A disturbance satellite target tracking and positioning method fusing binocular information

Abstract

The embodiment of the present application provides a kind of satellite target tracking and positioning method of disturbance fusion binocular data, comprising: based on the corner point information and ID information etc. of the disturbed sub-satellite target collected by camera, initialize target tracker;A group of KCF filters is used to construct the target tracker of the disturbed sub-satellite in this paper, mainly including four corner point trackers and a center tracker;Based on the tracker, the target response is calculated in real time in the image frame, and the current tracker position and pyramid level are updated at the highest response;Based on the tracking of sub-satellite target, the target credibility is constructed to represent the tracking stability of target;Finally, based on the target credibility, binocular position solving method or based on the pose solving method of target is selected, the fusion of binocular positioning and monocular target solving is completed, and the visual positioning for disturbed sub-satellite target is realized.According to the technical scheme provided by the embodiment of the present application, the fusion binocular data of disturbed sub-satellite target tracking and positioning can be provided with reference.

Description

Technical Field

[0001] This invention relates to the field of space robot technology, and in particular to a method for tracking and locating perturbed satellite targets by fusing binocular data. Background Technology

[0002] With the increasing demand for space exploration, the significance and role of autonomous on-orbit servicing through space robots are becoming increasingly prominent. These robots have been widely applied to various missions involving large spacecraft, becoming one of the most important methods for on-orbit servicing. On-orbit target acquisition is a crucial step in on-orbit operations, and accurate and efficient target acquisition requires high-precision visual servo acquisition technology. The accuracy of visual positioning directly affects the success rate of subsequent servo acquisition, making the research of visual positioning and detection methods that balance stability and accuracy of great practical significance. Currently, targets are widely used for spacecraft positioning. Target-based spacecraft positioning can obtain more accurate spacecraft pose information using arm-mounted cameras, facilitating subsequent servo acquisition operations.

[0003] However, existing target-based satellite positioning still relies on real-time single-frame calculations. While it performs well for hovering or slow-moving satellite targets, its success rate is low under conditions of rapid satellite perturbation or target image movement, and target attitude fluctuations frequently occur, significantly impacting the positioning performance of dynamic spacecraft and leading to insufficient stability and accuracy in the positioning process. Furthermore, this method, relying solely on monocular images for target positioning, cannot guarantee robustness and cannot fully utilize the binocular cameras mounted on the end effector of current space robotic arms. Therefore, proposing a perturbation-based satellite target tracking and positioning method that fuses binocular information is of significant research importance. Summary of the Invention

[0004] In view of this, an example of the present invention provides a method for tracking and locating a perturbed satellite target by fusing binocular information, comprising:

[0005] Based on the selected ID information and target parameters such as target size, the target features are first extracted from the image captured by the camera to obtain the information of the four corner points and the target number information.

[0006] Based on the corner information and target number information, the target tracker is initialized and a target tracker group is generated, that is, a group of trackers is used to form the target tracker in this paper;

[0007] Using the initialized target tracker, the target position changes are tracked in the subsequent real-time image stream, and a target response map is calculated; based on the response map, it is determined whether the target tracker has lost track; if not lost, the target tracker position is updated in real time; and if the tracker is lost, the target tracker is reinitialized based on the first two steps.

[0008] The reliability of the tracking is calculated in real time based on the position update of the trackers. The reliability directly represents the reliability of the tracking of the group of trackers. The final pose calculation strategy is switched according to the reliability. If the reliability is high, the target-based pose calculation method is selected, and if the reliability is low, the position calculation method based on stereo matching is selected.

[0009] Based on the trust index and the solution results of the pose calculation method, the pose information of the satellite target is output to complete the target localization.

[0010] In the above method, the target feature extraction process mainly extracts target information from the image. Each target information includes data such as the coordinates of the four corner points and the target ID information.

[0011] The content mainly represents the target's position information in the image, which can be expressed as:

[0012] For images The target objects detected can all be expressed by the following formula:

[0013]

[0014] Where c represents the image The pixel coordinates of the four corners of the marker m. Indicates the center coordinates of the mark, and also records The observed target area.

[0015] In the above method, a target tracker group is generated based on the target information. Each tracker group contains four target corner trackers and one target global tracker. Each initially defined tracker contains five target image patches. and The setting function Identifying images China and Israel Centered on, size is The image patch, which is the side length of the KCF tracker designed in this paper, is... Tracking images.

[0016] Based on the image patch, a global target tracker is generated using the KCF filtering algorithm. and four target corner trackers for:

[0017]

[0018] in This represents the KCF tracker at the center and four corners of the marker; it also records the current marker image pyramid level. This indicates the pyramid level at time t related to target m, which is the parameter representation used to characterize changes in image scale (away from or near).

[0019] In the above method, an initialized target tracker is used to track changes in the target position in subsequent real-time image streams, updating the target tracker position in real time. This mainly involves responding to the target tracker's response in subsequent images (t>1) based on the image response. The response map can be obtained by performing frequency domain correlation operations on the current image using the filter bank defined in the target tracker defined above. ,in Element-wise multiplication is represented by *, complex conjugation is represented by X, and X represents all sides of length X in the current image frame. A collection of patch blocks. The scaling factor, or pyramid hierarchy, needs to be considered during response calculation. To mitigate the impact of [the pandemic], a multi-scale parallel computation and weight fusion method is proposed for response update. Specifically, the response map is obtained from two adjacent pyramid scales. , The response graph is calculated as follows:

[0020] Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ;

[0021] Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ;

[0022] Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ;

[0023] And based on the weights, the response information is extracted to obtain a more accurate response; that is, the response map of the current target can be defined as...

[0024]

[0025] in The weights are on scale l. The final result is a response plot at time t on scale l. This is the response image from this detection. And based on the three pyramid scales calculated in this study... , and By comparing the responses in the image, the pyramid level with the largest response in the response map is selected as the new pyramid level recorded in the current tracking image. ;

[0026] Obtain the corresponding diagram Then, the position of the largest response is calculated and denoted as p. The new target tracker position can then be updated to... In this formula, PSR represents the filter centered at pixel p in image I. The response. Based on the response graph. The maximum response value calculation function is used to obtain the tracker corresponding to the center coordinate in the tracker group. Maximum response value (for the tracker used to track the center of the marker) Define the tracker loss threshold as If the maximum response value is lower than the loss threshold, the tag tracking is considered lost.

[0027] In the above method, the trustworthiness index mainly characterizes the clarity of the target image under the motion condition, and its calculation mainly includes:

[0028] The trustworthiness index mainly consists of two parts: firstly, the polygonal bit sequence of the target extracted based on the coordinates of the four corner points obtained by the KCF target tracker. and the tracking target position sequence detected during initial detection The credibility is initially determined by the normalized Hamming distance, based on , Constructing an F1 function to initially assess its reliability:

[0029]

[0030] Where H represents the normalized Hamming distance calculation function. The trust index for the first part is calculated as above.

[0031] Secondly, the loss status of the four corner filters created in the target tracker group can be considered. Based on the loss status of the four corner trackers created in the target tracker group, i.e., the response status of the four corner trackers, the reliability of the detection can also be determined. Based on this, an F2 function can be constructed to further determine the reliability.

[0032]

[0033] in

[0034]

[0035] Where PSR represents the maximum response of the tracker. This represents the threshold at which a target tracker is considered lost; if the target tracker's response is less than this threshold, the target tracker is considered lost.

[0036] Combining the F1 and F2 functions mentioned above, the metric for the trustworthiness of this paper is constructed as follows:

[0037]

[0038] Based on this level of trustworthiness, it is easy to assess the ambiguity and trustworthiness of the target being tracked by the tracker.

[0039] The above method, which uses a trust-based pose calculation approach, ultimately outputs the target's position or pose information. Its main calculation process is as follows:

[0040] Set a trust threshold As a criterion, when the confidence level is less than the threshold, the target image is considered to be highly blurred, and a binocular position calculation method is used to calculate the position information; when the confidence level is greater than the threshold, the target image is considered to be highly sharp, and a target pose calculation method is used to output the target pose information. The specific formula for this judgment can be expressed as follows:

[0041]

[0042] In the formula This represents the pose corresponding to m marked at time t, and the calculated target confidence level. Below the threshold When the problem is not in a state of equilibrium, a more robust binocular position calculation method is used; otherwise, a more accurate target-based pose calculation strategy is used.

[0043] Binocular position calculation method: When the reliability of the current target tracker is low, that is, when the target moves fast and the image becomes blurry, it is impossible to obtain a clear and accurate target image. However, the center pixel of the target can be obtained through the target tracker. Based on the pixel position in the binocular camera, the actual position information of the target can be obtained through binocular stereo matching. This method has excellent robustness.

[0044] Pose calculation strategy: When the current target tracker has a high degree of reliability, it means that the target motion is uniform and the target image is clear. A target image with good clarity can be obtained. Based on this image, the target pose calculation method based on OpenCV can be used to calculate the accurate target pose information.

[0045] Based on the pose information and the corresponding level of trust. Complete the localization of the perturbed substar target, output the calculated target pose information, and continuously execute the tracking and localization steps to achieve continuous target localization operation, thereby realizing continuous tracking and pose calculation of the perturbed substar target.

[0046] As can be seen from the above technical solutions, the embodiments of the present invention have the following beneficial effects:

[0047] This invention proposes a target tracker array to track randomly moving satellite targets. A set of KCF filters is constructed to track the moving satellite target information. The target's pixel positions are located in real time based on the tracking performance of the target tracker array. A tracking reliability mechanism is designed to determine the target's sharpness in real time, and then switch between a more robust binocular position calculation method or a more accurate target pose calculation method to obtain the target position / pose. This invention constructs a real-time visual pose calculation method for perturbed satellite airborne targets, balancing accuracy and robustness, and enabling stable visual localization of robots. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0049] Figure 1 This is an operation flowchart provided in the embodiments of the present invention;

[0050] Figure 2 This is a flowchart of the perturbation sub-star pose calculation algorithm provided in the embodiments of the present invention.

[0051] Figure 3 This is an initialization effect diagram of the sub-star target tracker provided in an embodiment of the present invention.

[0052] Figure 4 This is the real-time pose tracking curve of the sub-star perturbation target provided in the embodiment of the present invention. Detailed Implementation

[0053] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0054] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0055] This invention provides a method for tracking and locating perturbed satellite targets by fusing binocular data. The method includes the following steps; please refer to the operation flowchart. Figure 1 :

[0056] Step 101: Select the target ID and size parameters, extract the target features based on the image captured by the camera, and obtain the information of the four corner points of the target and the target number information.

[0057] Specifically, a target detection function is used to detect all target information in the camera image, and then for the image... All target objects detected can be expressed by the following formula:

[0058]

[0059] Where c represents the image The pixel coordinates of the four corners of the marker m. Indicates the center coordinates of the mark, and also records The observed target area.

[0060] The above steps allow us to extract the target corner position information from the initial frame and record it in a list.

[0061] Step 102: Based on the extracted target information, initialize the target tracker and generate a target filter bank, that is, use a set of filters to form the target tracker in this paper.

[0062] Please refer to Figure 2 This is a schematic diagram of the target tracker generated in this paper.

[0063] That is, after obtaining the target coordinates, first define The area of ​​the tracking block designed in this paper is used as the basis for the function. Identifying images China and Israel Centered on, size is The image patch is obtained by using this function, centered on the position information obtained from the target, to acquire the tracker's tracking image patch. Based on this tracked image patch, the tracker group for target m at time t can be defined as follows:

[0064]

[0065] in This represents the KCF tracker at the center and four corners of the marker; it also initializes and records the current marker image pyramid level here. This indicates the pyramid level at time t related to target m, which is the parameter representation used to characterize changes in image scale (away from or near).

[0066] Step 103 involves real-time tracking of the target position in the image stream based on the initialized tracker. This real-time tracking primarily includes: tracking the target position changes in subsequent real-time image streams based on the initialized target tracker, and updating the target tracker position in real time. This mainly involves responding to the target tracker's response in subsequent images (t>1) based on the image response. The response map can be obtained by performing frequency domain correlation operations on the current image using the filter bank defined in the previously defined target tracker. ,in Element-wise multiplication is represented by *, complex conjugation is represented by X, and X represents all sides of length X in the current image frame. A collection of patch blocks. The scaling factor, or pyramid hierarchy, needs to be considered during response calculation. To mitigate the impact of [the pandemic], a multi-scale parallel computation and weight fusion method is proposed for response update. Specifically, the response map is obtained from two adjacent pyramid scales. , The response graph is calculated as follows:

[0067] Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ;

[0068] Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ;

[0069] Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ;

[0070] And based on the weights, the response information is extracted to obtain a more accurate response; that is, the response map of the current target can be defined as...

[0071]

[0072] in The weights are on scale l. The final result is a response plot at time t on scale l. This is the response image from this detection. And based on the three pyramid scales calculated in this study... , and By comparing the responses in the image, the pyramid level with the largest response in the response map is selected as the new pyramid level recorded in the current tracking image. ;

[0073] After obtaining the response map Then, the position of the largest response is calculated, which is the new target filter position p. The new target tracker position will be updated to... In this formula, PSR represents the filter centered at pixel p in image I. The response. The corresponding target filter here. It will be updated to the position with the highest response in the current response graph. After all target filters have been updated, the pyramid response level of this frame will be updated to the pyramid level with the highest response level corresponding to the filter with the highest response in the current frame, based on the filter response. The maximum response value calculation function is used to obtain the tracker corresponding to the center coordinate in the tracker group. Maximum response value (for the tracker used to track the center of the marker) Define the tracker loss threshold as If the maximum response value is lower than the loss threshold, the tag tracking is considered lost, and tracking initialization needs to be re-performed, that is, the operations performed in steps 101 and 102 are repeated to re-track and locate.

[0074] Step 104: Calculate the trustworthiness of the target tracker in real time based on the target tracker's position update. The trustworthiness directly characterizes the current tracking reliability of the target tracker.

[0075] The specific reliability index for target tracking mainly consists of two parts: firstly, extracting the polygonal bit sequence of the target based on the coordinates of the four corner points obtained by the KCF target tracker. and the tracking target position sequence detected during initial detection The credibility is initially determined by the normalized Hamming distance, based on , Constructing an F1 function to initially assess its reliability:

[0076]

[0077] Where H represents the normalized Hamming distance calculation function. The trust index for the first part is calculated as above.

[0078] Secondly, the loss status of the four corner filters created in the target tracker group can be considered. Based on the loss status of the four corner trackers created in the target tracker group, i.e., the response status of the four corner trackers, the reliability of the detection can also be determined. Based on this, an F2 function can be constructed to further determine the reliability.

[0079]

[0080] in

[0081]

[0082] Where PSR represents the maximum response of the tracker. This represents the threshold at which a target tracker is considered lost; if the target tracker's response is less than this threshold, the target tracker is considered lost.

[0083] Combining the F1 and F2 functions mentioned above, the metric for the trustworthiness of this paper is constructed as follows:

[0084]

[0085] Based on this level of trustworthiness, it is easy to assess the ambiguity and trustworthiness of the target being tracked by the tracker.

[0086] Step 105: Switch the final pose calculation strategy based on the confidence level. If the confidence level is high, indicating a clear image and easy accurate localization, a more accurate target-based pose calculation method can be used. Conversely, if the confidence level is lower than the set detection threshold, a stereo positioning method based on binocular localization is used. For the relationship between the confidence-based switching strategy and the tracking and localization algorithm flow, please refer to [link to relevant documentation]. Figure 3 .

[0087] Specifically, target-based localization algorithms primarily rely on the PnP (Positioning and Positioning) algorithm to obtain the relative pose of the camera coordinate system with respect to the marker coordinate system. This type of target-based localization often requires sufficient image visibility for the camera to correctly detect the target. Therefore, it is essential to ensure a low relative speed between the satellite and the camera to accurately capture a clear image of the target. If the previous step yields a high degree of reliability and accurately locates the outline of the square target, the relative position and orientation can be determined using OpenCV's PnP pose calculation method for the target.

[0088] Binocular pose estimation methods can use 2D binocular images for stereo matching to reconstruct the position of points in the 3D world. This method has extremely high robustness, meaning it can obtain relatively accurate position information even when the relative velocity between the satellite and the camera is high, but it cannot obtain the satellite's attitude information. When the reliability of the previous step's estimation is low but the target is missing, satellite localization can be performed based on the target's center pixel in the left and right eye images.

[0089] Specifically, by using the position of the target pixel in the stereo image, a mapping relationship between the two-dimensional stereo image and the image coordinates in the display world can be constructed:

[0090]

[0091] in , , and , , These represent the parameters in the imaging models of the left and right cameras, respectively. , , , This indicates the internal parameters of the camera. , This represents the extrinsic parameters of the camera. All parameters in the above formula can be obtained through the calibration of a stereo camera, and their real-world coordinates can be solved by expanding the system of equations.

[0092] In summary, while the target-based pose calculation method presented in this paper can calculate relatively accurate pose information even when the target is moving at low speed, the binocular matching-based position calculation method can provide a more robust position calculation method when the target is tracked. Therefore, a judgment threshold is defined to determine when to use pose calculation and when to use position calculation. When the confidence level of the target tracking and identification in the current frame If the value is less than this threshold, the acquired target is considered to have a high degree of blur, and a binocular position calculation strategy is adopted; otherwise, a target-based pose calculation strategy is adopted. The specific judgment is shown in the following formula:

[0093]

[0094] In the formula This represents the pose corresponding to m marked at time t, when the calculated target confidence level... Below the threshold In the first case, a more robust binocular position calculation method is used; otherwise, a more accurate target-based pose calculation strategy is employed. Finally, the position and attitude information for the perturbed sub-star is obtained. The pose calculation results for the perturbed sub-star are as follows: Figure 4 As shown.

[0095] The technical solutions of the embodiments of the present invention have the following beneficial effects:

[0096] The technical solution of this invention proposes a method for tracking and locating perturbed satellite targets by fusing binocular data. First, a target tracker is initialized based on the perturbed satellite targets acquired by the camera. A set of filters is used to track the satellite targets, and tracking is then performed in parallel within the image, with the target position updated cyclically based on the responsivity. A target tracking reliability index is constructed to characterize the stability of the tracking and evaluate the reliability of the tracked target image. Finally, a pose calculation method is selected based on the reliability of the satellite tracking and the target tracking information. If the reliability is high, the target pose calculation method is selected; if the reliability is low, the binocular position calculation method is selected, achieving a perturbed satellite target localization that balances accuracy and robustness.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0098] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A method for tracking and locating a perturbed satellite target by fusing binocular information, characterized in that, The method includes: S1 selects the target ID and target size parameters. First, target features are extracted based on the image captured by the camera to obtain the information of the four corner points of the target and match the target ID information. S2 initializes the target tracker based on the corner information and target ID information, and generates a target tracker group, that is, a group of trackers is used to form a target tracker; S3 uses an initialized target tracker to track changes in the target position in a real-time image stream and calculates the target response map; S4 determines whether the target tracker has lost tracking based on the response graph; If the target tracker is lost in S5, then steps S1 and S2 are executed to reinitialize the target tracker; if the target tracker is lost, then the position of the target tracker is updated in real time, and the trustworthiness of the tracker is calculated based on the position update of the target tracker. S6 selects the final pose calculation strategy based on the trust level. If the trust level is higher than the set threshold, the pose calculation method based on the target is selected. If the trust level is lower than the threshold, the position calculation method based on binocular stereo matching is selected to calculate the relative position or pose information of the target. S7 outputs the pose information of the satellite target based on the trust level and the solution results, thus completing the target localization of the perturbed sub-satellite.

2. The method according to claim 1, characterized in that, Step S1 includes: S1.1 Select the target's ID information and target size as known target parameters; S1.2 Images are captured using a camera and recorded as images. Image matching and judgment are performed based on predefined target ID information and the identified target ID information. Which image patches in the image are target objects? The extracted target objects are denoted as... ; S1.3 Targeting the detected target objects Extract the feature information corresponding to the target, including the target's coordinates at the center of the image. and the pixel coordinates of the four corner points , where c represents the image The pixel coordinates of the four corners of the marker m.

3. The method according to claim 2, characterized in that, Step S2 includes: S2.1 Setting Function Identifying images China and Israel Centered on, size is Image blocks; S2.2 is based on the center coordinates described in S1.

3. and the pixel coordinates of the four corner points Generate five feature image patches and The image patch represents the side length of the designed KCF tracker. Tracking images; S2.3 Based on the image patch described in S2.1, a global target tracker is generated using the KCF filtering algorithm. and four target corner trackers ; S2.4, based on the target global tracker and four target corner trackers described in S2.3, is as follows: in This indicates the KCF trackers at the center and four corners; and records the pyramid level corresponding to the initialization of the trackers at this time. , representing the pyramid level related to target m at time t, the pyramid level Characterization parameters used to represent scale changes as an image moves closer to or further away.

4. The method according to claim 3, characterized in that, Step S3 includes: S3.1 Using a target tracker group, the response is calculated in real time within the image stream and represented as a response map. To accelerate the response calculation, a multi-scale parallel computing tracker group is used, as described in the pyramid hierarchy in step S2.

4. and the scale of the two adjacent pyramids , The response details include: Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ; Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ; Using the aforementioned target tracker set, based on pyramid levels Frequency domain correlation operations are performed on the real-time image to obtain the response map. ; in The asterisk (*) indicates element-wise multiplication, and the asterisk (*) indicates complex conjugation. This indicates that all sides of the image frame at the current pyramid level are of length 1. A collection of patch blocks; S3.2 Based on the response maps calculated from each pyramid level described in S3.1, the response map of the current image is calculated by weighting the responses according to their respective weights. in It is a scale Weight on, Representing scale The response plot at time t is the final result. This is the response image from this detection; S3.3 is based on the pyramid levels described in S3.

1. and , The response data is compared, and the pyramid level of the maximum response in the response map is updated to the new pyramid level recorded in the current tracking image. .

5. The method according to claim 4, characterized in that, Step S4 includes: S4.1 Design Function Used to calculate tracker groups The maximum response value corresponding to position p in the response graph, in this formula This represents the filter centered at pixel p in image I. The response; S4.2 Based on the response graph The maximum response value calculation function is used to obtain the tracker corresponding to the center coordinate in the tracker group. Maximum response value Define the tracker loss threshold as If the maximum response value is lower than the loss threshold, the tag tracking is considered lost.

6. The method according to claim 5, characterized in that, Step S5 includes: S5.1 is based on the judgment benchmark of S4.2, if the maximum response Below the loss threshold If the target tracker group is lost, it is considered to be lost and the target tracker group needs to be re-initialized and steps S1 and S2 need to be re-executed to re-track. S5.2 is based on the S4.2 judgment benchmark; if the maximum response... Above the loss threshold If the target is successfully tracked, the position of the new target in the image is updated to the position p of the maximum response of the tracker group in the image. The reliability of the tracking is calculated based on the position p. The reliability reflects the reliability of the tracking. The reliability index mainly consists of two parts.

7. The method according to claim 6, characterized in that, Step S6 includes: Based on the trustworthiness described in S6, a trustworthiness threshold is established. As a criterion, when the confidence level is less than this threshold, the target image is considered to be highly blurred, and a binocular position calculation method is used to calculate the position information; when the confidence level is greater than this threshold, the target image is considered to be highly sharp, and a target pose calculation method is used to output the target pose information. The specific formula for judgment is as follows: In the formula This represents the pose corresponding to m marked at time t, and the calculated target confidence level. Below the threshold When the problem is not in a state of flux, a more robust binocular position calculation method is used; otherwise, a more accurate target-based pose calculation strategy is used. Binocular position calculation method: When the current target tracker has low reliability, that is, when the target moves fast and the image becomes blurry, and a clear target image cannot be obtained, the center pixel of the target is obtained through the target tracker. Based on the pixel position in the binocular camera, the actual position information of the target is obtained by combining binocular stereo matching. This method has good robustness. Pose calculation strategy: When the current target tracker has high confidence, it means that the target motion is uniform and the target image is clear. Obtain a target image with high clarity, and use the target pose calculation method based on OpenCV to calculate accurate target pose information based on the image.

8. The method according to claim 7, characterized in that, Step S7 includes: Based on pose information and the corresponding level of trust. Complete the localization of the perturbed sub-star target, output the calculated target pose information, and repeatedly execute steps S5~S7 to perform continuous target localization operations to achieve continuous tracking and pose calculation of the perturbed sub-star target.

Images