A method and system for long-distance target recognition and azimuth perception based on optical markers

Abstract

The present invention discloses a method and system for long-distance target recognition and azimuth perception based on optical markers. The method includes: preprocessing a pre-acquired image and extracting target features; using an edge detection function to extract the target contour, and further excluding non-detected targets according to the geometric features of the target contour to obtain a preliminary contour of the detected target; fitting the pixel coordinates of the target centroid point in the image; tracking all targets to obtain the on / off state of the optical markers; calculating the on / off time of the lights and writing the corresponding flag bits into the light sequence; decoding the light sequence to match the target identity ID; solving the relative angle according to the camera internal parameters and the feature center position, and matching the relative angle of the target and the target identity ID to output the angle information of all targets in the image and their corresponding identity IDs. The present invention realizes the recognition of target identity IDs in the case of multiple targets, and real-time outputs the relative angle information between the target and the shooting point and its corresponding ID information.

Description

Technical Field

[0001] The present invention relates to the fields of target recognition and relative navigation, and particularly to a long-distance target recognition and orientation perception method and system based on optical markers. Background Art

[0002] High-precision navigation information is one of the keys to completing various long-distance target relative navigation tasks, and there are a large number of long-distance navigation task requirements in both civilian and military fields. Traditional relative navigation methods are mainly divided into two types: relative navigation methods based on absolute position difference and relative measurement sensors. The former mainly uses the method of inertial navigation / satellite integrated navigation to provide high-precision navigation information. However, the positioning accuracy will drop significantly when satellite signals are blocked, and it is also vulnerable to human interference in the complex wartime environment, and in severe cases, it will directly lead to the collapse of the entire navigation system. The latter mainly uses relative measurement sensors to realize the perception of targets in the environment. Commonly used measurement sensors include optoelectronic sensors that measure the angle of arrival, and lidar that has both relative ranging and angle measurement capabilities. Optoelectronic sensors have a fast response speed, good reliability, and high precision, but they are vulnerable to environmental interference and have certain limitations; the advantage of lidar lies in its powerful ranging and angle measurement capabilities, which can provide high-quality spatial information, but it is relatively expensive and not very flexible. In recent years, machine vision technology has developed rapidly. Due to the characteristics of flexibility, economy, non-contact, etc., visual sensing devices have been widely used in fields such as robots, autonomous driving, security, and intelligent wearable devices.

[0003] Since the visual measurement accuracy is closely related to the relative distance, the research of domestic and foreign scholars on visual measurement technology mainly focuses on the relative pose measurement in the near-distance case. In this range, the imaging area of the object is relatively large, and its visual characteristics can be well presented in the image. For example, in the aviation field, the pose of the refueling drogue is calculated using active or passive visual measurement methods to achieve the close-distance relative navigation task during the in-air refueling docking stage; in the industrial field, high-precision part size measurement and product quality control are realized using the image information of the target to be detected, with high stability and measurement accuracy; in the field of autonomous driving, visual three-dimensional measurement technology is used to achieve functions such as vehicle path planning and safe obstacle avoidance, effectively improving the reliability of autonomous driving technology.

[0004] However, the area occupied by a distant target in the camera image is very small, losing a large amount of texture and color features, and almost only having blurred contour information. The pose measurement method for close-range targets cannot be applied to distant targets, and monocular vision cannot obtain the depth information of the target. At present, ranging technologies such as UWB are relatively mature, which can achieve centimeter-level relative distance measurement within the communication range, with the advantages of low cost and high precision, and are widely used in the field of relative navigation. If the angle information of the target can be obtained and its identity can be recognized, the relative navigation information of the target can be restored by combining the distance information sensed by the ranging device. Therefore, the research on technologies such as long-distance high-precision relative angle measurement based on vision and identity recognition of multiple targets is particularly urgent. Summary of the Invention

[0005] Object of the Invention: The present invention provides a method and system for long-distance target recognition and orientation perception based on optical markers, which can realize the identification of the target identity ID in the case of multiple targets, and output the relative angle information between the target and the shooting point and its corresponding ID information in real time.

[0006] Technical Solution: A method for long-distance target recognition and orientation perception based on optical markers according to the present invention specifically includes the following steps:

[0007] (1) Preprocess the pre-acquired image and extract target features;

[0008] (2) Use the edge detection function to extract the target contour, exclude non-detected targets according to the geometric features of the target contour, and obtain the primary contour of the detected target;

[0009] (3) Use the centroid method to fit the pixel coordinates of the target centroid point in the image for the primary contour of the target;

[0010] (4) Determine whether it is a new target according to the position of the centroid point; detect the on / off state of the optical marker within the tracking frame of each frame of image;

[0011] (5) Calculate the on / off time of the light and write the corresponding flag bit into the light sequence;

[0012] (6) Decode the light sequence of each target, match and recognize it with the pre-set sequence, and obtain the target identity ID;

[0013] (7) Use the camera imaging model to calculate the azimuth angle and elevation angle of the target relative to the camera optical axis according to the target image coordinates;

[0014] (8) Match the relative angle of the target and the target identity ID, and output the angle information of all targets in the image and their corresponding identity IDs.

[0015] Furthermore, the implementation process of the step (1) is as follows:

[0016] Preprocess the collected images, including grayscale conversion and binarization operations. Calculate the threshold using a method that combines an adaptive threshold and a fixed threshold, and segment and extract all possible target points in the images.

[0017] Further, the implementation process of step (2) is as follows:

[0018] (a) Calculate the length h and width w of the minimum bounding rectangle of the target contour. The aspect ratio of the minimum bounding rectangle needs to satisfy the following conditions:

[0019]

[0020] In the formula, k d is a limiting parameter, and its value range is 0.5 < k d < 1;

[0021] (b) Calculate the area S of the target contour. The area of the target contour needs to satisfy the following conditions:

[0022] k 11 < S < k 12

[0023] In the formula, k 11 and k 12 are limiting parameters, which are set according to the specific size of the optical mark;

[0024] If the parameters related to the extracted target contour satisfy (a) and (b), then determine that the contour is the primary contour of the target to be detected.

[0025] Further, the implementation process of step (3) is as follows:

[0026] Regard the pixels of the image as the density function V(i, j). V(i, j) is the pixel value of the point (i, j) in the image. The coordinates C of the center of gravity of the obtained contour are expressed as:

[0027]

[0028] In the formula, m 00 is the zero-order moment of the image, m 10 and m 01 are the first-order moments of the image,

[0029] Further, the implementation process of step (4) is as follows:

[0030] Judge whether it is a new target according to the centroid position. If so, add it as an initial value to the target tracker; if not, update the position of the tracked target, and detect the on / off state of the optical mark within the tracking box of each frame of the image.

[0031] Furthermore, the new target determination method is as follows: during tracking initialization, a rectangular box that moves following the target point is generated based on the detected marker point coordinates; in subsequent frame image detections, first detect the centroid positions of all target points in the image, and determine one by one whether they are included in the already generated rectangular box. If not, the target is identified as a new target.

[0032] Furthermore, the implementation process of step (5) is as follows:

[0033] First, calculate the constant-on and off times of the optical marker based on the on-off states of the optical markers in consecutive frame images. Start calculating when the state of the optical marker in the tracking box changes, and stop until the state of the marker in the tracking box changes again; the constant-on durations of different lights correspond to flag bits "1", "2", "3"... respectively, and the flag bit for the off state is "0". Write the flag bits into the light sequence of the target according to the calculated on-off times of the marker; when the elapsed time of detecting the off state of the light exceeds the set threshold, it is determined that the tracking of the target fails, and the target is deleted from the tracker.

[0034] Furthermore, the implementation process of step (6) is as follows:

[0035] Regard one constant-on and one off of the light as one flash of the light. Design the light sequence using the method of single and combined flash speeds to meet the recognition requirements for multiple targets; match the obtained light sequence with the set sequence. After successful matching, enter the target confirmation stage. Judge the confidence level of the target identity ID according to the determination results in subsequent cycles. If two consecutive matches are for a new identity ID, modify the result to obtain the target identity ID.

[0036] Furthermore, the implementation process of step (7) is as follows:

[0037]

[0038] In the formula, α and β are the azimuth angle and elevation angle respectively, (x c , y c , z c ) are the coordinates of the target in the camera coordinate system, (u, v) are the coordinates of the target in the image coordinate system, f x , f y are the focal lengths in the x and y directions respectively, and (u0, v0) is the actual principal point coordinate.

[0039] A long-distance target recognition and orientation perception system based on optical markers according to the present invention includes an image acquisition module, an image preprocessing module, a feature preprocessing module, a multi-target tracking module, an identity decoding module, an orientation perception module, and an information matching module; where:

[0040] An image acquisition module drives a camera to collect dynamic image information containing ultraviolet feature markers in real time and transmits it frame by frame to an image preprocessing module;

[0041] An image preprocessing module receives the image information from the image acquisition module, preprocesses the image, separates the feature markers from the image, and transmits the extraction results of the feature markers to a feature preprocessing module;

[0042] A feature preprocessing module uses the geometric information of the marker contour as a constraint, filters out the primary contours of the detection targets, further fits the feature centers, and transmits the pixel coordinates of the feature centers to a multi-target tracking module;

[0043] A multi-target tracking module tracks the first detected marker coordinates as initial values respectively, records the on / off states of the optical markers in the image and transmits them to an identity decoding module, and transmits the position information of the feature markers in the image to an azimuth perception module;

[0044] An identity decoding module calculates the lighting on / off duration according to the states of the optical markers in consecutive frames, writes the corresponding flag bits into the lighting sequence, decodes the lighting sequence to match the target identity ID, and transmits the ID information of all targets to an information matching module;

[0045] An azimuth perception module uses the pixel coordinates of the target feature center in the image, combines the pre-calibrated camera parameters, solves the direction angle and elevation angle of the target, and transmits them to the information matching module;

[0046] An information matching module receives the relative angle information from the azimuth perception module and the target identity ID information from the identity decoding module, matches the two pieces of information, and finally outputs the angle information of all targets in the image and their corresponding identity IDs.

[0047] Advantageous effects: Compared with the prior art, the advantageous effects of the present invention are as follows:

[0048] 1. The present invention can complete the angle measurement task of multiple targets at a long distance by using a low-cost camera and an ultraviolet light source; compared with the existing technology, the method for visual measurement of angles and identification of identity IDs proposed by the present invention does not rely on data link communication, has strong anti-interference ability and high economic benefits, and can be combined with the existing ranging technology to provide high-precision relative navigation information;

[0049] 2. The present invention uses the method assisted by optical markers, accurately extracts the centroid coordinates of the targets through technologies such as image segmentation and feature screening, solves the problem of difficult feature extraction of small targets at a long distance, can realize the relative angle measurement of targets at a distance of one hundred meters, and the static angle measurement error is better than 0.15° at a distance of 50 - 150m, and the dynamic angle measurement error is better than 0.22°;

[0050] 3. The present invention encodes the target identity ID by using the stroboscopic characteristics of light, and can realize the decoding and recognition of the identity IDs of multiple targets within the field of view by analyzing the status information of optical markers in consecutive frame images, effectively solving the problem of difficult recognition due to the similar appearance between multiple targets, and can meet the identity recognition requirements of more than ten targets. Description of the Drawings

[0051] Figure 1 is a flowchart of a long-distance target recognition and azimuth perception method based on optical markers;

[0052] Figure 2 is a framework diagram of a long-distance target recognition and azimuth perception system based on optical markers;

[0053] Figure 3 is a schematic diagram of the hardware solution of a long-distance target recognition and azimuth perception system;

[0054] Figure 4 is a schematic diagram of the identity ID recognition results of different frame images of the captured video; among them, (a) is a schematic diagram of the identity ID recognition result of the 450th frame image of the captured video; (b) is a schematic diagram of the identity ID recognition result of the 1960th frame image of the captured video. Detailed Embodiment

[0055] The present invention will be further described in detail below with reference to the drawings.

[0056] As Figure 1 shown, the present invention proposes a long-distance target recognition and azimuth perception method based on optical markers, which can use visual measurement methods in a satellite-denied environment to output the relative angle information of multiple targets within the field of view and their matching identity IDs in real time; the specific process is as follows:

[0057] Step 1: Preprocess the image collected by the camera to extract target features.

[0058] After converting the collected image into a grayscale image, the image is processed by using the method of threshold binary segmentation to separate the feature target area from the background and obtain the feature targets in the image. The selection of the binary threshold adopts a method combining the adaptive threshold method and the global threshold method. The adaptive threshold method is used to detect the feature points brighter than the surrounding environment in the image, and the global threshold method is used to eliminate the interference in the environment and segment and extract all possible target points in the image.

[0059] Step 2: Screen the target points according to the geometric information of the target features, use the edge detection function to extract all the target contours in the binary image, and then assist in excluding other environmental interferences, such as the sun, reflective objects, etc., according to the aspect ratio of the minimum circumscribed rectangle of the target contour and the area size of the contour, and obtain the preliminary contour of the target to be detected.

[0060] Condition (a): Calculate the length h and width w of the minimum bounding rectangle of the target contour. The aspect ratio of the minimum bounding rectangle shall meet the following conditions:

[0061]

[0062] In the formula, k d is a limiting parameter, and its value range is 0.5 < k d < 1.

[0063] Condition (b): Calculate the area S of the target contour. The area of the target contour shall meet the following conditions:

[0064] k 11 < S < k 12

[0065] In the formula, k 11 and k 12 are limiting parameters, which are set according to the specific size of the optical mark.

[0066] If the parameters related to the target contour extracted meet Condition (a) and Condition (b), then determine that the contour is the primary contour of the target to be detected.

[0067] Step 3: After obtaining the primary contour of the detection target, use the extracted contour to fit the centroid point of the feature mark, and obtain the pixel coordinates of the target centroid point in the image.

[0068] Use the centroid method to obtain the centroid coordinates of the marked points, mainly by calculating using the moments of the image. Consider the pixels of the image as the density function V(i, j), where V(i, j) is the pixel value of the point (i, j) in the image. The obtained centroid coordinates C of the contour are expressed as:

[0069]

[0070] In the formula, m 00 is the zero-order moment of the image, m 10 and m 01 are the first-order moments of the image,

[0071] Step 4: Determine whether it is a new target according to the position of the centroid point; detect the on / off state of the optical mark within the tracking frame of each frame of the image.

[0072] Track all targets in each frame of the image. If it is the first frame of the image currently, set the target pixel coordinates obtained in step 3 as the initial values for tracking, and add them to the target tracker. The tracker contains information such as target pixel coordinates, tracking box positions, and target on / off states. If it is not the first frame, determine whether there are new targets based on the target coordinates, add the new targets to the tracker, and update the tracking information for the old targets. At a long distance, the displacement of the target between two adjacent frames is small. Directly detect the optical marker state within the tracking box area of the previous frame image to obtain the target optical marker state information.

[0073] During tracking initialization, a rectangular box that follows the target point will be generated based on the detected marker point coordinates. In the subsequent frame image detection, first detect the centroid positions of all target points in the image, and judge one by one whether they are included in the already generated rectangular box. If not, the target is considered a new target and added to the tracking.

[0074] Step 5, calculate the on / off time of the light and write the corresponding flag bits into the light sequence.

[0075] First, calculate the on and off times of the optical marker based on the on / off states of the optical marker in consecutive frame images. Start calculating when the optical marker state within the tracking box changes until the marker state within the tracking box changes again and stops. The on durations of different lights correspond to the flag bits "1", "2", "3"... respectively, and the flag bit for the off state is "0". Write the flag bits into the light sequence of the target according to the calculated on / off times of the marker. When the elapsed time for detecting the off state of the light exceeds the set threshold, it is determined that the target tracking fails, and the target is deleted from the tracker.

[0076] Before the optical marker is detected, the default initial state is off. So when it is detected for the first time and the state changes to on, start calculating the on time. When the previous frame light state is off and the current frame state is on, record the current time t1 and start calculating the on time. When the light state changes to off, record the current time t2. Then the on time of the light ton = t2 - t1. At this time, t2 is the starting point for calculating the off time of the light. When the light state changes back to on, record the current time to t1. Then the off time of the light toff = t1 - t2; and so on in a loop.

[0077] Step 6, decode the light sequence to match the target identity ID.

[0078] Match the light sequence counted in step 5 with the pre-set sequence to identify the target identity ID. Before the light sequence matching is successful, continuously monitor and match the sequence; after the matching is successful, enter the target confirmation stage, set the confidence level of the recognition result to the initial value, determine whether it is the sequence of the same ID in the new sequence cycle, and improve the confidence level of the target recognition result according to the matching results in subsequent cycles. If the continuous matching is a new recognition result, correct the target identity ID and re-enter the target confirmation stage. In this step, the influence of constantly lit non-detection targets in the environment can also be excluded.

[0079] The present invention regards one constant light and one extinguish of the light as one flash of the light, and adopts the method of single and combined flash speeds to design the light sequence to meet the recognition requirements of multiple targets. For example, set three flash speeds, and four flashes of the light are taken as one cycle. Then, a single light sequence within one cycle can be expressed as "10101010", and the combined light sequence can be expressed as "10202010". To prevent sequence matching errors caused by false detection or simple interference, high similarity should be avoided as much as possible in the setting of the light sequence.

[0080] Step 7: Solve the relative angle according to the position of the feature center in the camera.

[0081] Using the camera imaging model, calculate the direction angle and elevation angle of the target relative to the camera optical axis according to the target image coordinates. The calculation formula is as shown in the following formula:

[0082]

[0083] In the formula, α and β are the direction angle and elevation angle respectively, (x c , y c , z c ) is the coordinate of the target in the camera coordinate system, (u, v) is the coordinate of the target in the image coordinate system, f x , f y are the focal lengths in the x and y directions respectively, and (u0, v0) is the actual principal point coordinate.

[0084] Step 8: Output the relative angles of all targets and their matching target identity IDs.

[0085] For the targets whose identity IDs have been successfully matched, the navigation computer outputs their corresponding relative angle information and identity IDs, and at the same time outputs the confidence level of the identity ID recognition according to the confirmation result in the target confirmation stage. For the remaining targets that are difficult to identify due to long distance or have not been successfully matched yet, first output and save their measured relative angle information.

[0086] Such as Figure 2As shown in the figure, the present invention also provides a long-distance target recognition and orientation perception system based on optical markers, including an image acquisition module, an image preprocessing module, a feature preprocessing module, a multi-target tracking module, an identity decoding module, an orientation perception module, and an information matching module; where:

[0087] The image acquisition module drives the camera to collect dynamic image information containing ultraviolet feature markers in real time and transmits it frame by frame to the image preprocessing module.

[0088] The image preprocessing module receives the image information from the image acquisition module, separates the feature markers from the image after preprocessing the image, and transmits the extraction result of the feature markers to the feature preprocessing module.

[0089] The feature preprocessing module receives the feature marker information from the image preprocessing module, uses the geometric information of the marker contour as a constraint, filters out the qualified targets, and then further fits the feature center, and transmits the pixel coordinates of the feature center to the multi-target tracking module.

[0090] The multi-target tracking module receives the target feature information from the feature preprocessing module and tracks them respectively with the coordinates of the first detected marker as the initial value. Records the on / off state of the optical marker in the image and transmits it to the identity decoding module, and transmits the position information of the feature marker in the image to the orientation perception module.

[0091] The identity decoding module receives the state information of the optical marker from the multi-target tracking module, calculates the on / off duration of the light according to the state of the optical marker in consecutive frames, writes the corresponding flag bit into the light sequence, decodes the light sequence to match the target identity ID, and transmits the ID information of all targets to the information matching module.

[0092] The orientation perception module receives the position information of the optical marker from the multi-target tracking module, uses the pixel coordinates of the target feature center in the image, combines the pre-calibrated camera parameters, solves the azimuth angle and elevation angle of the target, and transmits them to the information matching module.

[0093] The information matching module receives the relative angle information from the orientation perception module and the target identity ID information from the identity decoding module, matches the two pieces of information, and finally outputs the angle information of all targets in the image and their corresponding identity IDs.

[0094] As Figure 3As shown in the figure, it is a hardware schematic diagram of the present invention, including parts such as an optical camera and a navigation computer of the vision measurement module, as well as an optical marking module arranged on the target, which includes an ultraviolet mark, a control switch, and a regulated power supply. The optical marking module uses an ultraviolet lamp strip with a wavelength of 395 nm as the target feature mark, designs the encoding of the ultraviolet light using the control switch, and the regulated power supply is used to supply power to the ultraviolet lamp strip. The vision measurement module installs an ultraviolet filter on the optical camera to achieve the effect of an ultraviolet camera. The camera is connected to the navigation computer and sends the collected image information containing the ultraviolet mark to the navigation computer for processing.

[0095] To verify the relative angle measurement accuracy of the method proposed by the present invention for the target, a set of ground vision angle measurement accuracy evaluation experimental platforms is constructed using equipment such as total stations to test the indicators of the system. Using the high-precision positioning information provided by the total station and comparing it with the angle information perceived by the vision camera, the angle measurement accuracy of the system is quantitatively analyzed. The specific experimental steps are as follows:

[0096] 1) Calibrate the conversion matrix between the total station coordinate system and the camera coordinate system. The camera and the total station are placed together, and the total station prism is fixed to the checkerboard for calibration. Adjust the position of the checkerboard in three-dimensional space, including changes in three directions, and record the captured images and total station information.

[0097] 2) Replace the checkerboard with an ultraviolet feature mark and install a 395 nm filter on the camera. Move the position of the ultraviolet mark and collect images at distances of 50 m, 100 m, and 150 m from the camera respectively. The total station records the three-dimensional position, and 10 groups of data are collected at each position.

[0098] 3) Similar to the static experiment, use a trolley to carry the feature mark board to simulate the low-speed movement of the target, collect an image sequence during the movement, and the total station automatically tracks and records the position. A total of three groups of experiments are carried out to test the dynamic angle measurement accuracy at 50 m, 100 m, and 150 m respectively.

[0099] The static experiment results obtained according to the above steps are shown in Table 1, and the dynamic experiment results are shown in Table 2.

[0100] Table 1 Statistical table of static multi-position vision relative angle measurement errors

[0101] Distance (m) Azimuth angle mean square error (°) Elevation angle mean square error (°) Relative angle mean square error (°) 50 0.1328 0.0623 0.1467 100 0.1105 0.0363 0.1163 150 0.1173 0.0409 0.1263

[0102] Table 2 Statistical table of dynamic experiment vision relative angle measurement errors

[0103] Distance (m) Azimuth angle mean square error (°) Elevation angle mean square error (°) Relative angle mean square error (°) 50 0.1833 0.1168 0.2174 100 0.1222 0.1012 0.1587 150 0.1286 0.0867 0.1550

[0104] The test results show that the relative angle measurement error of the target is better than 0.15° in the static case and better than 0.22° in the dynamic case. The method proposed in the present invention has relatively good measurement accuracy at long distances because the selection of the centroid coordinates of the target has a great influence on the measurement accuracy. When the feature points can be detected, the farther the target is from the camera, the smaller the proportion of the feature points in the image, the more accurate the position of the extracted centroid point, and the more accurate the calculated relative angle.

[0105] To verify the effect of the method proposed in the present invention on the identification of the target identity ID, an identification experiment for multiple targets was further carried out. The specific experimental steps are as follows:

[0106] 1) Set ten targets to be detected, install ultraviolet feature marks on the targets, preset the light sequence corresponding to the identity ID, and connect the switch to control the lights to flash in the set sequence.

[0107] 2) Arrange the targets to be detected evenly on the open ground, set two moving targets, and the rest are static targets. Place the camera about 100 meters away from the targets and shoot and identify the identity IDs of the targets within the field of view.

[0108] The experimental effect obtained according to the above experimental steps is as Figure 4 shown. Among them, (a) is a schematic diagram of the identity ID recognition result of the 450th frame image of the captured video; (b) is a schematic diagram of the identity ID recognition result of the 1960th frame image of the captured video. From the effect diagram, the ten targets in the image can all be recognized with the corresponding identity ID. It can be concluded that the method of encoding the target identity ID by using three combinations of flashing speeds in the present invention can meet the identification requirements of ten target identity IDs at a distance of 100 meters.

[0109] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for long-distance target recognition and azimuth perception based on optical markers, characterized in that, It includes the following steps: (1) Preprocess the pre-acquired image and extract target features; (2) Use an edge detection function to extract the target contour, exclude non-detection targets according to the geometric features of the target contour, and obtain the primary contour of the detection target; (3) Use the centroid method to fit the pixel coordinates of the target centroid point in the image for the primary contour of the target; (4) Judge whether it is a new target according to the centroid point position; detect the on / off state of the optical mark within the tracking frame of each frame of image; (5) Calculate the on / off time of the light and write the corresponding flag bit into the light sequence; (6) Decode the light sequence of each target, match and identify it with the pre-set sequence, and obtain the target identity ID; (7) Use the camera imaging model to calculate the azimuth angle and elevation angle of the target relative to the camera optical axis according to the target image coordinates; (8) Match the relative angle of the target and the target identity ID, and output the angle information of all targets in the image and their corresponding identity IDs.

2. The method for remote target recognition and azimuth perception based on optical markers according to claim 1, wherein The implementation process of step (1) is as follows: Preprocess the collected image, including grayscale conversion and binarization operations, calculate the threshold using a method that combines adaptive threshold and fixed threshold, and segment and extract all possible target points in the image.

3. A method for long-distance target recognition and azimuth perception based on optical markers according to claim 1, characterized in that The implementation process of step (2) is as follows: (a) Calculate the length h and width w of the minimum bounding rectangle of the target contour. The aspect ratio of the minimum bounding rectangle needs to meet the following conditions: where k d is a limiting parameter, and its value range is 0.5 < k d < 1; (b) Calculate the area S of the target contour. The area of the target contour needs to meet the following conditions: k 11 <S<k 12 where k 11 and k 12 are limiting parameters, which are set according to the specific size of the optical mark; If the relevant parameters of the extracted target contour meet (a) and (b), then determine that this contour is the primary contour of the target to be detected.

4. A method for long-distance target recognition and azimuth perception based on optical markers according to claim 1, characterized in that The implementation process of step (3) is as follows: Regard the pixels of the image as the density function V(i, j). V(i, j) is the pixel value of the point (i, j) in the image. The calculated centroid coordinates C of the contour are expressed as: where m 00 is the zero-order moment of the image, m 10 and m 01 are the first-order moments of the image, 5. A method for remote target recognition and azimuth perception based on optical markers according to claim 1, characterized in that The implementation process of step (4) is as follows: Judge whether it is a new target according to the centroid position. If so, add it as an initial value to the target tracker; if not, update the position of the tracked target and detect the on / off state of the optical mark within the tracking frame of each frame of image.

6. The method for remote target recognition and orientation perception based on optical markers according to claim 5, wherein The method for judging the new target is as follows: At the time of tracking initialization, generate a rectangular frame that follows the movement of the target point according to the detected marker point coordinates; in the subsequent frame image detection, first detect the centroid positions of all target points in the image, and judge one by one whether they are included in the already generated rectangular frame. If not, determine that this target is a new target.

7. A method for long-distance target recognition and azimuth perception based on optical markers according to claim 1, characterized in that, The implementation process of step (5) is as follows: First, calculate the on and off times of the optical mark according to the on / off state of the optical mark in consecutive frame images. Start calculating when the state of the optical mark within the tracking frame changes and stop until the state of the mark within the tracking frame changes again; the on durations of different lights respectively correspond to the flag bits "1", "2", "3"... The flag bit for the off state is "0". Write the flag bit into the light sequence of the target according to the calculated on / off time of the mark; when the elapsed time of detecting the off state of the light exceeds the set threshold, it is determined that the target tracking fails, and the target is deleted from the tracker.

8. A method for long-distance target recognition and azimuth perception based on optical markers according to claim 1, characterized in that The implementation process of step (6) is as follows: Regarding one constant light-on and one light-off of the light as one flash of the light, a method of single and combined flash speeds is used to design the light sequence to meet the recognition requirements of multiple targets; the obtained light sequence is matched with the set sequence, and after successful matching, it enters the target confirmation stage. The confidence level of the target identity ID is judged according to the judgment results in subsequent cycles. If the matching is continuously twice for a new identity ID, the result is modified to obtain the target identity ID.

9. A method for long-distance target recognition and azimuth perception based on optical markers according to claim 1, characterized in that The implementation process of step (7) is as follows: Where α and β are the azimuth angle and elevation angle respectively, (x c , y c , z c ) are the coordinates of the target in the camera coordinate system, (u, v) are the coordinates of the target in the image coordinate system, f x , f y are the focal lengths in the x and y directions respectively, and (u0, v0) are the actual principal point coordinates.

10. An optical marker-based long-distance target recognition and orientation perception system using the method according to claims 1 to 9, characterized in that, It includes an image acquisition module, an image preprocessing module, a feature preprocessing module, a multi-target tracking module, an identity decoding module, an azimuth perception module, and an information matching module; among them: The image acquisition module drives the camera to collect dynamic image information containing ultraviolet feature markers in real time and transmits it frame by frame to the image preprocessing module; The image preprocessing module receives the image information from the image acquisition module, separates the feature markers from the image after preprocessing the image, and transmits the extraction result of the feature markers to the feature preprocessing module; The feature preprocessing module uses the geometric information of the marker contour as a constraint, filters out the primary contour of the detected target and then further fits the feature center, and transmits the pixel coordinates of the feature center to the multi-target tracking module; The multi-target tracking module uses the coordinates of the markers detected for the first time as the initial values for tracking respectively, records the on-off state of the optical markers in the image and transmits it to the identity decoding module, and transmits the position information of the feature markers in the image to the azimuth perception module; The identity decoding module calculates the light-on and light-off duration according to the state of the optical markers in consecutive frames and writes the corresponding flag bits into the light sequence, decodes the light sequence to match the target identity ID, and transmits the ID information of all targets to the information matching module; The azimuth perception module uses the pixel coordinates of the target feature center in the image, combines the pre-calibrated camera parameters, solves the azimuth angle and elevation angle of the target and transmits them to the information matching module; The information matching module receives the relative angle information from the azimuth perception module and the target identity ID information from the identity decoding module, matches the two pieces of information, and finally outputs the angle information of all targets in the image and their corresponding identity IDs.

Images