Distance measuring method and device based on binocular vision and laser, and intelligent wearable equipment

Abstract

The invention discloses a distance measuring method and device based on binocular vision and laser, and intelligent wearable equipment. The method comprises the following steps: collecting a first image and a second image of a target object at different visual angles through a binocular vision device of the intelligent wearable equipment; extracting, matching and screening feature points of a target object based on the first image and the second image to obtain key point pairs, and calculating parallax according to the key point pairs; calculating a depth value of the intelligent wearable device and the target object according to the parallax; laser ranging values of the intelligent wearable device and the target object are collected through a laser device of the intelligent wearable device; fusing the depth value and the laser ranging value by adopting a Kalman filtering algorithm, and outputting a fused ranging result; the method solves the problems that a distance measurement scheme in the prior art is complex in algorithm and high in cost.

Description

Technical Field

[0001] The present invention relates to the field of distance measurement technology, and in particular to a binocular vision and laser-based distance measurement method, device, and intelligent wearable device. Background Art

[0002] Existing technologies typically use a single device to obtain distance measurement information, such as monocular vision, laser ranging, or multi-camera (binocular) vision ranging. However, these single devices have limitations in specific applications. For example, monocular vision ranging infers depth information by detecting and matching image feature points. While achieving certain accuracy and real-time performance, errors can increase significantly at longer distances and in complex environments (such as uneven lighting and target motion). Furthermore, monocular vision is sensitive to changes in the device's height and angle, making it susceptible to environmental fluctuations. Laser ranging technology offers high accuracy and is particularly suitable for short- and medium-range measurements. However, laser ranging equipment is relatively expensive and, during long-range measurements, is significantly affected by interference from reflectivity, angle, and external light, resulting in reduced ranging accuracy. Furthermore, laser ranging typically has a narrow field of view, making it difficult to capture complete distance information in multi-object or complex spatial environments. While using multi-camera (binocular) vision images to estimate target depth can achieve relatively good accuracy, it still faces challenges such as visual occlusion, parallax noise, and environmental interference. Especially when the target is in motion, the real-time matching and calculation of multi-viewing will become unstable, affecting the ranging effect.

[0003] Patent document CN118115945A proposes a substation construction safety management and control system and method based on binocular vision ranging, wherein the system includes: a safety management and control object determination subsystem, used to determine the safety management and control objects at the substation construction site; a target hazard source determination subsystem, used to determine the target hazard source at the substation construction site; a real-time distance detection subsystem, used to calibrate the binocular camera and obtain the real-time distance between the safety management and control objects and the target hazard source through the binocular camera; a management and control subsystem, used to determine the warning information for management and control based on the real-time distance and the safety distance against the warning template. This method requires calibration and the algorithm is complex. Patent document CN117647219A proposes a field of power inspection technology based on drones, specifically a distance measurement method based on multi-sensor fusion. This method uses a binocular camera, ultrasonic sensors, an inertial sensor (IMU) module, and a barometer to measure the drone's altitude. Using the initial IMU and barometer values as a benchmark, the method uses changes in IMU and barometer data during takeoff and landing to determine altitude. Altitude information measured by the ultrasonic sensor and binocular camera is used for auxiliary correction, achieving multi-sensor information fusion for altitude measurement. This method incorporates multiple sensors, resulting in high production costs and complex algorithms. Summary of the Invention

[0004] The present invention provides a binocular vision and laser-based ranging method, device and intelligent wearable device to solve the problems of complex algorithms and high costs in the existing ranging solutions.

[0005] A ranging method based on binocular vision and laser, comprising:

[0006] The first image and the second image of the target object at different viewing angles are collected by a binocular vision device of the smart wearable device;

[0007] Extracting, matching, and screening feature points of the target object based on the first image and the second image to obtain key point pairs, and calculating disparity based on the key point pairs;

[0008] Calculating a depth value between the smart wearable device and the target object according to the disparity;

[0009] The laser distance measurement value between the smart wearable device and the target object is collected by the laser device of the smart wearable device;

[0010] The depth value and the laser ranging value are fused using a Kalman filter algorithm to output a fused ranging result.

[0011] Furthermore, the laser distance measurement value of the target object is collected by the laser device, including:

[0012] The laser distance measurement value is obtained by collecting multiple sets of distance measurement values from the target object through the laser device and averaging the multiple sets of distance measurement values.

[0013] Furthermore, extracting, matching, and screening feature points of the target object based on the first image and the second image to obtain key point pairs includes:

[0014] Extracting feature points from the first image and the second image based on the SIFT algorithm, respectively, to obtain a plurality of first feature points and descriptors belonging to the first image and a plurality of second feature points and descriptors belonging to the second image;

[0015] Matching the first feature points and the second feature points based on the descriptor similarity to obtain a matching result;

[0016] The matching result pairs are screened to obtain key point pairs.

[0017] Furthermore, matching the first feature point and the second feature point based on the descriptor similarity to obtain a matching result includes:

[0018] Performing L2 normalization on the first descriptor set corresponding to the first feature point and the second descriptor set corresponding to the second feature point;

[0019] Establishing a first KD-tree index based on the first descriptor set;

[0020] Performing a nearest neighbor search on each second descriptor in the second descriptor set in the space of the first KD tree index to find a first nearest neighbor point and a first-order nearest neighbor point;

[0021] Calculating a first distance between each second descriptor and the corresponding first nearest neighbor point and a second distance between each second descriptor and the first nearest neighbor point, and calculating a first ratio of the first distance to the second distance, retaining the second descriptor and its corresponding first nearest neighbor point whose first ratio is less than a preset ratio as a first initial matching result;

[0022] Establishing a second KD-tree index based on the second descriptor set;

[0023] Performing a nearest neighbor search on each first descriptor in the first descriptor set in the space indexed by the second KD tree to find a second nearest neighbor point and a second nearest neighbor point;

[0024] Calculating a third distance between each first descriptor and the corresponding second nearest neighbor point and a fourth distance between each first descriptor and the second nearest neighbor point, and calculating a second ratio of the third distance to the fourth distance, and retaining the first descriptor and its corresponding second nearest neighbor point whose second ratio is less than a preset ratio as a second initial matching result;

[0025] The identical parts of the first initial matching result and the second initial matching result are taken as the final matching result.

[0026] Furthermore, the matching result includes a first matching feature point set belonging to the first image and a second matching feature point set belonging to the second image;

[0027] The matching results are screened to obtain key point pairs, including:

[0028] Randomly selecting a first matching feature point and a corresponding second matching feature point from the first matching feature point set and the second matching feature point set to form a matching feature point pair, and randomly selecting multiple matching feature point pairs to form multiple first minimum sample sets respectively;

[0029] Randomly selecting the first minimum sample set to calculate a first basic matrix, and calculating a first epipolar constraint error between each of the first basic matrices and all matching feature point pairs;

[0030] The matching feature point pairs whose first pair of epipolar constraint errors are less than a preset threshold are determined as initial inliers, and multiple initial inliers are randomly selected from all the obtained initial inliers to form multiple second minimum sample sets respectively;

[0031] randomly selecting the second minimum sample set to calculate a second fundamental matrix, and calculating a second epipolar constraint error between each second fundamental matrix and all initial inliers;

[0032] The initial inliers whose second epipolar constraint errors are less than a preset threshold are determined as optimal inliers, and the number of optimal inliers corresponding to each second fundamental matrix is counted;

[0033] The optimal inliers corresponding to the second fundamental matrix with the largest number of optimal inliers are determined as key point pairs.

[0034] Furthermore, the number of the key point pairs is multiple;

[0035] The method further comprises:

[0036] Perform disparity calculation based on multiple key point pairs to obtain multiple disparities;

[0037] A weighted average value is calculated for the multiple disparities, and the weighted average value is used as the depth value between the smart wearable device and the target object.

[0038] Furthermore, a Kalman filter algorithm is used to fuse the depth value with the laser ranging value, and a fused ranging result is output, including:

[0039] The actual distance between the smart wearable device and the target object and the relative speed between the smart wearable device and the target object are used as the state vector, and the covariance matrix, state transfer matrix and process noise matrix are assigned values;

[0040] Taking the depth value and the laser ranging value as observation quantities, constructing an observation model, and obtaining an observation matrix;

[0041] The observation noise matrix is constructed based on the depth measurement standard deviation of the binocular vision device and the nominal accuracy of the laser equipment;

[0042] Performing state prediction based on the initialized state vector and state transfer matrix to obtain a predicted state quantity, and performing covariance prediction based on the state transfer matrix, process noise matrix, and covariance matrix to obtain a prediction error covariance matrix;

[0043] Update the Kalman gain according to the prediction error covariance matrix, the observation matrix and the observation noise matrix;

[0044] Calculating based on the updated Kalman gain, the observed value, and the predicted state quantity to obtain a state estimate and update the covariance matrix;

[0045] The state estimation value is used as the fused ranging result.

[0046] Furthermore, the method further comprises:

[0047] Obtaining the acceleration of the smart wearable device through a posture detection device of the smart wearable device;

[0048] A compensation value is calculated according to the acceleration, and the observed quantity is compensated according to the compensation value.

[0049] A distance measuring device based on binocular vision and laser, comprising:

[0050] An acquisition module, configured to control the binocular vision device of the smart wearable device to acquire a first image and a second image of the target object at different viewing angles;

[0051] a feature extraction module, configured to extract, match, and filter feature points of an object based on the first image and the second image, obtain key point pairs, and calculate disparity based on the key point pairs;

[0052] A depth calculation module, configured to calculate a depth value between the smart wearable device and the target object based on the parallax;

[0053] The laser ranging module is used to control the laser device of the smart wearable device to collect the laser ranging value between the smart wearable device and the target object;

[0054] The fusion module is used to fuse the depth value with the laser ranging value using a Kalman filter algorithm and output a fused ranging result.

[0055] Furthermore, the laser ranging module collects the laser ranging value of the target object through the laser device, including:

[0056] The laser distance measurement value is obtained by collecting multiple sets of distance measurement values from the target object through the laser device and averaging the multiple sets of distance measurement values.

[0057] Furthermore, the feature extraction module extracts, matches, and filters feature points of the target object based on the first image and the second image to obtain key point pairs, including:

[0058] Extracting feature points from the first image and the second image based on the SIFT algorithm, respectively, to obtain a plurality of first feature points and descriptors belonging to the first image and a plurality of second feature points and descriptors belonging to the second image;

[0059] Matching the first feature points and the second feature points based on the descriptor similarity to obtain a matching result;

[0060] The matching result pairs are screened to obtain key point pairs.

[0061] Furthermore, the feature extraction module matches the first feature point and the second feature point based on the descriptor similarity to obtain a matching result, including:

[0062] Performing L2 normalization on the first descriptor set corresponding to the first feature point and the second descriptor set corresponding to the second feature point;

[0063] Establishing a first KD-tree index based on the first descriptor set;

[0064] Performing a nearest neighbor search on each second descriptor in the second descriptor set in the space of the first KD tree index to find a first nearest neighbor point and a first-order nearest neighbor point;

[0065] Calculating a first distance between each second descriptor and the corresponding first nearest neighbor point and a second distance between each second descriptor and the first nearest neighbor point, and calculating a first ratio of the first distance to the second distance, retaining the second descriptor and its corresponding first nearest neighbor point whose first ratio is less than a preset ratio as a first initial matching result;

[0066] Establishing a second KD-tree index based on the second descriptor set;

[0067] Performing a nearest neighbor search on each first descriptor in the first descriptor set in the space indexed by the second KD tree to find a second nearest neighbor point and a second nearest neighbor point;

[0068] Calculating a third distance between each first descriptor and the corresponding second nearest neighbor point and a fourth distance between each first descriptor and the second nearest neighbor point, and calculating a second ratio of the third distance to the fourth distance, and retaining the first descriptor and its corresponding second nearest neighbor point whose second ratio is less than a preset ratio as a second initial matching result;

[0069] The identical parts of the first initial matching result and the second initial matching result are taken as the final matching result.

[0070] Furthermore, the matching result includes a first matching feature point set belonging to the first image and a second matching feature point set belonging to the second image;

[0071] The feature extraction module screens the matching results to obtain key point pairs, including:

[0072] Randomly selecting a first matching feature point and a corresponding second matching feature point from the first matching feature point set and the second matching feature point set to form a matching feature point pair, and randomly selecting multiple matching feature point pairs to form multiple first minimum sample sets respectively;

[0073] Randomly selecting the first minimum sample set to calculate a first basic matrix, and calculating a first epipolar constraint error between each of the first basic matrices and all matching feature point pairs;

[0074] The matching feature point pairs whose first pair of epipolar constraint errors are less than a preset threshold are determined as initial inliers, and multiple initial inliers are randomly selected from all the obtained initial inliers to form multiple second minimum sample sets respectively;

[0075] randomly selecting the second minimum sample set to calculate a second fundamental matrix, and calculating a second epipolar constraint error between each second fundamental matrix and all initial inliers;

[0076] The initial inliers whose second epipolar constraint errors are less than a preset threshold are determined as optimal inliers, and the number of optimal inliers corresponding to each second fundamental matrix is counted;

[0077] The optimal inliers corresponding to the second fundamental matrix with the largest number of optimal inliers are determined as key point pairs.

[0078] Furthermore, the number of the key point pairs is multiple;

[0079] The depth calculation module is further configured to:

[0080] Perform disparity calculation based on multiple key point pairs to obtain multiple disparities;

[0081] A weighted average value is calculated for the multiple disparities, and the weighted average value is used as the depth value between the smart wearable device and the target object.

[0082] Furthermore, the fusion module uses a Kalman filter algorithm to fuse the depth value with the laser ranging value and outputs a fused ranging result, including:

[0083] The actual distance between the smart wearable device and the target object and the relative speed between the smart wearable device and the target object are used as the state vector, and the covariance matrix, state transfer matrix and process noise matrix are assigned values;

[0084] Taking the depth value and the laser ranging value as observation quantities, constructing an observation model, and obtaining an observation matrix;

[0085] The observation noise matrix is constructed based on the depth measurement standard deviation of the binocular vision device and the nominal accuracy of the laser equipment;

[0086] Performing state prediction based on the initialized state vector and state transfer matrix to obtain a predicted state quantity, and performing covariance prediction based on the state transfer matrix, process noise matrix, and covariance matrix to obtain a prediction error covariance matrix;

[0087] Update the Kalman gain according to the prediction error covariance matrix, the observation matrix and the observation noise matrix;

[0088] Calculating based on the updated Kalman gain, the observed value, and the predicted state quantity to obtain a state estimate and update the covariance matrix;

[0089] The state estimation value is used as the fused ranging result.

[0090] Furthermore, the fusion module is also used to:

[0091] Controlling the posture detection device of the smart wearable device to obtain the acceleration of the smart wearable device;

[0092] A compensation value is calculated according to the acceleration, and the observed quantity is compensated according to the compensation value.

[0093] An intelligent wearable device includes a binocular vision device and a laser device, and also includes a processor and a storage medium. The storage medium stores multiple instructions, and the processor is used to read the instructions and execute the above method.

[0094] The binocular vision and laser-based ranging method, device, and smart wearable device provided by the present invention have at least the following beneficial effects:

[0095] (1) Based on the binocular vision device and laser device of the smart wearable device, the depth value and laser ranging value are obtained respectively, and the two are fused to solve the limitation of single sensor in ranging. The Kalman filter is used to fuse the two, which has a simple algorithm and low cost.

[0096] (2) For images captured by the binocular vision device, after extracting features using the SIFT algorithm, feature points are matched through nearest neighbor search, and key points are screened based on epipolar constraints to improve the accuracy of feature point selection and thus improve the accuracy of subsequent ranging results;

[0097] (3) In feature point matching, the matching results are determined through bidirectional verification to further improve the accuracy of feature point matching;

[0098] (4) In the process of selecting key points based on the epipolar constraint, the quadratic fitting of the basic matrix is performed based on the interior points to improve the precision and thus improve the accuracy of key point selection;

[0099] (5) The attitude detection device based on the smart wearable device compensates the observation quantity in the Kalman filter to further improve the accuracy of ranging fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] Figure 1 The present invention provides a flow chart of an embodiment of a distance measurement method based on binocular vision and laser.

[0101] Figure 2 The present invention provides a flowchart of an embodiment of feature extraction in a ranging method based on binocular vision and laser.

[0102] Figure 3 The present invention provides a flowchart of an embodiment of feature matching in a ranging method based on binocular vision and laser.

[0103] Figure 4 The present invention provides a flowchart of an embodiment of feature screening in the ranging method based on binocular vision and laser.

[0104] Figure 5 This is a flowchart of an embodiment of ranging fusion in the ranging method based on binocular vision and laser provided by the present invention.

[0105] Figure 6 This is a structural schematic diagram of an embodiment of the ranging method based on binocular vision and laser provided by the present invention. DETAILED DESCRIPTION

[0106] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0107] refer to Figure 1 In some embodiments, a ranging method based on binocular vision and laser is provided, comprising:

[0108] S1, collecting a first image and a second image of a target object at different viewing angles through a binocular vision device of a smart wearable device;

[0109] S2. Extract, match, and filter feature points of the target object based on the first image and the second image to obtain key point pairs, and calculate disparity based on the key point pairs;

[0110] S3. Calculating a depth value between the smart wearable device and the target object based on the disparity;

[0111] S4. Collecting the laser distance value between the smart wearable device and the target object through the laser device of the smart wearable device;

[0112] S5. Use a Kalman filter algorithm to fuse the depth value and the laser ranging value, and output a fused ranging result.

[0113] Specifically, in step S1, a binocular vision device on the smart wearable device collects a first image and a second image of the same target object at different viewing angles.

[0114] Further, refer to Figure 2 In step S2, feature points of the target object are extracted, matched, and screened based on the first image and the second image to obtain key point pairs, including:

[0115] S21. Extract feature points from the first image and the second image based on the SIFT algorithm, respectively, to obtain a plurality of first feature points and descriptors belonging to the first image and a plurality of second feature points and descriptors belonging to the second image;

[0116] S22, matching the plurality of first feature points and second feature points based on descriptor similarity to obtain a matching result;

[0117] S23: Screen the matching result pairs to obtain key point pairs.

[0118] Furthermore, in step S21 , the SIFT algorithm is used to extract feature points and their descriptors from the first image and the second image.

[0119] Specifically, the SIFT algorithm is used to extract feature points and their descriptors from the first image and the second image, including: performing multi-scale Gaussian blur processing on the first image and the second image respectively to generate multiple groups of sub-images of different scales, each group of sub-images containing multiple gradually blurred Gaussian blurred images.

[0120] In each group of sub-images, Gaussian blurred images of adjacent scales are subtracted to obtain Gaussian difference images.

[0121] For each Gaussian difference image, each pixel is compared with its neighboring pixels within a preset neighborhood range. If the pixel value of the pixel is a maximum or minimum value, the pixel is used as a candidate feature point.

[0122] The contrast of the candidate feature points is compared, and the candidate feature points with contrast lower than the preset contrast value are determined as noise and removed. The curvature of the remaining candidate feature points is calculated using the Hessian matrix, and the points with strong edge response (points with a principal curvature ratio greater than the preset value) are removed, and the points with stable corner-like features are retained as the final feature points.

[0123] For the sub-image of the final feature point, the gradient magnitude and direction (typically using the L2 norm) of the pixels within its neighborhood window are calculated. The directions are divided into 36 bins (one bin every 10°) and a gradient direction histogram is generated. The peak direction of the gradient direction histogram is used as the main direction of the feature point. If other peaks exist (such as a secondary peak that is greater than 80% of the main peak), feature points with multiple directions are generated for the same location to enhance rotation invariance.

[0124] Each feature point neighborhood is divided into 4×4 sub-regions (a total of 16 blocks), and the gradient histogram (8-dimensional vector) in 8 directions is calculated for each sub-region to form a 128-dimensional (16×8) SIFT descriptor.

[0125] Further, refer to Figure 3In step S22, matching the first feature point and the second feature point based on the descriptor similarity to obtain a matching result includes:

[0126] S221, performing L2 normalization processing on a first descriptor set corresponding to a first feature point and a second descriptor set corresponding to a second feature point;

[0127] S222: Establish a first KD-tree index based on the first descriptor set;

[0128] S223: Perform a nearest neighbor search on each second descriptor in the second descriptor set in the space of the first KD tree index to find a first nearest neighbor point and a first secondary nearest neighbor point;

[0129] S224, respectively calculating a first distance between each second descriptor and the corresponding first nearest neighbor point and a second distance between each second descriptor and the first nearest neighbor point, and calculating a first ratio of the first distance to the second distance, retaining the second descriptor and its corresponding first nearest neighbor point for which the first ratio is less than a preset ratio as a first initial matching result;

[0130] S225. Establish a second KD-tree index based on the second descriptor set;

[0131] S226. Perform a nearest neighbor search on each first descriptor in the first descriptor set in the space of the second KD tree index to find a second nearest neighbor point and a second nearest neighbor point;

[0132] S227, respectively calculating a third distance between each first descriptor and its corresponding second nearest neighbor and a fourth distance between each first descriptor and its corresponding second nearest neighbor, and calculating a second ratio of the third distance to the fourth distance, retaining the first descriptor and its corresponding second nearest neighbor for which the second ratio is less than a preset ratio as the second initial matching result;

[0133] S228: Take the identical parts of the first initial matching result and the second initial matching result as the final matching result.

[0134] Specifically, in the matching algorithm, the first descriptor set corresponding to the first feature point and the second descriptor set corresponding to the second feature point undergo L2 normalization, which can eliminate fluctuations in the feature vector amplitude caused by changes in illumination and scale. During the search phase, the nearest neighbor and next-nearest neighbor points are searched for each descriptor in the corresponding feature space, and a distance ratio is introduced to filter valid matches. This step can filter out approximately 60% of false matches. Finally, a bidirectional match verification is performed based on the corresponding descriptor sets of the two images, retaining only the point pairs that are the nearest neighbors in the bidirectional search.

[0135] The final matching results include the nearest neighbor feature points and their descriptors in the first and second images. After the above calculations, there may be multiple nearest neighbor feature points and their descriptors in the first and second images. These nearest neighbor feature points in the first and second images are used as matching feature points to establish a first set of matching feature points for the first image and a second set of matching feature points for the second image, respectively.

[0136] Further, refer to Figure 4 , in step S23, the matching result includes a first matching feature point set belonging to the first image and a second matching feature point set belonging to the second image;

[0137] The matching results are screened to obtain key point pairs, including:

[0138] S231, randomly selecting a first matching feature point and a corresponding second matching feature point from the first matching feature point set and the second matching feature point set to form a matching feature point pair, and randomly selecting multiple matching feature point pairs to form multiple first minimum sample sets respectively;

[0139] S232: randomly select the first minimum sample set to calculate a first basic matrix, and calculate a first epipolar constraint error between each first basic matrix and all feature matching point pairs;

[0140] S233, determining the matching feature point pairs whose first pair of epipolar constraint errors are less than a preset threshold as initial inliers, and randomly selecting multiple initial inliers from all the obtained initial inliers to form multiple second minimum sample sets respectively;

[0141] S234, randomly selecting the second minimum sample set to calculate a second fundamental matrix, and calculating a second epipolar constraint error between each second fundamental matrix and all initial inliers;

[0142] S235, determining the initial inliers whose second epipolar constraint errors are less than a preset threshold as optimal inliers, and counting the number of optimal inliers corresponding to each second fundamental matrix;

[0143] S236. Determine the optimal interior points corresponding to the second basic matrix with the largest number of optimal interior points as key point pairs.

[0144] Specifically, in step S231, a first matching feature point and a corresponding second matching feature point are selected from the first matching feature point set and the second matching feature point set respectively to form a matching feature point pair, and multiple matching feature point pairs are randomly selected to form multiple first minimum sample sets. The number of matching feature point pairs in the first minimum sample set is at least 8 pairs, and the first minimum sample set is used to calculate the first basic matrix F1.

[0145] Furthermore, in step S232, a first minimum sample set is randomly selected. Based on the first minimum sample set, the first basic matrix F1 is first calculated. The basic matrix is a 3×3 singular matrix (rank 2) used to describe the epipolar geometric constraints of the binocular camera imaging. For the first matching feature point x l =(x l ,y l ,1) T and the second matching feature point x r =(x r ,y r ,1) T , the fundamental matrix F satisfies:

[0146]

[0147] The basic matrix F has 9 unknowns (due to scale equivalence, the actual degrees of freedom are 7), but each pair of matching feature points provides a linear constraint. Based on the linear constraint, a system of equations is established, and the system of equations is solved based on the singular value decomposition method to obtain the basic matrix.

[0148] To improve numerical stability, the coordinates of the matched feature points need to be normalized (for example, the coordinates are translated to the center of the image and scaled to the range of [-1, 1]).

[0149] The basic matrix F obtained by the singular value decomposition method needs to be forced to have a rank of 2, that is, the singular values of F are corrected (the smallest singular value is set to zero).

[0150] Furthermore, based on each calculated first fundamental matrix, first epipolar constraint errors of all corresponding first matching feature points and second matching feature points in the first matching feature point set and the second matching feature point set are calculated.

[0151] The epipolar constraint error e is calculated using the following formula:

[0152]

[0153] Where e represents the epipolar constraint error.

[0154] Furthermore, in step S233, all matching feature point pairs whose first epipolar constraint error is less than a preset threshold are determined as initial inliers. The number of inliers reflects the goodness of fit of the fundamental matrix F to the overall matching feature points. The more inliers there are, the closer the fundamental matrix is to the true geometric relationship.

[0155] Furthermore, a plurality of initial inliers are randomly selected from all the obtained initial inliers to form a plurality of second minimum sample sets respectively. Similarly, the number of inliers in the second minimum sample set is at least 8.

[0156] Furthermore, in step S234, the second minimum sample set is randomly selected to calculate the second fundamental matrix. Similarly, a system of equations is established based on formula (1), and the system of equations is solved using the singular value decomposition method to obtain the second fundamental matrix. The second epipolar constraint error between each second fundamental matrix and all initial inliers is calculated according to formula (2).

[0157] Furthermore, in step S235, the initial inliers whose second epipolar constraint errors are less than a preset threshold are determined as optimal inliers, and the number of optimal inliers corresponding to each second basic matrix is counted. The greater the number, the better the second basic matrix.

[0158] In step S236, the optimal inliers corresponding to the second basic matrix with the largest number of optimal inliers are determined as key point pairs, and the obtained key point pairs may be multiple pairs.

[0159] Furthermore, in step S2, the parallax is calculated as follows:

[0160] d=x L -x R ; (3)

[0161] Where d represents the disparity, x L Indicates the horizontal coordinate of the key point in the first image in the key point pair, x R Indicates the horizontal coordinate of the keypoint in the keypoint pair belonging to the second image.

[0162] In step S3, triangulation is used to convert the parallax into a depth value based on known camera parameters (such as focal length f and baseline length B). The depth value Z is calculated as:

[0163]

[0164] Where Z is the depth value, f is the focal length of the camera in pixels, B is the distance between the two cameras in the binocular vision device (baseline length), and d is the parallax in pixels.

[0165] In some embodiments, there may be multiple key point pairs, and thus multiple disparities may be calculated.

[0166] Therefore, in some embodiments, the method further comprises:

[0167] Perform disparity calculation based on multiple key point pairs to obtain multiple disparities;

[0168] A weighted average value is calculated for the multiple disparities, and the weighted average value is used as the depth value between the smart wearable device and the target object.

[0169] Specifically, the weighted average D f Calculated using the following formula:

[0170] D f =w1·D1+w2·D2+…+w n ·D n ; (5)

[0171] Among them, D1, D2, ..., D n The depth value calculated for each disparity point, w1,w2,…,w n is the corresponding weight.

[0172] Furthermore, in step S4, the laser distance measurement value of the target object is collected by the laser device of the smart wearable device, including:

[0173] The laser distance measurement value is obtained by collecting multiple sets of distance measurement values from the target object through the laser device and averaging the multiple sets of distance measurement values.

[0174] Specifically, the laser ranging algorithm calculates distance by emitting a light beam and measuring the round-trip time difference between the beam and the target. This method is suitable for high-precision measurement over short distances and can also complement the accuracy of multi-viewing algorithms in long-range measurement.

[0175] In actual scenarios, laser reflection may fluctuate due to the properties of the target surface (such as reflectivity), resulting in unstable distance measurement results. To this end, the system usually performs multiple measurements and takes the average to improve data reliability. Let the distance measurement result of each measurement be D i , the mean value D of n measurement results avg for:

[0176]

[0177] Due to interference from the external environment (such as sunlight or other infrared sources), the laser ranging data may be affected by noise. The signal can be smoothed by a low-pass filter. Assuming that the ranging signal is S(t), the frequency response of the low-pass filter is H(f), and the filtered signal S filtered(t) for:

[0178] S filtered (t)=S(t)*H(f); (7)

[0179] Further, refer to Figure 5 In step S5, the depth value and the laser ranging value are fused using a Kalman filter algorithm to output a fused ranging result, including:

[0180] S51, taking the actual distance between the smart wearable device and the target object and the relative speed between the smart wearable device and the target object as the state vector, and assigning values to the covariance matrix, the state transfer matrix, and the process noise matrix;

[0181] S52, using the depth value and the laser ranging value as observation quantities, constructing an observation model, and obtaining an observation matrix;

[0182] S53, constructing an observation noise matrix based on the depth measurement standard deviation of the binocular vision device and the nominal accuracy of the laser equipment;

[0183] S54, performing state prediction according to the initialized state vector and state transfer matrix to obtain a predicted state quantity, performing covariance prediction according to the state transfer matrix, process noise matrix and covariance matrix to obtain a prediction error covariance matrix;

[0184] S55, updating the Kalman gain according to the prediction error covariance matrix, the measurement matrix and the observation noise matrix;

[0185] S56, performing calculations based on the updated Kalman gain, the observed value, and the predicted state quantity to obtain a state estimate, and updating the covariance matrix;

[0186] S57: Use the state estimation value as the fused ranging result.

[0187] Specifically, in step S51, in order to reflect the dynamic changes in the target distance, the state vector is designed to include a combination of the distance and its rate of change (speed), that is, the actual distance between the smart wearable device and the target object and the relative speed between the smart wearable device and the target object. The depth value of binocular vision and the laser ranging value are both affected by motion (such as device jitter and target movement). By introducing the speed term, the dynamic change trend of the distance can be captured. If only a single distance value is used as the state quantity, the acceleration effect in the motion cannot be modeled, resulting in a lag in the filter's response to rapidly changing distances.

[0188] The process noise matrix Q is:

[0189]

[0190] Where q is the acceleration variance (unit: m 2 / s 4 ), calibrated by experiments (typical value q = 0.01), Δt is the frame time interval.

[0191] Assuming that the relative motion between the target object and the smart wearable device conforms to the uniform acceleration model, the state transfer equation is:

[0192]

[0193] Wherein, Δt is the frame interval time (unit: seconds), which is determined by the sensor sampling frequency (e.g., Δt = 0.033s at 30Hz); w k-1is process noise, reflecting model uncertainty (such as target acceleration fluctuation), is the state transfer matrix, X k|k-1 is the state quantity.

[0194] Specifically, the covariance matrix can be assigned an initial value or the covariance matrix updated in the previous round of calculation.

[0195] Furthermore, in step S52, the depth value and the laser ranging value are used as observation quantities to construct an observation model:

[0196]

[0197] Among them, Y k is the observed quantity, X k is the state quantity, v k is the observation noise, is the observation matrix.

[0198] Furthermore, in step S53, the observation noise matrix R is:

[0199]

[0200] Among them, σ vision is the standard deviation of depth measurement of binocular vision device, σ laser It is the nominal accuracy of the laser equipment’s ranging, obtained through measurement statistics.

[0201] Furthermore, in step S54, the predicted state quantity is as follows:

[0202]

[0203] in, represents the predicted state quantity at time k, represents the predicted state quantity at time k-1, F k represents the state transition matrix, B k represents the control input matrix, u k Indicates the control quantity.

[0204] The prediction error covariance matrix is shown below:

[0205]

[0206] in, represents the prediction error covariance matrix at time k, F k represents the state transition matrix, P k-1 represents the covariance matrix at time k-1, Q k represents the process noise matrix.

[0207] Furthermore, in step S55, the updated Kalman gain is as follows:

[0208]

[0209] Among them, K k is the Kalman gain at time k, represents the prediction error covariance matrix at time k, H k represents the observation matrix, R k represents the observation noise matrix.

[0210] Furthermore, in step S56, the state estimation value is as follows:

[0211]

[0212] in, is the estimated value of the state at time k, Yk represents the observed value, is the predicted state quantity at time k, and Hk represents the observation matrix.

[0213] The updated covariance matrix looks like this:

[0214]

[0215] Among them, P k represents the updated covariance matrix and I represents the identity matrix.

[0216] Furthermore, in step S57, the state estimation value is used as the fused ranging result, and the filter is updated once per frame (e.g., 30 Hz). By combining the real-time disparity calculation of binocular vision and the fast sampling of laser ranging (e.g., completed within 10 ms), continuous distance estimation in a dynamic environment is achieved.

[0217] Kalman filtering is used to predict and update system states in real time, typically in a predict-update loop to optimize ranging results. During the Kalman filtering process, the system predicts the current state based on sensor measurements and makes corrections based on new observations.

[0218] Furthermore, in some embodiments, the method further comprises:

[0219] Obtaining the acceleration of the smart wearable device through a posture detection device of the smart wearable device;

[0220] A compensation value is calculated according to the acceleration, and the observed quantity is compensated according to the compensation value.

[0221] Specifically, assuming the acceleration of the smart wearable device is a x ,a y ,a z , then the compensation value △D is:

[0222] ΔD=k x a x +k y a y +k z a z ; (17)

[0223] Among them, k x , k y , k z is the correction coefficient, which is determined based on experimental data.

[0224] The compensation value obtained is mainly used to correct the ranging results, especially when the device posture changes (such as acceleration or rotation). In the actual ranging process, the posture detection device of the smart wearable device, including the accelerometer and gyroscope, monitors the posture changes, which will affect the accuracy of the ranging data. Therefore, a compensation model is needed to correct these errors. This means that every time the device posture changes (for example, the acceleration changes), the calculated compensation value ΔD will adjust the current ranging result. In this way, the compensation value will correct the ranging error in real time, making the ranging result more accurate.

[0225] In this case, the compensation value (based on the attitude compensation of the accelerometer and gyroscope) can be used as part of the prediction error and applied to the state update process of the Kalman filter. Specifically, the compensation value can play a role in the error correction step of the Kalman filter.

[0226] During the update process of the Kalman filter, the obtained compensation value ΔD is used as additional correction information and updated in combination with the observation value, for example, the compensation value is added to the observation value. k The state estimate is adjusted based on this compensation information.

[0227] Specifically, the compensation value can also be used to modify the observation noise matrix and the process noise matrix. For example, to modify the observation noise matrix, the corresponding component can be added based on the compensation value to indicate an increase in "uncertainty":

[0228] R t =R0+f(△D); (18)

[0229] Among them, R t represents the observation noise matrix after correction, R0 represents the observation noise matrix before correction, and f(△D) represents a function related to the compensation value.

[0230] When the acceleration is significant, the process noise matrix can be modified by adding the corresponding components based on the compensation value:

[0231] Q t=Q0+g(△D); (19)

[0232] Among them, Q t represents the process noise matrix after correction, Q0 represents the process noise matrix before correction, and g(△D) represents a function related to the compensation value.

[0233] The Kalman filter can smooth out the ranging error of the device in a dynamic environment, and combined with the compensation information provided by the accelerometer and gyroscope, the updated ranging result is more accurate.

[0234] Each time the Kalman filter's ranging value is updated, the compensation value ΔD is combined with the observation data to reduce the system's sensitivity to acceleration or attitude changes, further improving the ranging accuracy.

[0235] The dynamic error of the system is compensated by Kalman filtering, and the ranging error is adjusted in combination with the real-time data of the accelerometer and gyroscope, so that the system can dynamically respond to environmental changes.

[0236] In some embodiments, reference Figure 6 , also provides a ranging device based on binocular vision and laser, including:

[0237] An acquisition module 201 is configured to control a binocular vision device of the smart wearable device to acquire a first image and a second image of a target object at different viewing angles;

[0238] A feature extraction module 202 is configured to extract, match, and filter feature points of an object based on the first image and the second image to obtain key point pairs, and calculate disparity based on the key point pairs;

[0239] A depth calculation module 203 is configured to calculate a depth value between the smart wearable device and the target object based on the parallax;

[0240] The laser ranging module 204 is used to control the laser device of the smart wearable device to collect the laser ranging value between the smart wearable device and the target object;

[0241] The fusion module 205 is configured to fuse the depth value with the laser ranging value using a Kalman filter algorithm and output a fused ranging result.

[0242] Furthermore, the laser ranging module 204 collects the laser ranging value of the target object through the laser device, including:

[0243] The laser distance measurement value is obtained by collecting multiple sets of distance measurement values from the target object through the laser device and averaging the multiple sets of distance measurement values.

[0244] Furthermore, the feature extraction module 202 extracts, matches, and filters feature points of the target object based on the first image and the second image to obtain key point pairs, including:

[0245] Extracting feature points from the first image and the second image based on the SIFT algorithm, respectively, to obtain a plurality of first feature points and descriptors belonging to the first image and a plurality of second feature points and descriptors belonging to the second image;

[0246] Matching the first feature points and the second feature points based on the descriptor similarity to obtain a matching result;

[0247] The matching result pairs are screened to obtain key point pairs.

[0248] Furthermore, the feature extraction module 202 matches the first feature point and the second feature point based on the descriptor similarity to obtain a matching result, including:

[0249] Performing L2 normalization on the first descriptor set corresponding to the first feature point and the second descriptor set corresponding to the second feature point;

[0250] Establishing a first KD-tree index based on the first descriptor set;

[0251] Performing a nearest neighbor search on each second descriptor in the second descriptor set in the space of the first KD tree index to find a first nearest neighbor point and a first-order nearest neighbor point;

[0252] Calculating a first distance between each second descriptor and the corresponding first nearest neighbor point and a second distance between each second descriptor and the first nearest neighbor point, and calculating a first ratio of the first distance to the second distance, retaining the second descriptor and its corresponding first nearest neighbor point whose first ratio is less than a preset ratio as a first initial matching result;

[0253] Establishing a second KD-tree index based on the second descriptor set;

[0254] Performing a nearest neighbor search on each first descriptor in the first descriptor set in the space indexed by the second KD tree to find a second nearest neighbor point and a second nearest neighbor point;

[0255] Calculating a third distance between each first descriptor and the corresponding second nearest neighbor point and a fourth distance between each first descriptor and the second nearest neighbor point, and calculating a second ratio of the third distance to the fourth distance, and retaining the first descriptor and its corresponding second nearest neighbor point whose second ratio is less than a preset ratio as a second initial matching result;

[0256] The identical parts of the first initial matching result and the second initial matching result are taken as the final matching result.

[0257] Furthermore, the matching result includes a first matching feature point set belonging to the first image and a second matching feature point set belonging to the second image;

[0258] The feature extraction module 202 filters the matching results to obtain key point pairs, including:

[0259] Randomly selecting a first matching feature point and a corresponding second matching feature point from the first matching feature point set and the second matching feature point set to form a matching feature point pair, and randomly selecting multiple matching feature point pairs to form multiple first minimum sample sets respectively;

[0260] Randomly selecting the first minimum sample set to calculate a first basic matrix, and calculating a first epipolar constraint error between each of the first basic matrices and all matching feature point pairs;

[0261] The matching feature point pairs whose first pair of epipolar constraint errors are less than a preset threshold are determined as initial inliers, and multiple initial inliers are randomly selected from all the obtained initial inliers to form multiple second minimum sample sets respectively;

[0262] randomly selecting the second minimum sample set to calculate a second fundamental matrix, and calculating a second epipolar constraint error between each second fundamental matrix and all initial inliers;

[0263] The initial inliers whose second epipolar constraint errors are less than a preset threshold are determined as optimal inliers, and the number of optimal inliers corresponding to each second fundamental matrix is counted;

[0264] The optimal inliers corresponding to the second fundamental matrix with the largest number of optimal inliers are determined as key point pairs.

[0265] Furthermore, the number of the key point pairs is multiple;

[0266] The depth calculation module 203 is further configured to:

[0267] Perform disparity calculation based on multiple key point pairs to obtain multiple disparities;

[0268] A weighted average value is calculated for the multiple disparities, and the weighted average value is used as the depth value between the smart wearable device and the target object.

[0269] Furthermore, the fusion module 205 uses a Kalman filter algorithm to fuse the depth value with the laser ranging value, and outputs a fused ranging result, including:

[0270] The actual distance between the smart wearable device and the target object and the relative speed between the smart wearable device and the target object are used as the state vector, and the covariance matrix, state transfer matrix and process noise matrix are assigned values;

[0271] Taking the depth value and the laser ranging value as observation quantities, constructing an observation model, and obtaining an observation matrix;

[0272] The observation noise matrix is constructed based on the depth measurement standard deviation of the binocular vision device and the nominal accuracy of the laser equipment;

[0273] Performing state prediction based on the initialized state vector and state transfer matrix to obtain a predicted state quantity, and performing covariance prediction based on the state transfer matrix, process noise matrix, and covariance matrix to obtain a prediction error covariance matrix;

[0274] Update the Kalman gain according to the prediction error covariance matrix, the observation matrix and the observation noise matrix;

[0275] Calculating based on the updated Kalman gain, the observed value, and the predicted state quantity to obtain a state estimate and update the covariance matrix;

[0276] The state estimation value is used as the fused ranging result.

[0277] Furthermore, the fusion module 205 is further configured to:

[0278] Controlling the posture detection device of the smart wearable device to obtain the acceleration of the smart wearable device;

[0279] A compensation value is calculated according to the acceleration, and the observed quantity is compensated according to the compensation value.

[0280] In some embodiments, a smart wearable device is also provided, including a binocular vision device and a laser device, as well as a processor and a storage medium, wherein the storage medium stores multiple instructions, and the processor is used to read the instructions and execute the above method.

[0281] The smart wearable device provided in the above embodiment can be applied in power line construction. For example, in high-risk operations across highways, the smart wearable device can detect the distance between the spanning frame and the object being spanned (highway, live wires). For spanning frames, the device can detect the conductor sag height, the height of the conductor suspension point, the spanning frame length and width, etc. For another example, for a fallen tower, the smart wearable device can be used to detect the distance from the fixed point to the center point of the tower, the distance between any two points on the fallen tower, and the distance between the tower and the live wire.

[0282] The binocular vision and laser-based ranging method, device, and smart wearable device provided in the above embodiments have at least the following beneficial effects:

[0283] (1) Based on the binocular vision device and laser device of the smart wearable device, the depth value and laser ranging value are obtained respectively, and the two are fused to solve the limitation of single sensor in ranging. The Kalman filter is used to fuse the two, which has a simple algorithm and low cost.

[0284] (2) For images captured by the binocular vision device, after extracting features using the SIFT algorithm, feature points are matched through nearest neighbor search, and key points are screened based on epipolar constraints to improve the accuracy of feature point selection and thus improve the accuracy of subsequent ranging results;

[0285] (3) In feature point matching, the matching results are determined through bidirectional verification to further improve the accuracy of feature point matching;

[0286] (4) In the process of selecting key points based on the epipolar constraint, the quadratic fitting of the basic matrix is performed based on the interior points to improve the precision and thus improve the accuracy of key point selection;

[0287] (5) The attitude detection device based on the smart wearable device compensates the observation quantity in the Kalman filter to further improve the accuracy of ranging fusion.

[0288] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such changes and modifications as fall within the scope of the claims and their equivalents.

Claims

1. A distance measurement method based on binocular vision and laser, characterized in that: include: The first image and the second image of the target object at different viewing angles are collected by a binocular vision device of the smart wearable device; Extracting, matching, and screening feature points of the target object based on the first image and the second image to obtain key point pairs, and calculating disparity based on the key point pairs; Calculating a depth value between the smart wearable device and the target object according to the disparity; The laser distance measurement value between the smart wearable device and the target object is collected by the laser device of the smart wearable device; The depth value and the laser ranging value are fused using a Kalman filter algorithm to output a fused ranging result.

2. The method according to claim 1, characterized in that The laser distance measurement value of the target object is collected by the laser device, including: The laser distance measurement value is obtained by collecting multiple sets of distance measurement values from the target object through the laser device and averaging the multiple sets of distance measurement values.

3. The method according to claim 1, characterized in that Extracting, matching, and screening feature points of the target object based on the first image and the second image to obtain key point pairs includes: Extracting feature points from the first image and the second image based on the SIFT algorithm, respectively, to obtain a plurality of first feature points and descriptors belonging to the first image and a plurality of second feature points and descriptors belonging to the second image; Matching the first feature points and the second feature points based on the descriptor similarity to obtain a matching result; The matching result pairs are screened to obtain key point pairs.

4. The method according to claim 3, characterized in that Matching the first feature point and the second feature point based on the descriptor similarity to obtain a matching result includes: Performing L2 normalization on the first descriptor set corresponding to the first feature point and the second descriptor set corresponding to the second feature point; Establishing a first KD-tree index based on the first descriptor set; Performing a nearest neighbor search on each second descriptor in the second descriptor set in the space of the first KD tree index to find a first nearest neighbor point and a first-order nearest neighbor point; Calculating a first distance between each second descriptor and the corresponding first nearest neighbor point and a second distance between each second descriptor and the first nearest neighbor point, and calculating a first ratio of the first distance to the second distance, retaining the second descriptor and its corresponding first nearest neighbor point whose first ratio is less than a preset ratio as a first initial matching result; Establishing a second KD-tree index based on the second descriptor set; Performing a nearest neighbor search on each first descriptor in the first descriptor set in the space indexed by the second KD tree to find a second nearest neighbor point and a second nearest neighbor point; Calculating a third distance between each first descriptor and the corresponding second nearest neighbor point and a fourth distance between each first descriptor and the second nearest neighbor point, and calculating a second ratio of the third distance to the fourth distance, and retaining the first descriptor and its corresponding second nearest neighbor point whose second ratio is less than a preset ratio as a second initial matching result; The identical parts of the first initial matching result and the second initial matching result are taken as the final matching result.

5. The method according to claim 3, characterized in that The matching result includes a first matching feature point set belonging to the first image and a second matching feature point set belonging to the second image; The matching results are screened to obtain key point pairs, including: Randomly selecting a first matching feature point and a corresponding second matching feature point from the first matching feature point set and the second matching feature point set to form a matching feature point pair, and randomly selecting multiple matching feature point pairs to form multiple first minimum sample sets respectively; Randomly selecting the first minimum sample set to calculate a first basic matrix, and calculating a first epipolar constraint error between each of the first basic matrices and all matching feature point pairs; The matching feature point pairs whose first pair of epipolar constraint errors are less than a preset threshold are determined as initial inliers, and multiple initial inliers are randomly selected from all the obtained initial inliers to form multiple second minimum sample sets respectively; randomly selecting the second minimum sample set to calculate a second fundamental matrix, and calculating a second epipolar constraint error between each second fundamental matrix and all initial inliers; The initial inliers whose second epipolar constraint errors are less than a preset threshold are determined as optimal inliers, and the number of optimal inliers corresponding to each second fundamental matrix is counted; The optimal inliers corresponding to the second fundamental matrix with the largest number of optimal inliers are determined as key point pairs.

6. The method according to claim 1, characterized in that The number of the key point pairs is multiple; The method further comprises: Perform disparity calculation based on multiple key point pairs to obtain multiple disparities; A weighted average value is calculated for the multiple disparities, and the weighted average value is used as the depth value between the smart wearable device and the target object.

7. The method according to claim 1, characterized in that The depth value and the laser ranging value are fused using a Kalman filter algorithm to output a fused ranging result, including: The actual distance between the smart wearable device and the target object and the relative speed between the smart wearable device and the target object are used as the state vector, and the covariance matrix, state transfer matrix and process noise matrix are assigned values; Taking the depth value and the laser ranging value as observation quantities, constructing an observation model, and obtaining an observation matrix; The observation noise matrix is constructed based on the depth measurement standard deviation of the binocular vision device and the nominal accuracy of the laser equipment; Performing state prediction based on the initialized state vector and state transfer matrix to obtain a predicted state quantity, and performing covariance prediction based on the state transfer matrix, process noise matrix, and covariance matrix to obtain a prediction error covariance matrix; Update the Kalman gain according to the prediction error covariance matrix, the observation matrix and the observation noise matrix; Calculating based on the updated Kalman gain, the observed value, and the predicted state quantity to obtain a state estimate and update the covariance matrix; The state estimation value is used as the fused ranging result.

8. The method according to claim 7, characterized in that The method further comprises: Obtaining the acceleration of the smart wearable device through a posture detection device of the smart wearable device; A compensation value is calculated according to the acceleration, and the observed quantity is compensated according to the compensation value.

9. A distance measuring device based on binocular vision and laser, characterized in that: include: An acquisition module, configured to control the binocular vision device of the smart wearable device to acquire a first image and a second image of the target object at different viewing angles; a feature extraction module, configured to extract, match, and filter feature points of an object based on the first image and the second image, obtain key point pairs, and calculate disparity based on the key point pairs; A depth calculation module, configured to calculate a depth value between the smart wearable device and the target object based on the parallax; The laser ranging module is used to control the laser device of the smart wearable device to collect the laser ranging value between the smart wearable device and the target object; The fusion module is used to fuse the depth value with the laser ranging value using a Kalman filter algorithm and output a fused ranging result.

10. A smart wearable device, characterized in that: It includes a binocular vision device and a laser device, and also includes a processor and a storage medium, wherein the storage medium stores multiple instructions, and the processor is used to read the instructions and execute any method according to claims 1-8.

Images