A multi-optical target recognition method based on the statistical characteristics of the number of BLOB regions

Abstract

The present invention discloses a multi-optical target recognition method based on the statistical quantity feature of BLOB regions; it solves the technical problem of low recognition efficiency existing in the prior art, which adopts the method of first occluding one optical target and then recognizing the other optical target; it includes steps such as the acquisition of collimated images, and the binary processing, the first digital morphological processing, the second digital morphological processing, and target recognition of the acquired images. This method is a multi-optical target recognition image processing algorithm based on the extraction of quantitative feature parameters, realizing the recognition of simulated optical targets and main laser targets in the simulated optical collimation process, with a processing time less than 1 second, meeting the requirements for accuracy and efficiency in the optical path docking and collimation process of large laser devices; moreover, the multi-optical target recognition image processing algorithm based on the extraction of quantitative feature parameters proposed in this paper is of great significance for improving the optical path adjustment efficiency of large laser device emission experiments.

Description

Technical Field

[0001] The present invention relates to a multi-optical target recognition method, and in particular to a multi-optical target recognition method based on the statistical feature of the number of BLOB regions. Background Art

[0002] The optical path alignment and collimation of a large laser device is one of the three collimation processes (optical path auto-collimation, simulated light collimation, and optical path alignment and collimation). For the three different collimation processes, the image features collected by each collimation process are completely different. Compared with the other two collimation processes, the collimation images processed by the optical path alignment and collimation have different characteristics. The main characteristics of the optical targets in the optical path alignment and collimation are as follows:

[0003] 1) The collimation image contains two optical targets, namely the simulated light target and the main laser target;

[0004] 2) The beam quality of the simulated light target is good. Theoretically, the simulated light target is a solid circular light spot. However, due to the unstable output of the simulated light source and the interference of optical elements in the optical path transmission, the simulated light target detected by the collimation CCD is an irregular, solid target without a definite geometric shape;

[0005] 3) The beam quality of the main laser target is poor, and the spot shape is extremely irregular, mainly manifested in: the beam has uncertain textures, the edges are tortuous, discontinuous, and the total area is large;

[0006] 4) The light intensity distribution of the main laser target is very unstable, and the spot shape, intensity, and position change with time;

[0007] 5) The sizes, relative positions, and intensities of the simulated light and the main laser targets are uncertain, and will change with the process of optical path collimation.

[0008] Based on the characteristics of the simulated light collimation image above, the simulated light collimation image processing algorithm not only needs to determine the relative positions of the two targets, but also needs to determine which target is the main laser target and which target is the simulated light target. That is to say, the simulated light collimation image processing algorithm needs to realize the recognition of two different optical targets.

[0009] Most of the existing optical target recognitions are for individual targets, that is, first block one optical target and recognize the other optical target. After the recognition is completed, the type of the other optical target is also determined accordingly. However, when blocking one of the optical targets, it is necessary to arrange corresponding optical components and adjust the parameters of the arranged optical components to make them meet the blocking function. However, the processes such as arranging optical components result in a low recognition efficiency of optical targets. Summary of the Invention

[0010] The object of the present invention is to solve the technical problem of low recognition efficiency in the prior art where one optical target is blocked first and then the other optical target is recognized, and to provide a multi-optical target recognition method based on BLOB region quantity feature statistics.

[0011] The concept of the present invention is as follows:

[0012] Geometric features, such as area, shape, edges, etc., cannot be used to recognize two optical targets. However, there are significant differences between the two optical targets. As long as the characteristics of the two optical targets are fully explored and quantitative feature parameters that can distinguish the two optical targets are proposed, the distinction between the two optical targets, namely the simulated light target and the main laser target, can be completed. This is an important prerequisite for successfully completing the optical path alignment and collimation.

[0013] In order to extract the quantitative feature parameters of the two optical targets, it is necessary to further analyze the characteristics of the two optical targets. The quantifiable feature parameters of the two targets are as follows:

[0014] 1) The simulated light target is a solid optical target with a continuous target area and no hole in the center of the target area;

[0015] 2) The simulated light target has an edge that is not very tortuous, and the area surrounded by the edge contains only one complete connected domain;

[0016] 3) The main laser target is a non-solid optical target with a discontinuous target area and many holes in the target area;

[0017] 4) The main laser target has multiple tortuous edges, and each edge is not connected to each other; at the same time, although the area surrounded by the edges also contains only one complete connected domain, there may also be holes in the connected domain.

[0018] Through the above analysis, it can be found that the obvious difference between the two optical targets is that the number of connected domains contained in the two optics is different. The simulated light target contains only one connected domain, and the main laser target contains at least two connected domains. Based on this feature, it is possible to distinguish which target is the simulated light target and which is the main laser target by statistically analyzing and comparing the number of connected domains contained in their respective target areas.

[0019] To achieve the above concept, the technical solution adopted by the present invention is:

[0020] A multi-optical target recognition method based on BLOB region quantity feature statistics, which is characterized in that:

[0021] Step 1: Collect the collimated image and perform binarization processing on the collected image;

[0022] Step 2: First digital morphological processing

[0023] Perform an erosion operation on the image obtained after Step 1;

[0024] Step 3: Second digital morphological processing

[0025] Perform a dilation operation on the image obtained after Step 2 to obtain the full image, as well as the number N of BLOB regions in the full image and the relevant information of each BLOB region; the relevant information includes the area area i , the coordinates C of the center position xi and C yi and the length L in the X-axis direction xi and the length L in the Y-axis direction yi , where N ≥ 3, 1 ≤ i ≤ N;

[0026] Step 4: Target recognition

[0027] 4.1 Set the two BLOB regions with the largest areas among the N BLOB regions as candidate recognition BLOB region 1 and candidate recognition BLOB region 2 respectively, where the center position coordinates of candidate recognition BLOB region 1 are defined as C xa and C ya and the length in the X-axis direction is L xa and the length in the Y-axis direction is L ya ; the center position coordinates of candidate recognition BLOB region 2 are C xb and C yb and the length in the X-axis direction is L xb and the length in the Y-axis direction is L yb , where 1 ≤ a ≤ N, 1 ≤ b ≤ N, a ≠ b;

[0028] 4.2 Based on C xa 、C ya 、L xa 、L ya 、C xb 、C yb 、L xb and L yb , respectively obtain the extended rectangular region information Region1 of candidate recognition BLOB region 1 and the extended rectangular region information Region2 of candidate recognition BLOB region 2;

[0029] 4.3 Respectively obtain the number of the center position coordinates C xi and C yi of the N BLOB regions located in the extended rectangular region information Region1 and the extended rectangular region information Region2, and perform target recognition on the simulated light target and the main laser target.

[0030] Further, in step 3, the area is obtained. i The coordinates C of the center position xi and C yi and the length L in the X-axis direction xi and the length L in the Y-axis direction yi are obtained as follows:

[0031] A Binary Large Object (BLOB) region refers to a set of pixels in an image that have similar features (such as texture, color, etc.) and are spatially connected, i.e., a connected domain. Through BLOB analysis, the target object can be separated from the background, and then object feature parameters can be extracted, such as centroid, center of gravity, perimeter, area, dimensions in the horizontal / vertical directions, number of pixels, etc. The feature information of each BLOB region in a collimated image is stored in a linked list, i.e., the BLOB region is the same as the linked list, represented by blobcount. Then, the area area of each BLOB region i The coordinates C of the center position xi and C yi and the length L in the X-axis direction xi and the length L in the Y-axis direction yi are expressed by the formula:

[0032]

[0033] In the formula, endsNumber i is the number of rows of the line segment table corresponding to the chain code table of the i-th BLOB region, pPoint[k].x i is the starting position of the horizontal line segment table of the i-th BLOB region, pPoint[k + 1].x i is the ending position of the horizontal line segment table of the i-th BLOB region, pPoint[k].y i is the starting position of the vertical line segment table of the i-th BLOB region, pPoint[k + 1].y i is the ending position of the vertical line segment table of the i-th BLOB region, point-sum i is the number of pixels included in the i-th BLOB region, k represents the k-th row in the line segment table corresponding to the chain code table, is the intermediate value, defined as

[0034] Further, step 4.2 specifically includes the following steps:

[0035] 4.2.1 Obtain the left upper endpoint value and the right lower endpoint value of the rectangle to be expanded for candidate recognition BLOB region 1 and candidate recognition BLOB region 2 respectively:

[0036] Calculate the upper left endpoint value and the lower right endpoint value of the rectangle to be expanded for the candidate recognition BLOB region 1:

[0037] The upper left endpoint value is (C xa -L xa , C ya -L ya );

[0038] The lower right endpoint value is (C xa +L xa , C ya +L ya );

[0039] The method for obtaining the upper left endpoint value and the lower right endpoint value of the rectangle to be expanded for the candidate recognition BLOB region 2 is the same as the method for obtaining the upper left endpoint value and the lower right endpoint value of the rectangle to be expanded for the candidate recognition BLOB region 1;

[0040] 4.2.2 Based on the upper left endpoint value and the lower right endpoint value obtained in step 4.2.1, draw the expanded rectangle regions for the selected recognition BLOB region 1 and the candidate recognition BLOB region 2 respectively, and obtain the expanded rectangle region information Region1 and the expanded rectangle region information Region2.

[0041] Further, step 4.3 specifically includes the following steps:

[0042] 4.3.1 Statistically count the number BlobCount1 of the central position coordinates C xi and C yi located in the expanded rectangle region information Region1 and the number BlobCount2 located in the expanded rectangle region information Region2 for N BLOB regions respectively;

[0043] 4.3.2 If BlobCount1 > BlobCount2, the candidate recognition BLOB region 1 corresponding to BlobCount1 is the main laser target, and the candidate recognition BLOB region 2 corresponding to BlobCount2 is the simulated light target; if BlobCount1 < BlobCount2, the candidate recognition BLOB region 1 corresponding to BlobCount1 is the simulated light target, and the candidate recognition BLOB region 2 corresponding to BlobCount2 is the main laser target.

[0044] The beneficial effects of the present invention are:

[0045] 1. The present invention proposes a multi-optical target recognition method based on the statistical characteristics of the number of BLOB regions, which is a multi-optical target recognition image processing algorithm based on the extraction of quantization feature parameters, and realizes the recognition of the simulated light target and the main laser target in the simulated light collimation process. Moreover, the multi-optical target recognition image processing algorithm based on the extraction of quantization feature parameters proposed in this paper is of great significance for improving the adjustment efficiency of the experimental optical path of large laser devices and ensuring the success of physical experiments.

[0046] 2. According to the automatic collimation design index of the optical path of large devices, it is required that the target recognition processing time of each collimation image of the optical path docking is less than 1 second. The present invention proposes a multi-optical target recognition method based on the statistical characteristics of the number of BLOB regions. After binarization processing, erosion operation, dilation operation and target recognition, the time used is less than 1 second, meeting the automatic collimation design index of the optical path of large devices.

[0047] 3. The present invention proposes a multi-optical target recognition method based on the statistical characteristics of the number of BLOB regions, which can simultaneously recognize two optical targets (main laser target and simulated light target) in a collimation image, and there is no need to block the light source of another optical target generated in order to generate only one optical target in the collimation image.

[0048] 4. The present invention proposes a multi-optical target recognition method based on the statistical characteristics of the number of BLOB regions, which can recognize two optical targets in the collimation image collected at any time during the optical path docking and collimation process without human intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is the flowchart of the embodiment of the present invention;

[0050] Figure 2 is the original image of the optical path docking and collimation in the first embodiment of the present invention;

[0051] Figure 3 is the Figure 2 image obtained after binarization processing;

[0052] FIG. 4(a) is the Figure 3 simulated light target image obtained after cropping;

[0053] FIG. 4(b) is the image obtained after erosion operation on FIG. 4(a);

[0054] FIG. 4(c) is the Figure 3 main laser target image obtained after cropping;

[0055] FIG. 4(d) is the image obtained after erosion operation on FIG. 4(c);

[0056] Figure 4(e) is Figure 3 each BLOB information in

[0057] Figure 4(f) is each BLOB information corresponding to Figure 4(e) in Figure 4(d);

[0058] Figure 4(g) is the number of BLOB regions in the whole image and the relevant information of each BLOB region after Figure 4(d) is completed;

[0059] Figure 5 is the number of BLOB regions in the whole image and the relevant information of each BLOB region after the image of Figure 4(d) undergoes dilation operation;

[0060] Figure 6 is the process diagram of the two regions with the largest search area in the first embodiment of the present invention. Among them, 1 is the binary image, 2 is the retrieval of BLOB region information, and 3 is the region with the largest area searched;

[0061] Figure 7 is the diagram of the original rectangular region and the extended rectangular region in the first embodiment of the present invention;

[0062] Figure 8 is the target recognition result diagram of the first embodiment of the present invention;

[0063] Figure 9 is the target recognition result diagram of the second embodiment of the present invention. Specific embodiments

[0064] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0065] The present invention provides a multi-optical target recognition method based on BLOB region quantity feature statistics through the following embodiments, as Figure 1 shown, including the following steps:

[0066] Step 1: Collect a collimated image and perform binary processing on the collected collimated image;

[0067] The original collimated image of the optical path docking is as Figure 2 shown. It can be seen from Figure 2 that the collimated image of the optical path docking contains two optical path targets. The left one is the simulated light target, and the right one is the main laser target. Among them, the simulated light target is a solid optical target with a continuous target area and no holes in the center of the target area; while the main laser target is a non-solid optical target with a discontinuous target area and many holes in the target area.

[0068] Perform binary processing on the original collimated image of the optical path docking, and the result is as Figure 3As shown in the figure, when observing the binary image, the connected region features of the two targets are more obvious: that is, the simulated light target region only contains one connected region, while the main laser target region contains multiple connected regions (more than 5), and the center of the largest connected region is non-solid and an irregular connected region containing many holes;

[0069] Step 2: First digital morphological processing

[0070] For Figure 3 the BLOB information in is statistically analyzed;

[0071] To highlight the connectivity of each BLOB region, the main laser target (on the right) is split into more non-connected BLOB regions. At this time, the binary image needs to be processed by erosion operation for the first digital morphological processing.

[0072] Specifically: To highlight the processing results of the erosion operation on the simulated light target and the main laser target, the simulated light target and the main laser target are separately collected and processed. Among them, the results before and after the erosion of the simulated light target are shown in Figures 4(a) and 4(b). By comparing the images of the simulated light target before and after the erosion operation, it is found that the simulated light target hardly changes in shape and still has only one large pixel point, designated as No. 1. This is because the simulated light target is solid and there are no discrete pixels at the edge, and the erosion operation basically does not change the geometric shape of the simulated light target.

[0073] The results before and after the erosion of the main laser target are shown in Figures 4(c) and 4(d), which contain 8 pixel points. By comparing the images of the main laser target before and after the erosion operation, it is found that although the main laser target changes little in shape, there are obvious changes in the edge region of the main laser target, mainly manifested as: 1) The discrete pixel points in the eroded image disappear, such as points A and B corresponding in Figures 4(e) and 4(f); 2) The areas of some BLOB regions in the eroded image become smaller, such as points C and D corresponding in Figures 4(e) and 4(f). The reason for the obvious change in the main laser target before and after the erosion operation is mainly that the main laser target is non-solid and the target region is non-continuous, and each target contains many holes.

[0074] Regarding the obvious features of the simulated light target and the main laser target, this just provides a basis for the recognition and feature extraction of the two optical targets.

[0075] As shown in Figure 4(g), after the erosion operation, the relevant information of the two largest BLOB regions in the obtained full image is shown in Table 1. Among them, the positive X-axis direction extends from the upper left corner to the right, and the positive Y-axis direction extends from the upper left corner downwards;

[0076] Table 1

[0077] Number Area Central position coordinates Axial length 1 10053 (140,154) [132126] 2 4006 (424,272)

[9586]

[0078] It is initially inferred that the BLOB regions corresponding to No. 1 and No. 2 are the simulated light target and the main laser target, but which one is the simulated light target and which one is the main laser target need to be further determined.

[0079] Step 3: Second digital morphological processing

[0080] According to the digital morphology theory, the dilation algorithm is defined as a set operation. The dilation of Q by U is the set composed of the origin positions of all structural elements. The dilation of Q by U is denoted as Defined as

[0081]

[0082] In the formula, is an empty set, U is a structural element, and z is the entire set.

[0083] In order to merge the main laser targets containing irregular textures into a larger and more complete target, in this embodiment, a 5*5 structural element is taken as an example for explanation. The structural element is shown as matrix U:

[0084]

[0085] The dilation processing is performed on the binary image after the erosion operation, which is represented by the following formula:

[0086] where f delate (x, y) represents the image function after dilation, and f bin (x, y) represents the image function before dilation, and U is the structural element;

[0087] As Figure 5 shown, the second digital morphological processing is performed. Through large-scale dilation operation, the non-connected main laser targets are dilated to generate a connected region with a larger area. After the second digital morphological processing is completed, the BLOB detailed information of the image after large-scale dilation is statistically analyzed for the second time, and the relevant information of each BLOB region in the first and second times is summarized and compared. The comparison results are shown in Table 2:

[0088] Table 2

[0089]

[0090] As can be seen from Table 2, after the dilation operation, each BLOB region has two aspects of changes: 1) The area of each BLOB region becomes larger. For example, the area of No. 1 changes from 10053 to 10857, and the area of No. 2 changes from 4006 to 5964; 2) The axis lengths in the horizontal and vertical directions become larger. For example, the horizontal axis length of No. 1 changes from 132 to 135, and the vertical axis length changes from 126 to 129; the horizontal axis length of No. 2 changes from 95 to 101, and the vertical axis length changes from 86 to 89.

[0091] The large-size dilation operation is to dilate and enlarge each BLOB region. For the simulated light target, since the simulated light target is a solid optical target, the large-size dilation operation only enlarges the simulated light target proportionally around it, and the geometric shape of the simulated light target hardly changes; while for the main laser target, since the main laser target is a non-solid optical target, the main changes of the main laser target after the large-size dilation operation are: 1) Each BLOB region becomes a larger connected region; 2) Since the area of each BLOB region becomes larger, it is possible that the BLOB regions will stick together, and several BLOB regions may merge into a BLOB region with a larger area; 3) Due to the dilation operation, the holes in each BLOB region may become smaller or even disappear.

[0092] The comparison of the main laser target before and after the large-size dilation operation is shown in Figure 4(g) and Figure 5 as shown.

[0093] From Figure 5 it can be seen that the connected region areas of the BLOB regions corresponding to No. 3 and No. 4 become larger; the BLOB region corresponding to No. 10 disappears; the BLOB region corresponding to No. 2 not only has a larger area, but also the 4 holes in the BLOB region disappear, and the characteristic change of the BLOB region corresponding to this number is the largest; the connected region area of the BLOB region corresponding to No. 5 becomes larger; the BLOB regions corresponding to No. 11 and No. 12 merge into a larger connected region, that is to say, after the large-size dilation operation, the BLOB regions corresponding to No. 11 and No. 12 are merged into the BLOB region corresponding to No. 2; the connected region areas of the BLOB regions corresponding to No. 8 and No. 9 also become larger. After the erosion operation, the disappeared pixel points will be displayed again after the dilation operation, such as No. 6 and No. 7.

[0094] Step 4: Target recognition

[0095] The purpose of target recognition is to use quantitative metrics, namely the number of BlobCount of each BLOB region obtained from two digital morphological processes, based on the detailed information of each BLOB region. Assume the number of center coordinates corresponding to the simulated optical target located in candidate recognition BLOB region 1 and candidate recognition BLOB region 2 is BlobCount1, and the number of center coordinates corresponding to the main laser target located in candidate recognition BLOB region 1 and candidate recognition BLOB region 2 is BlobCount2. Compare the sizes of BlobCount1 and BlobCount2. If BlobCount1 < BlobCount2, the candidate recognition BLOB region 1 or candidate recognition BLOB region 2 corresponding to BlobCount1 is the simulated optical target, and the candidate recognition BLOB region 2 or candidate recognition BLOB region 1 corresponding to BlobCount2 is the main laser target; if BlobCount1 > BlobCount2, the candidate recognition BLOB region 1 or candidate recognition BLOB region 2 corresponding to BlobCount1 is the main laser target, and the candidate recognition BLOB region 2 or candidate recognition BLOB region 1 corresponding to BlobCount2 is the simulated optical target.

[0096] Therefore, target recognition mainly consists of the following steps:

[0097] 1) Search for the two regions with the largest area, set them as candidate recognition BLOB region 1 and candidate recognition BLOB region 2, and set the recognition flag MngFlag to 3;

[0098] 2) Search for the rectangular regions of each BLOB region;

[0099] 3) Count the number of BLOB regions whose center coordinates are located in candidate recognition BLOB region 1 and candidate recognition BLOB region 2 after the second digital morphological process, denoted as BlobCount1 and BlobCount2 respectively, compare the sizes of BlobCount1 and BlobCount2, and perform target recognition on the simulated optical target and the main laser target.

[0100] Specifically as follows:

[0101] 1) Search for the two regions with the largest area, set them as candidate recognition BLOB region 1 and candidate recognition BLOB region 2, and use them as candidate targets for the simulated optical target and the main laser target. Since the areas of No. 1 and No. 2 are the largest, which are 10857 and 5964 respectively, the BLOB regions corresponding to No. 1 and No. 2 are candidate targets for the simulated optical target and the main laser target.

[0102] Set the MngFlag of the two BLOB regions with the largest areas to 3, specifically as shown in the 1st and 2nd rows, 5th column of Table 3 MngFlag. The search results of the two BLOB regions with the largest areas are as Figure 6 shown in the third column of the figure.

[0103] Table 3

[0104]

[0105] 2) Search for the corresponding rectangular regions of the two BLOB regions with the largest areas

[0106] Search for the corresponding rectangular regions of candidate recognition BLOB region 1 and candidate recognition BLOB region 2, which is to prepare for counting the number of each BLOB whose center coordinates are located within candidate recognition BLOB region 1 and candidate recognition BLOB region 2.

[0107] As Figure 7 shown, the corresponding rectangular regions A and B of candidate recognition BLOB region 1 and candidate recognition BLOB region 2, expressed as [left, top][right, bottom], respectively represent [upper left X, upper left Y][lower right X, lower right Y] of the BLOB region, and are expressed by the formula: BLOB region left = center coordinate X - X-axis length; BLOB region top = center coordinate Y - Y-axis length; BLOB region right = center coordinate X + X-axis length; BLOB region bottom = center coordinate Y + Y-axis length.

[0108] As shown in Table 3, the BLOB rectangular region corresponding to No. 1 = [4, 26][274, 284], and the BLOB rectangular region corresponding to No. 2 = [324, 184][526, 362]. For example, the BLOB region corresponding to No. 1 = [139 - 135, 155 - 129][139 + 135, 155 + 129] = [4, 26][274, 284]. The BLOB region corresponding to No. 1 is as Figure 7 the area selected by rectangular frame A, where A' is the corresponding area of the original size of this BLOB region before size magnification. The BLOB region corresponding to No. 2 is as Figure 7 the area selected by rectangular frame B, where B' is the corresponding area of the original size of this BLOB region before size magnification.

[0109] 3) After the second digital morphological processing, count the number of BLOB regions with the center coordinates of all BLOB regions in candidate recognition BLOB region 1 and candidate recognition BLOB region 2, denoted as BlobCount1 and BlobCount2 respectively, compare the sizes of BlobCount1 and BlobCount2, and perform target recognition on the simulated light target and the main laser target. The statistical results are shown in Table 4;

[0110] Table 4

[0111]

[0112] The target recognition decision-making process for the simulated light target and the main laser target is mainly achieved by counting the number of BLOBs with the center coordinates of each BLOB located in the extended rectangle region information Region1 and the extended rectangle region information Region2.

[0113] For the schematic diagram of the target recognition process, as shown in Figure 7 and Table 4. For the BLOB region numbered 1, among the center coordinates of the 9 BLOB regions, only the center (139, 155) of number 1 is located within the rectangle A. Therefore, the number of BLOBs with the center coordinates of the largest area numbered 1 located in the two largest areas BlobCount1 = 1.

[0114] For the BLOB region corresponding to number 2, among the center coordinates of the 9 BLOB regions, 8 coordinates, namely the center coordinates of number 2 (425, 274), the center coordinates of number 3 (416, 215),..., the center coordinates of number 9 (449, 326) are located within the range [324, 526][184, 362] of the rectangle B.

[0115] Therefore, the number of BLOBs with the center coordinates of the largest area numbered 2 located in the two largest areas BlobCount2 = 8, as shown in the last row of Table 4.

[0116] 4) Compare the number of center coordinates located in the two largest BLOB regions

[0117] After obtaining the numbers BlobCount1 and BlobCount2 of the center coordinates of each BLOB located in the extended rectangular region information Region1 and the extended rectangular region information Region2, compare the sizes of BlobCount1 and BlobCount2. If BlobCount1 < BlobCount2, then the candidate recognition BLOB region 1 or candidate recognition BLOB region 2 corresponding to BlobCount1 is the simulated light target, and the candidate recognition BLOB region 2 or candidate recognition BLOB region 1 corresponding to BlobCount2 is the main laser target; conversely, the candidate recognition BLOB region 1 or candidate recognition BLOB region 2 corresponding to BlobCount1 is the main laser target, and the candidate recognition BLOB region 2 or candidate recognition BLOB region 1 corresponding to BlobCount2 is the simulated light target.

[0118] The target recognition results obtained in this experiment are as Figure 8 shown.

[0119] As Figure 6 、 Figure 8 shown, the simulated light target is on the left, the center of the simulated light target is (139.4, 155.1), and the recognition flag MngFlag is set to 1. The main laser target is on the right, the center of the main laser target is (425.0, 273.6), and the recognition flag MngFlag is set to 2. The target recognition results are shown in Table 4, where 1 represents the collected image, 2 represents the image after collision processing, and 3 represents the image obtained after the target recognition is completed.

[0120] Result analysis:

[0121] 1. Analysis of the repeated recognition results of multiple optical targets based on feature extraction

[0122] To verify the effectiveness of the multi-optical target recognition algorithm based on feature extraction in this scheme for repeated recognition, this article provides Example 2: As Figure 9 shown, four additional collimation images after target recognition processing are added and compared with the images in Example 1. These 5 collimation images have the same characteristics: (1) The collimation image contains two optical targets, namely the simulated light target and the main laser target; (2) The simulated light target is a solid and irregular optical target; (3) The main laser target is a non-solid optical target with many holes in the center of the image and a zigzag edge, and there is no adhesion or connection between the bright spots in the edge region of the optical target and the high-energy center region.

[0123] For the above characteristics, aligning and collimating the image not only requires identifying which of the two targets is the analog light target and which is the main laser target, but also for the main laser target, the bright spot area at the edge and the high-energy center area of the main laser target must be regarded as the same optical target.

[0124] The optical target recognition results for 5 different aligning and collimating images are as Figure 9 and shown in Table 5, where Figure 9 the first column is the original image, the second column is the image after digital morphological processing, and the third column is the target recognition result.

[0125] Table 5

[0126]

[0127] It can be seen from Table 5 that the analog light target and the main laser target in the 5 docking images are all recognized.

[0128] In Image 1, the area of the BLOB corresponding to No. 1 is larger than the area of the BLOB corresponding to No. 2. The BlobCount of the whole image = 6. The BlobCount of the Blobs located in the two regions with the largest areas at the center, the BlobCount1 of the BLOB corresponding to No. 1 is 1, and the BlobCount2 of the BLOB corresponding to No. 2 is 5. Since BlobCount1 is less than BlobCount2, the BLOB corresponding to No. 1 is the analog light target, and the BLOB corresponding to No. 2 is the main laser target.

[0129] For Image 2, BlobCount1 = 2, BlobCount2 = 4. Since BlobCount1 is less than BlobCount2, the BLOB corresponding to No. 1 is the analog light target, and the BLOB corresponding to No. 2 is the main laser target. The detection results of Image 3 and Image 4 are the same as those of Image 1 and Image 2, that is, the BLOB corresponding to No. 1 is the analog light target, and the BLOB corresponding to No. 2 is the main laser target.

[0130] For Image 5, the area of the BLOB corresponding to No. 1 is larger than the area of the BLOB corresponding to No. 2. The BlobCount of the whole image = 10; the BlobCount of the Blobs located in the two regions with the largest areas at the center, the BlobCount1 of the BLOB corresponding to No. 1 is 9, and the BlobCount2 of the BLOB corresponding to No. 2 is 1. Since BlobCount1 is greater than BlobCount2, the BLOB corresponding to No. 1 is the main laser target, and the BLOB corresponding to No. 2 is the analog light.

[0131] 2. Processing time performance analysis of the multi-optical target recognition algorithm based on feature extraction

[0132] For the multi-optical target recognition algorithm, not only the recognition accuracy and precision of the recognition algorithm need to be ensured, but also the processing time of the multi-optical target recognition algorithm needs to meet the requirements of the large-scale device optical path automatic alignment process for the processing time. According to the large-scale device optical path automatic alignment design index, it is required that the target recognition processing time for each optical path docking and alignment image is less than 1 second. The target recognition process of an optical path docking and alignment image is divided into 4 steps: 1) Binarization processing; 2) First digital morphology processing; 3) Second digital morphology processing; 4) Target recognition. Among them, the second digital morphology processing includes two processes: dilation operation and acquisition of relevant information.

[0133] The processing times of the multi-optical target recognition algorithm for 5 optical path docking and alignment images are shown in Table 6.

[0134] Table 6 (unit: second)

[0135]

[0136] It can be seen from Table 6 that the total target recognition time for Image 1 is 0.656 seconds, among which the binarization time is 0.016 seconds, the erosion operation time is 0.234 seconds, the dilation operation time is 0.219 seconds, the acquisition time of relevant information is 0.078 seconds, and the target separation is 0.109 seconds. Among them, the most time-consuming are the erosion operation and the dilation operation, and the binarization, feature extraction, and target separation times are less time-consuming. For the other 4 images, the dilation operation takes a relatively long time, all greater than 0.15 seconds, the longest is 0.281 seconds, and the shortest is 0.187 seconds. Thus, it can be seen that reducing the time-consuming of the digital morphology processing process, that is, the erosion operation and the dilation operation, is an important measure to improve the efficiency of multi-optical target recognition and reduce the multi-optical target recognition time.

[0137] For the 5 docking and alignment images, among the 4 steps of all multi-optical target recognition, the average time-consuming of the dilation operation is the longest, which is 0.2184 seconds, and the processing times of binarization and acquisition of relevant information are the shortest, both less than 0.1 second. The shortest total time for the multi-optical target recognition of the 5 docking and alignment images is 0.344 seconds, the longest total time is 0.703 seconds, and the average total time is 0.5596 seconds.

[0138] Thus, it can be seen that the multi-optical target recognition algorithm proposed in this paper can not only accurately recognize the simulated light target and the main laser target, but also the total processing time of multi-optical target recognition is all less than 1 second, meeting the requirements of the multi-optical target recognition algorithm for the target recognition processing time in the optical path docking and alignment process of large-scale devices.

[0139] Generally speaking, this paper proposes a multi-optical target recognition algorithm based on the statistical features of the number of BLOB regions, which is used to identify the simulated optical target and the main laser target in the optical path docking and collimation process. The main steps are as follows: 1) Perform erosion operation on the binarized collimation image, so that each BLOB region in the peripheral interval of each main laser target is separated from the main laser center region, and count the coordinates C of the center positions of all BLOB regions in the whole image x2 and C y2 ; 2) Perform dilation operation with a large size on the binarized image, so that each BLOB region in the peripheral interval of the main laser target adheres to and merges with the main laser center into a larger connected region; 3) Search for the two BLOB regions with the largest areas in the whole image as the candidate target recognition BLOB regions, and extract the extended rectangular region information Region of the two candidate BLOB regions; 4) Count the coordinates of the center positions of each BLOB region and the number located in the two candidate BLOB regions Region with the largest areas. The candidate BLOB region with a smaller number is the main laser target, and the candidate BLOB region with a larger number is the simulated optical target.

[0140] The experimental results show that the multi-optical target recognition algorithm based on the statistical features of the number of BLOB regions proposed in this paper realizes the recognition of the simulated optical target and the main laser target, and the processing time is less than 1 second, meeting the requirements of the multi-optical target recognition algorithm for the target recognition processing time in the optical path docking and collimation process of large-scale devices.

Claims

1. A multi-optical target recognition method based on BLOB region quantity feature statistics, characterized in that: Step 1: Collect collimated images and perform binarization processing on the collected images; Step 2: First digital morphological processing Perform erosion operation on the image obtained after Step 1; Step 3: Second digital morphological processing Perform a dilation operation on the image obtained after completing Step 2 to obtain the full image, as well as the number N of BLOB regions in the full image and the relevant information of each BLOB region; the relevant information includes the area area i , the coordinates C of the center position xi and C yi and the length L in the X-axis direction xi and the length L in the Y-axis direction yi , where N≥3, 1≤i≤N; Step 4: Target recognition 4.1 Set the two BLOB regions with the largest areas among the N BLOB regions as candidate recognition BLOB region 1 and candidate recognition BLOB region 2, where the central position coordinates of candidate recognition BLOB region 1 are defined as C xa and C ya and the length along the X-axis is L xa and the length along the Y-axis is L ya ; the central position coordinates of candidate recognition BLOB region 2 are C xb and C yb and the length along the X-axis is L xb and the length along the Y-axis is L yb , where 1 ≤ a ≤ N, 1 ≤ b ≤ N, a ≠ b; 4.2 Based on C xa 、C ya 、L xa 、L ya 、C xb 、C yb 、L xb and L yb respectively obtain the extended rectangle region information Region1 of the candidate recognition BLOB region 1 and the extended rectangle region information Region2 of the candidate recognition BLOB region 2; 4.3 Obtain the central position coordinates C of N BLOB regions respectively xi and C yi The quantities located in the extended rectangle region information Region1 and the extended rectangle region information Region2, and perform target recognition on the simulated light target and the main laser target; The specific steps of Step 4.2 are as follows: 4.2.1 Obtain the upper left endpoint value and the lower right endpoint value of the expansion rectangle of the candidate recognition BLOB region 1 and the candidate recognition BLOB region 2 respectively: Calculate the upper left endpoint value and the lower right endpoint value of the expansion rectangle of the candidate recognition BLOB region 1: The left upper endpoint value is (C xa -L xa , C ya -L ya ); The right lower endpoint value is (C xa + L xa , C ya + L ya ); The method for obtaining the upper left endpoint value and the lower right endpoint value of the expansion rectangle of the candidate recognition BLOB region 2 is the same as the method for obtaining the upper left endpoint value and the lower right endpoint value of the expansion rectangle of the candidate recognition BLOB region 1; 4.2.2 Based on the two groups of upper left endpoint values and lower right endpoint values obtained in Step 4.2.1, draw the expansion rectangle regions of the candidate recognition BLOB region 1 and the candidate recognition BLOB region 2 respectively, and obtain the expansion rectangle region information Region1 and the expansion rectangle region information Region2; The specific steps of Step 4.3 are as follows: 4.3.1 Statistically calculate the central position coordinates C of N BLOB regions separately xi and C yi The quantity BlobCount1 located in the extended rectangle region information Region1 and the quantity BlobCount2 located in the extended rectangle region information Region2; 4.3.2 If BlobCount1 > BlobCount2, then the candidate recognition BLOB region 1 corresponding to BlobCount1 is the main laser target, and the candidate recognition BLOB region 2 corresponding to BlobCount2 is the simulated light target; if BlobCount1 < BlobCount2, then the candidate recognition BLOB region 1 corresponding to BlobCount1 is the simulated light target, and the candidate recognition BLOB region 2 corresponding to BlobCount2 is the main laser target.

2. The multi-optical target recognition method based on BLOB region quantity feature statistics according to claim 1, characterized in that: In step 3, obtain the area area i , the coordinates C of the center position xi and C yi as well as the length L in the X-axis direction xi and the length L in the Y-axis direction yi are as follows; The target object can be separated from the background through BLOB analysis to extract the object feature parameters. The feature parameter information of each BLOB region is stored in the chain code table, and the area area of each BLOB region i , the coordinates C of the center position xi and C yi as well as the length L in the X-axis direction xi and the length L in the Y-axis direction yi are expressed by the formula as follows: where endsNumber i is the number of lines in the line segment table corresponding to the chain code table of the i-th BLOB region, pPoint[k].x i is the starting position of the horizontal line segment table of the i-th BLOB region, pPoint[k + 1].x i is the ending position of the horizontal line segment table of the i-th BLOB region, pPoint[k].y i is the starting position of the vertical line segment table of the i-th BLOB region, pPoint[k + 1].y i is the ending position of the vertical line segment table of the i-th BLOB region, point_sum i is the number of pixels included in the i-th BLOB region, k represents the k-th row in the line segment table corresponding to the chain code table, is the intermediate value, defined as

Images