An event camera based irregular fragment mass inversion method

Abstract

The application provides an irregular fragment mass inversion method based on an event camera, first, binocular event stream data are subjected to density clustering, event clusters corresponding to each fragment are divided, and a centroid trajectory is tracked; then, the event clusters are converted into a spatial density map, discrete contour point sets are obtained by extracting closed contour lines; subsequently, significant corner points are screened in the point sets, are arranged according to the trend, and a normalized edge length sequence is calculated, so that a contour topological feature is generated; then, based on the feature, multiple frame contours of the same fragment are matched between continuous frames, and the largest area is selected as the optimal single-frame contour; finally, the equivalent image plane diameter is converted from the area of the optimal contour, the real physical size is calculated in combination with the depth and the focal length, and the mass of the fragment is finally inverted according to the preset thickness and the material density.

Description

Technical Field

[0001] This invention relates to a visual measurement technique, and more particularly to a method for inverting the quality of irregular fragments based on an event camera. Background Technology

[0002] The assessment of the destructive performance of high-speed fragment swarms is a crucial aspect of impact dynamics. Fragment mass and velocity are core physical parameters for calculating fragment kinetic energy and evaluating damage effects. In recent years, non-contact optical measurement methods based on high-speed cameras have been widely applied. However, the intense flash of light generated at the moment of explosion can cause saturation overexposure of traditional frame-based camera detectors, obscuring fragment details; simultaneously, the frame-based exposure principle leads to unavoidable motion blur and inter-frame information loss. Event-based visual measurement technology offers an innovative solution for high spatiotemporal resolution observation of the explosion process. Event cameras employ a biomimetic working principle, asynchronously outputting an event stream with pixel-level brightness changes. They possess core advantages such as extremely high dynamic range, extremely high temporal resolution, and extremely low data redundancy, effectively overcoming the oversaturation problem caused by the intense flash of explosion light.

[0003] In the prior art, patent application CN118196001A discloses a method and device for post-target damage assessment based on a binocular event camera. The method acquires event flow data of the fragment field behind the target using a binocular event camera and performs time integration to form a fragment field image. The centroid of the light spot is extracted, and the three-dimensional coordinates of the fragment in the world coordinate system are constructed based on the parallax method. The fragment velocity is obtained by time difference. Although this method uses the high dynamic range advantage of the event camera to solve the problem of saturation overexposure of traditional cameras under the strong light of the explosion, its processing flow essentially still treats the fragment as a point mass, and what is obtained is the velocity of the centroid. The final output is only the position and velocity information of the fragment, and does not involve extracting the fragment contour features from the event cluster and performing quality inversion, so it cannot complete the complete damage assessment. The core difficulty of the above patent in being unable to invert the quality of irregular fragments is that: (1) the fragment is a randomly broken polygon, and the shape cannot be assumed a priori; (2) the fragment rotates at high speed around the centroid during flight, and what is observed in a single frame is only the projected contour at a certain moment, which cannot represent the real shape. Summary of the Invention

[0004] The purpose of this invention is to provide a method for inverting the quality of irregular fragments based on an event camera, comprising: Step S100: The event stream data of the stereo event camera is spatially aggregated and divided into different event clusters by density clustering, the centroid of each event cluster is calculated to obtain the trajectory of the centroid, and the original event data of each event cluster is cached, wherein each event cluster corresponds to a fragment. Step S200: Preprocess each event cluster to generate spatial density values, extract closed contour lines from the spatial density values ​​as the projection contours of the fragments, and sample and sparsify the closed contour lines to obtain a discrete contour point set. Step S300: Extract corner points from the discrete contour point set and calculate the interior angles of each corner point. Retain corner points within a preset range as significant corner points, merge significant corner points with a distance less than a threshold, and obtain a significant corner point set. Arrange the corner points in the significant corner point set according to the contour direction to obtain the arrangement order of the significant corner points. Calculate the ratio of the side length of adjacent corner points to the length of the first and last closed side to obtain a normalized side length sequence. Generate a contour topology feature containing the number of significant corner points, the arrangement order, and the normalized side length sequence. Step S400: Perform multi-constraint matching between consecutive frames based on contour topology features to establish the correspondence between multiple frame projection contours; for the multiple frame projection contours of the same fragment that are successfully matched, select the single frame contour with the largest contour area as the optimal single frame contour. Step S500: Based on the contour area of ​​the optimal single frame contour, calculate the equivalent image plane diameter. Combine the depth and focal length to use the equivalent image plane diameter to inversely calculate the true physical size of the fragment. Output the final mass of the fragment according to the preset fragment thickness parameters and material density.

[0005] Furthermore, step S100 specifically includes: Step S110: Slice the event stream data of the left and right eyes according to the time window. In each time window, use the DBSCAN density clustering algorithm to divide the clustered events into different event clusters based on the spatial distance between events in the event stream data. Each event cluster corresponds to a fragment. Step S120: For each event cluster, calculate the weighted average of the coordinates of all events within it as the centroid coordinates of the cluster, process them in the order of the time window, and obtain the centroid trajectory of each fragment on the left and right eye planes respectively. Step S130: Cache the raw event stream data of each event cluster for the left and right eyes respectively. The raw event stream data includes the position coordinates of each event on the image plane, the trigger timestamp, and the polarity of the brightness change.

[0006] Further, step S200 includes: Step S210: Extract a local region centered on the centroid of the event cluster, and count the number of events occurring at each pixel location to generate a spatial density value. , , in, Indicates pixel position Spatial density value at that location, Area per unit pixel This represents the number of times the event is triggered at this pixel location. Step S220: Obtain all pixels within the local region. maximum value Corresponding closed contour lines As the effective projection contour boundary of the fragment. , in, This is the proportional threshold.

[0007] Further, step S300 includes: Step S310, for contour lines The discrete contour point set of the frame is obtained by equidistant sampling and sparsification of the pixels. , , in, For the first The pixel coordinates of each contour point ( x i , y i ), This represents the total number of contour points. For the number of frames; Step S320: Use the Douglas-Peucker algorithm to process the discrete contour point set. Key corner points are obtained by performing polygon fitting, and the key corner points are arranged clockwise to obtain a set of corner points. , , in, For the first Pixel coordinates of the corner points , This represents the total number of corner points; Step S330, calculate the first corner points interior angle of , , in, Indicates from point to The vector, Indicates from point to vectors, interior angles The value range is from 0° to 180°; Step S340: Set the lower limit threshold of the interior angle. and upper limit threshold The filtered set of significant corner points is obtained. , ; Step 350: Calculate the Euclidean distance between the two corner points. Less than The two corner points are merged into a new feature point, and the position of the merged new feature point is... Take the average of the two points. , in, and These are the pixel coordinates of two adjacent corner points; Step S360, obtain A new set of significant corner points : , in, For the first The pixel coordinates of the new significant corner points The total number of significant corner points; Step S370: Generate a normalized side length sequence of adjacent new significant corner points. ,in , in, Indicates the first From the first significant corner point to the... Euclidean distance between the three significant corner points Indicates the first The length of the first and last closed side from each significant corner point to the first significant corner point. , This indicates that the ratio of the length of the first and last sides to itself is always 1.

[0008] Further, the multi-constraint matching in step S400 includes: Matching constraint 1, the difference in the number of new significant corner points between adjacent frames satisfies: , in, Indicates the first The number of new salient corner points in the frame, Indicates the first The number of new significant corner points in the frame; Matching constraint 2, calculate the normalized side length sequence similarity. , , in, The similarity of the normalized side length sequences of adjacent frames. For the first The normalized side length sequence of the frame, For the first The normalized side length sequence of the frame, For the first The first frame A normalized side length value, For the first The first frame One normalized side length value; Greater than the threshold This constraint must be satisfied at that time; Matching constraint 3: The deviation between the average displacement of all corner points and the predicted displacement of the centroid does not exceed twice the historical standard deviation; If all three constraints are satisfied simultaneously, a multi-frame corner point correspondence table is established. It was confirmed that consecutive frames belonged to the same fragment. , in, For the first The first frame The pixel coordinates of the three significant corner points This represents the total number of frames involved in the fusion process.

[0009] Furthermore, in step S400, for the multiple frames of projected contours of the same fragment that have been successfully matched, the area of ​​each frame contour is calculated. The single-frame contour area with the largest area is selected as the optimal single-frame contour. , , , in, It's a formula for polygonal shoelaces. It is the image area of ​​the optimal single-frame contour. This indicates taking the maximum value. This indicates the index corresponding to the maximum value. , The number of frames corresponding to the optimal single-frame contour .

[0010] Further, step S500 includes: Step S510: Within the same time window, based on the acquired left and right eye centroid trajectories, stereo matching is performed on the centroids of the same fragment in the left and right eyes according to epipolar constraints, and the parallax of the centroid abscissa of the fragment on the left and right eye image planes is calculated. ,in, and These are the x-coordinates of the centroid of the same fragment on the left and right eye-viewing planes, respectively. Step S520, combining the known baseline distance of the stereo camera and camera focal length The depth of the fragment's location was obtained by inversion based on the principle of triangulation. , ; Step S530: The image area of ​​the optimal single-frame contour. Converted to equivalent image plane diameter , ; Step S540, combined with depth and camera focal length Inverted projected diameter , ; Step S550: Calculate the volume using an equivalent thin-plate model. : , , , in, For the actual projected area, This is the curvature correction factor. Projected area and equivalent thickness in the image The preset fragment thickness parameter; Step S560: Calculate the quality corresponding to the optimal single-frame contour as the final quality. , , in, The density of the fragment material.

[0011] Compared with the prior art, the present invention has the following advantages: (1) The present invention characterizes irregular shapes through contour topology features and utilizes the invariance of contour topology features to realize non-contact measurement of the quality of irregular fragments based on event cameras; (2) The present invention constructs a multi-frame tracking and optimal single-frame screening mechanism, completes fragment identification and trajectory association through continuous multi-frame multi-constraint matching, selects the single frame with the largest area as the optimal observation result, and solves the problem of single-frame projection deformation caused by high-speed fragment flipping; (3) The present invention does not require the laying of recycling media and manual screening and weighing, which significantly shortens the test cycle and reduces labor costs. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0013] Figure 2 This is a schematic diagram of the spatial density map of a random frame in this invention.

[0014] Figure 3 This is a schematic diagram of the closed contour line extraction result of a random frame in this invention.

[0015] Figure 4 This is a schematic diagram of the discrete contour point set and corner point extraction results of a random frame in this invention.

[0016] Figure 5 This is a schematic diagram of an optimal single-frame fragment outline in this invention.

[0017] Figure 6 This is a schematic diagram showing the relevant data and the top 10 areas of an optimal single-frame fragment outline in this invention. Detailed Implementation

[0018] Combination Figure 1 A method for inverting the quality of irregular fragments based on an event camera includes the following steps: Step S100, fragment detection and trajectory calculation based on spatial density clustering: the event stream data of the binocular event camera is spatially aggregated and divided into different event clusters by density clustering, the centroid of each event cluster is calculated to obtain the trajectory of the centroid, and the original event data of each event cluster is cached, wherein each event cluster corresponds to a fragment. Step S200, Single-frame fragment contour extraction and sampling sparsification: Preprocessing is performed on each event cluster to generate a spatial density map, and closed contour lines in the spatial density map are extracted as the projected contours of the fragments. The closed contour lines are sampled and sparsified to obtain a discrete contour point set. Step S300, fragment contour topology feature generation: extract corner points from the discrete contour point set and calculate the interior angles of each corner point. Retain corner points whose interior angles are within a preset range as significant corner points. Merge significant corner points that are too close to each other to obtain a significant corner point set. Arrange the corner points in the significant corner point set according to the contour direction to obtain the arrangement order of the significant corner points. Calculate the ratio of the side length of adjacent corner points to the length of the first and last closed side to obtain a normalized side length sequence. Generate contour topology features containing the number of significant corner points, the arrangement order, and the normalized side length sequence. Step S400, Multi-frame tracking and optimal single-frame selection: Based on the contour topology features, perform multi-constraint matching between consecutive frames to establish the correspondence between multi-frame projected contours; for the multi-frame projected contours of the same fragment that are successfully matched, select the single-frame contour with the largest contour area as the optimal single-frame contour. Step S500, Fragment size inversion and quality calculation: Based on the contour area of ​​the optimal single frame contour, the equivalent image plane diameter is converted. Combining depth and focal length, the true physical size of the fragment is inverted using the equivalent image plane diameter. The final quality of the fragment is output according to the preset fragment thickness parameters and material density.

[0019] Step S100 specifically includes: Step S110: Slice the event stream data of the left and right eyes according to the time window. In each time window, use the DBSCAN density clustering algorithm to divide the clustered events into different event clusters based on the spatial distance between events in the event stream data. Each event cluster corresponds to a fragment. Step S120: For each event cluster, calculate the weighted average of the coordinates of all events within it as the centroid coordinates of the cluster, process them in the order of the time window, and obtain the centroid trajectory of each fragment on the left and right eye planes respectively. Step S130: Cache the raw event stream data of each event cluster for the left and right eyes respectively. The raw event stream data includes the position coordinates of each event on the image plane, the trigger timestamp, and the polarity of the brightness change.

[0020] The centroid trajectory obtained in step S100 is used for motion consistency constraints in multi-frame tracking and optimal single-frame selection in step S400; the original event data of each event cluster is cached as input for contour extraction in step S200.

[0021] Step S200 specifically includes the following steps: Step S210: Extract a local region centered on the centroid of the event cluster, and generate a spatial density map by counting the number of events at each pixel location, such as... Figure 2 As shown, , in, Indicates pixel position Spatial density value at that location, Area per unit pixel This represents the number of times an event is triggered at that pixel location. The spatial density map reflects the degree of event clustering at the edge of the fragment on the image plane: the brightness changes drastically at the edge of the fragment, resulting in high event density, while the event density is low inside the fragment and in the background area.

[0022] Step S220: Take the maximum density value. That is, among all pixels in a local region To address the issue of absolute density differences between different fragments, a proportional threshold is set, referencing the hysteresis threshold principle in Canny edge detection. Extract density value equals Closed contour lines This is used as the effective projection contour boundary of the fragment, such as Figure 3 The above, .

[0023] Step S300 specifically includes the following steps: Step S310, for contour lines The pixels on the frame are sampled at equal intervals to reduce the number of pixels to a few dozen discrete points, i.e., one point is taken every 2 to 3 pixels. This reduces the original hundreds of consecutive pixels to dozens of discrete points, resulting in the discrete contour point set of the frame. ,like Figure 4 As shown, , in, For the first pixel coordinates of each contour point , This represents the total number of contour points. The number of frames.

[0024] Step S320: The Douglas-Peucker algorithm is used to process the dense set of discrete contour points. Polygon fitting is performed to simplify the dense set of discrete contour points into a small number of key corner points while preserving the overall contour shape, resulting in a set of corner points arranged clockwise. ,like Figure 4 As shown, , in, For the first Pixel coordinates of the corner points , This represents the total number of corner points.

[0025] Step S330, for each corner point Calculate its interior angles , , in, Indicates from point to The vector, Indicates from point to vectors, interior angles The value range is from 0° to 180°.

[0026] Step S340: Set the lower limit threshold of the interior angle. and upper limit threshold The filtered set of significant corner points is obtained. , , In this embodiment, the following is taken , The selection logic is as follows: remove overly sharp corners with an interior angle of less than 30° (their position is extremely sensitive to discretization error) and overly flat corners with an interior angle of greater than 150° (close to a straight line and lacking distinctiveness), and retain corners with moderate angles as stable feature points.

[0027] Step S350: For corner points that are too close, i.e., the Euclidean distance between the two corner points... , , in, and These are the pixel coordinates of any two adjacent corner points; Less than The corner points are merged into one feature point. The minimum distance threshold is set to 3-5 pixels, because two points smaller than 3 pixels basically overlap on the image and cannot be distinguished, so no effective information is lost after merging.

[0028] The position of the new feature point after merging is the average of the two points. , in, These are the pixel coordinates of the new feature points after merging.

[0029] Step S360, finally obtained A new set of significant corner points : , in, For the first The pixel coordinates of the new significant corner points This represents the total number of significant corner points.

[0030] Step S370: Generate a normalized side length sequence of adjacent new significant corner points. ,in, , in, Indicates the first From the first significant corner point to the... Euclidean distance between the three significant corner points Indicates the first The length of the first and last closed side from each significant corner point to the first significant corner point. , This indicates that the ratio of the length of the first and last sides to itself is always 1.

[0031] Normalized side length sequence These are scale-independent topological invariants. Regardless of the distance of the fragment or the size of the image plane, the normalized side length sequence of the same fragment remains unchanged. Therefore, the normalized side length sequence of the same fragment... The sequence remains constant across different distances and different frames.

[0032] In step S400, matching is performed after satisfying three matching constraints. The multi-constraint matching includes: Matching constraint 1, quantity consistency: The difference in the number of new significant corner points between adjacent frames satisfies the following inequality, , in, Indicates the first The number of new salient corner points in the frame, Indicates the first The number of new salient corner points in the frame. This constraint is based on the fact that the number of corner points remains relatively stable during the fragment flipping process, and only disappears briefly when a corner rotates to an invisible side.

[0033] Matching constraint 2, topological consistency: Calculate the similarity of normalized edge length sequences. , , in, The similarity of the normalized side length sequences of adjacent frames. For the first The normalized side length sequence of the frame, For the first The normalized side length sequence of the frame, For the first The first frame A normalized side length value, For the first The first frame The normalized side length value is calculated to be greater than the threshold. This confirms that the constraint is satisfied. Here, a threshold value is used. This formula calculates the average relative change rate of the side length sequence of adjacent frames, and a match is considered successful when the similarity is greater than 0.85. The value of 0.85 is based on the fact that the side length ratio fluctuation of the same fragment due to small changes in attitude between two consecutive frames is usually less than 15%, and the threshold of 0.85 can effectively distinguish the same fragment from different fragments.

[0034] Matching constraint 3, motion consistency: follows the statistical principle of "3 σ The principle is that the deviation between the average displacement of all corner points and the predicted displacement of the centroid should not exceed twice the historical standard deviation. The specific operation is as follows: Calculate the arithmetic mean of the Euclidean distances between the matching salient corner points in the current frame and the previous frame, and use this as the average displacement of all corner points. Based on the centroid trajectory obtained in step S120, the centroid position of the current frame is predicted using a uniform motion model. The distance between this predicted position and the centroid position of the previous frame is the predicted centroid displacement. ;Compare and ,like If the match is incorrect, it needs to be re-verified. This represents the standard deviation of historical displacement.

[0035] Matching output: When all three constraints are satisfied simultaneously, establish a multi-frame corner point correspondence table. It was confirmed that consecutive frames belonged to the same fragment. , in, For the first The first frame The pixel coordinates of the corner points Number the corner points. Frame number This represents the total number of frames involved in the fusion. The table records the number of consecutive frames for each corner point. The positional changes within the frame form a sub-trajectory for each corner point.

[0036] In step S400, for the multiple projected contours of the same fragment that have been successfully matched, the area of ​​each contour frame is calculated. The single-frame contour with the largest area is selected as the optimal single-frame contour, such as... Figure 5 , Figure 6 As shown, , , , in, It's a formula for polygonal shoelaces. It is the image area of ​​the optimal single-frame contour. This indicates taking the maximum value. This indicates the index corresponding to the maximum value. , The number of frames corresponding to the optimal single-frame contour .

[0037] The frame with the largest area corresponds to the moment when the fragment is directly facing the camera. At this moment, the projection profile distortion is the smallest and closest to the true largest projection surface. Therefore, the profile of this frame is selected as the basis for the final size inversion.

[0038] If some corner points in a frame are invisible due to occlusion, then topological consistency is utilized, based on the normalized side length sequence in step 2.8. The recorded side length ratios are used to interpolate and complete the missing projected contour edges using visible corner point positions. Specifically, the method involves: knowing the position of the missing edge in the normalized side length sequence... and the corresponding ratio value By combining the actual distances between the first and last visible corner points, we can deduce the location where the missing corner point should be.

[0039] Step S500 specifically includes the following steps: Step S510: Within the same time window, based on the acquired left and right eye centroid trajectories, stereo matching is performed on the centroids of the same fragment in the left and right eyes according to epipolar constraints, and the parallax of the centroid abscissa of the fragment on the left and right eye image planes is calculated. ,in, and These are the x-coordinates of the centroid of the same fragment on the left and right eye image planes, respectively.

[0040] Step S520, combining the known baseline distance of the stereo camera and camera focal length The depth of the fragment's location was obtained by inversion based on the principle of triangulation. , .

[0041] Step S530: The image area of ​​the optimal single-frame contour. Converted to equivalent image plane diameter , , Step S540, combined with depth and camera focal length Inverting the true physical size , .

[0042] Step S550: Calculate the volume using an equivalent thin-plate model. , , , , in, For the actual projected area, Here, we take the curvature correction factor. , Projected area and equivalent thickness in the image This is the preset fragment thickness parameter.

[0043] Step S560: Calculate the quality corresponding to the optimal single-frame contour as the final quality. : , in, The density of the fragment material.

Claims

1. A method for quality inversion of irregular fragments based on event cameras, characterized in that, include: Step S100: The event stream data of the stereo event camera is spatially aggregated and divided into different event clusters by density clustering, the centroid of each event cluster is calculated to obtain the trajectory of the centroid, and the original event data of each event cluster is cached, wherein each event cluster corresponds to a fragment. Step S200: Preprocess each event cluster to generate spatial density values, extract closed contour lines from the spatial density values ​​as the projection contours of the fragments, and sample and sparsify the closed contour lines to obtain a discrete contour point set. Step S300: Extract corner points from the discrete contour point set and calculate the interior angle of each corner point. Retain corner points within a preset range as significant corner points, and merge significant corner points with a distance less than a threshold to obtain a set of significant corner points. Arrange the corner points in the set of significant corner points according to the contour direction to obtain the arrangement order of the significant corner points; calculate the ratio of the side length of adjacent corner points to the length of the first and last closed side to obtain the normalized side length sequence; generate contour topological features containing the number of significant corner points, the arrangement order, and the normalized side length sequence. Step S400: Perform multi-constraint matching between consecutive frames based on contour topology features to establish the correspondence between multiple frame projection contours; for the multiple frame projection contours of the same fragment that are successfully matched, select the single frame contour with the largest contour area as the optimal single frame contour. Step S500: Based on the contour area of ​​the optimal single frame contour, calculate the equivalent image plane diameter. Combine the depth and focal length to use the equivalent image plane diameter to inversely calculate the true physical size of the fragment. Output the final mass of the fragment according to the preset fragment thickness parameters and material density.

2. The method according to claim 1, characterized in that, Step S100 specifically includes: Step S110: Slice the event stream data of the left and right eyes according to the time window. In each time window, use the DBSCAN density clustering algorithm to divide the clustered events into different event clusters based on the spatial distance between events in the event stream data. Each event cluster corresponds to a fragment. Step S120: For each event cluster, calculate the weighted average of the coordinates of all events within it as the centroid coordinates of the cluster, process them in the order of the time window, and obtain the centroid trajectory of each fragment on the left and right eye planes respectively. Step S130: Cache the raw event stream data of each event cluster for the left and right eyes respectively. The raw event stream data includes the position coordinates of each event on the image plane, the trigger timestamp, and the polarity of the brightness change.

3. The method according to claim 2, characterized in that, Step S200 includes: Step S210: Extract a local region centered on the centroid of the event cluster, and count the number of events occurring at each pixel location to generate a spatial density value. , ， in, Indicates pixel position Spatial density value at that location, Area per unit pixel This represents the number of times the event is triggered at that pixel location. Step S220: Obtain all pixels within the local region. maximum value Corresponding closed contour lines As the effective projection contour boundary of the fragment. , in, This is the proportional threshold.

4. The method according to claim 3, characterized in that, Step S300 includes: Step S310, for contour lines The discrete contour point set of the frame is obtained by equidistant sampling and sparsification of the pixels. , ， in, For the first pixel coordinates of each contour point , The total number of contour points. For the number of frames; Step S320: The Douglas-Peucker algorithm is used to process the discrete contour point set. Polygon fitting is performed to obtain key corner points, and the key corner points are arranged clockwise to obtain a set of corner points. , ， in, For the first Pixel coordinates of the corner points , This represents the total number of corner points; Step S330, calculate the first corner points interior angle of , ， in, Indicates from point to The vector, Indicates from point to vectors, interior angles The value ranges from 0° to 180°; Step S340: Set the lower limit threshold of the interior angle. and upper limit threshold The filtered set of significant corner points is obtained. , ； Step 350: Calculate the Euclidean distance between the two corner points. Less than The two corner points are merged into a new feature point, and the position of the merged new feature point is... Take the average of the two points. ， in, and These are the pixel coordinates of two adjacent corner points; Step S360, obtain A new set of significant corner points : ， in, For the first The pixel coordinates of the new significant corner points The total number of significant corner points; Step S370: Generate a normalized side length sequence of adjacent new significant corner points. ,in ， in, Indicates the first From the first significant corner point to the... Euclidean distance between the three significant corner points Indicates the first The length of the first and last closed side from each significant corner point to the first significant corner point. , This indicates that the ratio of the length of the first and last sides to its own length is always 1.

5. The method according to claim 4, characterized in that, The multi-constraint matching in step S400 includes: Matching constraint 1, the difference in the number of new significant corner points between adjacent frames satisfies: ， in, Indicates the first The number of new salient corner points in the frame, Indicates the first The number of new significant corner points in the frame; Matching constraint 2, calculate the normalized side length sequence similarity. , ， in, The similarity of the normalized side length sequences of adjacent frames. For the first The normalized side length sequence of the frame, For the first The normalized side length sequence of the frame, For the first The first frame A normalized side length value, For the first The first frame One normalized side length value; Greater than the threshold This constraint must be satisfied at that time; Matching constraint 3: The deviation between the average displacement of all corner points and the predicted displacement of the centroid does not exceed twice the historical standard deviation; If all three constraints are satisfied simultaneously, a multi-frame corner point correspondence table is established. It was confirmed that consecutive frames belonged to the same fragment. , in, For the first The first frame Pixel coordinates of a significant corner point This represents the total number of frames involved in the fusion process.

6. The method according to claim 5, characterized in that, In step S400, for the multiple frames of projected contours of the same fragment that have been successfully matched, the area of ​​each frame contour is calculated. The single-frame contour area with the largest area is selected as the optimal single-frame contour. ， ， ， in, It's a formula for polygonal shoelaces. It is the image area of ​​the optimal single-frame contour. This indicates taking the maximum value. This indicates the index corresponding to the maximum value. , The number of frames corresponding to the optimal single-frame contour .

7. The method according to claim 6, characterized in that, Step S500 includes: Step S510: Within the same time window, based on the acquired left and right eye centroid trajectories, stereo matching is performed on the centroids of the same fragment in the left and right eyes according to epipolar constraints, and the parallax of the centroid abscissa of the fragment on the left and right eye image planes is calculated. ,in, and These are the x-coordinates of the centroid of the same fragment on the left and right eye-viewing planes, respectively. Step S520, combining the known baseline distance of the stereo camera and camera focal length The depth of the fragment's location was obtained by inversion based on the principle of triangulation. , ； Step S530: The image area of ​​the optimal single-frame contour. Converted to equivalent image plane diameter , ； Step S540, combined with depth and camera focal length Inverted projected diameter , ； Step S550: Calculate the volume using an equivalent thin-plate model. : ， ， ， in, For the actual projected area, This is the curvature correction factor. Projected area and equivalent thickness in the image The preset fragment thickness parameter; Step S560: Calculate the quality corresponding to the optimal single-frame contour as the final quality. , ， in, The density of the fragment material.

Images