platform to perform a wide range of overall pitch movements. In this embodiment, the rotary actuator is preferably a servo motor with an encoder. The control unit can drive the shared pitch platform to perform pitch adjustments from -90° (vertically downward) to +90° (vertically upward) around the pitch axis by controlling the rotation angle of the motor.

[0029] A left camera is fixedly mounted at one end of the shared pitch platform, serving as the reference point for the binocular system; its pose relative to the platform is fixed. A linear guide rail is mounted on the shared pitch platform along the horizontal direction (i.e., the baseline direction), and a sliding slider is mounted on the linear guide rail. The right camera is mounted on the slider. The baseline distance between the left and right cameras is controlled by a linear actuator. In this embodiment, the linear actuator is preferably a stepper motor coupled with a lead screw and nut mechanism. One end of the lead screw is connected to the shared pitch platform, and the lead screw and nut mechanism is fixedly connected to the slider. The control unit can precisely drive the slider to move along the linear guide rail by controlling the number of rotation steps and direction of the motor, thereby changing the horizontal distance between the right camera and the left camera, and thus achieving the desired baseline length. Within the preset physical range It enables continuous and precise adjustment.

[0030] Throughout the adjustment process, the rotary actuator drives the entire shared pitch platform and its mounted components for overall pitch adjustment. Inside the shared pitch platform, the linear actuator controls the horizontal position of the right camera, thereby adjusting the baseline length. This design achieves control and adjustment of both the pitch angle and the horizontal baseline distance using only two independent actuators, resulting in a compact structure and manageable cost.

[0031] Example 2 This embodiment provides a biomimetic binocular vision perception method, the specific steps of which are as follows: Step S1: At the beginning of each control cycle, the control unit acquires a pair of synchronized left and right images in real time using a binocular camera. Using SGM or deep learning algorithms, a real-time depth map of the scene is calculated. Subsequently, the control unit analyzes the depth map and extracts the following important parameters: The first is the nearest obstacle distance, which represents the immediate environmental risk. To avoid interference and ensure the stability of this value, the system does not simply take the global minimum of the depth map. Instead, it first defines a forward "collision cone" region in the image space based on the UAV's current flight speed and physical size. This region represents the danger zone where the UAV is most likely to collide in the short term. The system only analyzes the depth values ​​within this region, and after removing noise points caused by matching errors through median filtering and other methods, it takes the minimum valid depth value within this region as the minimum depth value. .therefore This is the core parameter for quantifying the collision risk in the current environment.

[0032] The second is the image coordinates of the target point guiding the task direction. These coordinates are derived from the target navigation point of this mission, combined with the UAV's real-time pose and camera parameters, through coordinate system transformation and projection calculations, thus converting the navigation target into a specific visual tracking object.

[0033] The third is safe course. The system marks all pixels with depth values ​​greater than a preset safe distance threshold on the depth map as "passable regions." Then, it uses a connected component analysis algorithm to find the largest connected region, representing the safest path for the current task. The geometric center point of this region is projected back into 3D space to obtain a direction vector pointing to the safest space, denoted as the safe heading. .

[0034] Step S2: This step is the core of the invention. Based on the different parameters extracted in step S1, the control unit simultaneously sends adjustment commands to the two actuators, converting the sensed parameters into active adjustments of the device. Baseline control: The control unit controls and calculates the baseline length according to the following formula. ,in This is the environmental gain coefficient, pre-calibrated according to different scenarios (e.g., indoor, outdoor). It also incorporates physical constraints. Perform amplitude limiting, i.e. The principle of this control law is that it correlates the baseline length with the environmental risk. Related. When drones fly in open areas The system automatically increases the baseline to improve ranging accuracy for distant objects, enabling it to see further and provide earlier warnings; this is especially important when the drone enters narrow or obstacle-filled areas. If the value is small, the system automatically reduces the baseline to avoid nearby blind spots, achieving both clear visibility and accurate avoidance.

[0035] At the same time, in order to avoid in To address the mechanical wear and screen oscillations caused by frequent motor fine-tuning during fluctuations, this invention introduces a hysteresis control strategy. The system is configured with a hysteresis threshold. The control unit monitors the current actual baseline length in real time. Only if the condition is met Only when the condition is met will a drive command be sent to the linear actuator; otherwise, the current baseline will remain unchanged.

[0036] Pitch angle control: The control unit uses proportional control to calculate the pitch angle adjustment amount. ,in The ordinate of the image center. Here is the ordinate of the projected coordinates of the target point. The principle behind this formula is to eliminate the vertical deviation between the mission target and the current position of the UAV through negative feedback adjustment. Regardless of how the UAV flies, the system will actively adjust the pitch angle to lock the mission target in the center of the field of view for tracking. The control unit sends pitch angle adjustment commands to the rotary actuator.

[0037] Step S3: To ensure high accuracy, the system triggers an online calibration process after each change in baseline or pitch angle. This step is crucial because even minor gaps in the mechanical structure can cause the pre-calibrated extrinsic parameters to fail. Specifically, the system first extracts and matches sparse feature points such as ORB or SIFT in the left and right images of the current frame. To prevent dynamic interference, this step uses conventional dynamic point culling techniques to filter feature points: for example, calculating epipolar distances using IMU data to remove points that do not meet epipolar constraints; or using optical flow analysis to analyze the motion vectors of feature points to remove dynamic features inconsistent with the background motion (such as pedestrians and vehicles), retaining only static environmental features. Subsequently, stereo matching is performed using the retained static feature points. The system constructs a bundle adjustment optimization model to minimize the reprojection error of feature points on the image, quickly calculating and updating the relative translation vector between the left and right cameras. (i.e., actual baseline length) and rotation correction matrix This allows for real-time compensation for any minor errors that may arise during adjustments. This step ensures that the accuracy of subsequent depth calculations remains guaranteed even during frequent dynamic adjustments.

[0038] Step S4: After the vision module is optimized, the control unit makes a decision using the perception parameters. The key to this step is determining whether to fly towards the target or avoid obstacles, two sometimes conflicting behaviors. Therefore, this invention introduces a dynamic weighting mechanism. Parameters are used for judgment.

[0039] The system is based on the formula The formula for calculating weights utilizes the properties of the Sigmoid function. This continuously changing parameter is mapped to a numerical value that smoothly varies between 0 and 1. Wherein To adjust the sensitivity coefficient, This is the safe distance threshold. This determines the steepness of the weight change. Larger... The value indicates that the system is more sensitive to changes in distance and that weights switch more quickly; This defines the critical point for switching from "free flight" to "obstacle avoidance flight".

[0040] when Much greater than the safe distance hour, Approaching 0 means environmental safety; when Close to or less than hour, It rapidly approaches 1.

[0041] Specifically: when When considering (environmental safety), the index term... The value tends towards infinity, resulting in an extremely large denominator, which makes... This means that when far from obstacles, the obstacle avoidance weight is extremely low, and the system determines that the environment is safe.

[0042] when At the critical point, the exponential term approaches... , making This means that when the distance to the obstacle is just within the safe threshold, flying towards the target and avoiding the obstacle have equal priority.

[0043] when When danger is imminent, the exponential term The denominator tends towards 0, causing it to tend towards 1, thus... This means that as obstacles approach, obstacle avoidance becomes the dominant factor.

[0044] Finally, the system uses the heading fusion formula. The heading will directly point to the mission objective. and safe course Perform linear combinations.

[0045] In a safe time ( Drones are mainly composed of Driven by the ability to fly efficiently to the destination via the shortest path; In dangerous situations ( The drone's heading is from Its primary task is to perform obstacle avoidance, thereby ensuring flight safety.

[0046] This smooth transition mechanism based on the Sigmoid function avoids the sudden changes in flight attitude caused by hard switching in traditional logic, and achieves a smooth and consistent flight trajectory.

[0047] Step S5: Calculated Ultimately, the data is sent to the underlying PID flight controller, which calculates the underlying motor control commands that drive the drone to roll, yaw, and move forward, completing the entire closed-loop control of "perception-decision-action".

[0048] Example 3 This embodiment provides a specific control process for a bridge inspection task: Setting parameters 10cm =50cm, environmental gain =0.1, safe distance 5m, hysteresis threshold cm Phase A: Approaching the Bridge Pier from a Distant Distance: The drone flies from its takeoff point to a distant bridge pier. The drone first ascends to the same height as the target pier and then begins to approach it. At this point, the collision cone... The detection range is 100 meters, according to , 0.1 × 100m = 10m, after amplitude limiting It was truncated to the maximum physical baseline of 50cm. Due to |50cm- |> The linear actuator drives the right camera to move to the farthest end of the guide rail. The pitch angle is calculated using the formula... Adjust the position to center the bridge pier in your field of vision. At this point... =100m The weight formula is calculated as follows (Close to 0), heading fusion result The drone flew directly to the target at full speed.

[0049] Phase B: Close-range obstacle avoidance: When the drone approaches the bridge pier, assuming that currently... Reduced to 3m, system calculation theoretical baseline Assuming the baseline was 50cm at the previous moment, the difference at this moment is... This triggers an adjustment action, shrinking the baseline to 30cm to eliminate near-field binocular blind spots. During this process, the online calibration program corrects for parameter errors caused by baseline changes in real time. Simultaneously, it calculates... Because the exponent term becomes smaller, It rapidly increases and approaches 1.0 (assuming) ),course More on safe course The decision was made to deviate the drone from its straight path and fly along a plane at the same height as the bridge pier to avoid obstacles.

[0050] Phase C: Inspecting the Bridge Base from Above: When the drone reaches the area beneath the bridge piers, the mission objective is updated to the bridge base, and the drone begins its descent to the preset bridge base height. At this point, the inspection objective is above the drone, utilizing... The formula is used to calculate and generate a large adjustment command, driving the pitch angle to nearly 90 degrees to achieve a downward view; simultaneously, assuming the sensor detects a distance of 8 meters from the bottom of the bridge, the calculation... After being limited, it is set to If the current baseline is still 30cm, the adjustment is triggered again to lengthen the baseline to 50cm for optimal imaging accuracy. Throughout the process, the drone hovers horizontally or moves smoothly, relying on the deformation of the vision module to complete the detection task from complex perspectives. After completing the detection in this area, the pitch angle is adjusted again to near 0 degrees to achieve a level view, and then the drone flies to another target area to be detected. During this period, the drone repeats the above operations for obstacle avoidance flight and to perform inspection tasks.

[0051] The above specific 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 examples, 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 scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A biomimetic binocular vision perception device based on baseline and pitch angle control, characterized in that: include: A bionic binocular vision module, comprising a base; a shared pitch platform, which is pivotally connected to the base via a pitch pivot; a left camera, which is fixedly mounted on one end of the shared pitch platform; and a linear guide rail, which is fixedly mounted on the shared pitch platform in a horizontal direction. A slider is slidably connected to the linear guide rail; a right camera is fixedly mounted on the slider. A rotary actuator for driving the shared pitch platform to rotate about a pitch axis; a linear actuator connecting the shared pitch platform and the slider to drive the slider to move along a linear guide; and a control unit communicatively connected to the left camera, right camera, rotary actuator, and linear actuator, the control unit being configured to perform the following operations: The system controls the left and right cameras to acquire images and calculates a real-time depth map; based on the depth map and a preset collision cone region, it extracts the distance to the nearest obstacle. ; Calculate the projected coordinates of the target point in the image coordinate system based on the target point and the current pose of the device. ; and calculate the safe course based on the passable area in the depth map. ; Based on the nearest obstacle distance The desired baseline length is calculated using the baseline control law. And drive the linear actuator to adjust the distance between the left and right cameras; simultaneously, based on the projection coordinates The vertical deviation from the image center is used to calculate the pitch angle adjustment amount through the pitch control law. And drive the rotary actuator to adjust the angle of the shared pitch platform; After adjusting the baseline length or pitch angle, static sparse feature points in the left and right images are extracted and matched, and the external parameter matrix of the binocular system is updated by minimizing the reprojection error. Based on the nearest obstacle distance Calculate dynamic weights Utilizing the dynamic weights Desired heading toward the global target and safe course By performing weighted fusion, the final target heading is obtained. .

2. The biomimetic binocular vision perception device based on baseline and pitch angle control according to claim 1, characterized in that: The linear actuator includes a stepper motor fixed to the shared pitch platform and a lead screw and nut mechanism connected to the slider.

3. The biomimetic binocular vision perception device based on baseline and pitch angle control according to claim 2, characterized in that: When driving the linear actuator, the control unit employs a hysteresis control strategy: According to the formula Calculate the expected baseline length, where The preset environmental gain coefficient, This is the physical minimum baseline of the bionic binocular vision module. This is the physical maximum baseline of the bionic binocular vision module; Real-time monitoring of the current actual baseline length Only if the condition is met Only then are drive commands sent, among which This is the preset hysteresis threshold.

4. The biomimetic binocular vision perception device based on baseline and pitch angle control according to claim 1, characterized in that: The pitch control law in the control unit is configured to use proportional control, and the formula is: in, This is the pitch control proportional coefficient. The vertical center coordinates of the image. The vertical coordinate is the projected coordinate of the target point; the rotation actuator adjusts the pitch angle independently of the baseline length adjustment.

5. The biomimetic binocular vision perception device based on baseline and pitch angle control according to claim 1, characterized in that: When updating the external parameter matrix of the binocular system, the control unit is specifically configured as follows: Detect sparse feature points in the left and right images of the current frame; IMU data is used to remove feature points belonging to dynamic objects, retaining only feature points of the static environment; Using epipolar constraints, stereo matching is performed on the preserved static environmental feature points; An optimization equation is constructed with the goal of minimizing the reprojection error. Using the matched feature point pairs, the relative translation vector T and rotation correction matrix R between the left and right cameras are solved and updated.

6. The biomimetic binocular vision perception device based on baseline and pitch angle control according to claim 1, characterized in that: The formula for calculating the dynamic weight w configured in the control unit is: ; in, To adjust the sensitivity coefficient, The preset safe distance threshold; The formula for calculating the target heading is: ; when Much larger hour, The flight decision of the device tends towards 0. ; when and When approaching, w is approximately equal to 0.5, and the flight heading of the device is... and average value; when Close to or less than hour, Approaching 1.0, the device's flight decision-making focuses on .

7. The biomimetic binocular vision perception device based on baseline and pitch angle control according to claim 1, characterized in that: The control unit extracts the distance to the nearest obstacle. The specific method is as follows: Based on the device’s current flight speed and physical size, a forward collision cone region is defined in the image space, which represents the danger zone in which the device is most likely to collide in the short term. The depth values ​​within the collision cone region are analyzed, and noise points caused by matching errors are removed using median filtering. The minimum valid depth value within this region is then taken as the minimum depth value. .

8. A biomimetic binocular vision perception method based on baseline and pitch angle control, characterized in that: An application to a biomimetic binocular vision sensing device based on baseline and pitch angle control as described in any one of claims 1-7, comprising the following steps: Step S1: Control the left and right cameras to acquire images, and calculate a real-time depth map using SGM or deep learning algorithms; based on the depth map, define a forward collision cone region in the image space, combined with the device's current flight speed and physical size, and extract the distance to the nearest obstacle from it. Based on the target point and the current pose of the device, calculate the projected coordinates of the target point in the image coordinate system. ; and calculate the safe course based on the passable area in the depth map. ; Step S2: Based on the distance to the nearest obstacle The desired baseline length is calculated using the baseline control law. And drive the linear actuator to adjust the distance between the left and right cameras; simultaneously, based on the projection coordinates The vertical deviation from the image center is used to calculate the pitch angle adjustment amount through the pitch control law. And drive the rotary actuator to adjust the angle of the shared pitch platform; Step S3: After adjusting the baseline length or pitch angle, extract and match sparse feature points in the left and right images, remove feature points belonging to dynamic objects by combining IMU data, perform stereo matching on the retained static feature points using epipolar constraints, and update the external parameter matrix of the binocular system by minimizing the reprojection error. Step S4: Based on the distance to the nearest obstacle Calculate the dynamic weight w, and use the dynamic weight w to determine the desired heading towards the global target. and the aforementioned safe course By performing weighted fusion, the final target heading is obtained. And fly based on the target heading control device.

Images