CN114419342A - High-resolution image multi-target multi-feature real-time extraction method based on FPGA - Google Patents
High-resolution image multi-target multi-feature real-time extraction method based on FPGA Download PDFInfo
- Publication number
- CN114419342A CN114419342A CN202210110677.5A CN202210110677A CN114419342A CN 114419342 A CN114419342 A CN 114419342A CN 202210110677 A CN202210110677 A CN 202210110677A CN 114419342 A CN114419342 A CN 114419342A
- Authority
- CN
- China
- Prior art keywords
- target
- pixel
- pixels
- label
- mark
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of aerospace or satellite remote sensing or infrared high-resolution image target real-time detection, and particularly relates to a high-resolution image multi-target multi-feature real-time extraction method based on an FPGA (field programmable gate array), which comprises the following steps of: segmenting the received high-resolution original image to obtain a plurality of pixels, carrying out binary labeling on each pixel, and labeling binary information on the highest position of each pixel to obtain a plurality of pixels with binary information; aiming at a plurality of pixels with binary information, establishing a mark table; according to the established mark table, performing connected domain marking and mark merging on the target, performing connected marking to form a target area, establishing a target characteristic set according to the target area, and extracting basic characteristic attributes of the target in the target characteristic set; and after traversing the received original image once, outputting all target areas in the original image and the basic characteristic attributes of the corresponding targets.
Description
Technical Field
The invention belongs to the technical field of aerospace or satellite remote sensing or infrared high-resolution image target real-time detection, and particularly relates to a high-resolution image multi-target multi-feature real-time extraction method based on an FPGA.
Background
The target feature extraction is a precondition of target confirmation and is a key step of target identification. The target feature extraction is to mark the discrete target area with the suppressed background as a complete target and extract target features, and relates to real-time calculation and output of connected area marks and area features. The connected domain mark and the connected domain shape and structure have close relation, the mark algorithm also contains a large amount of judgment, comparison and shift operation, although the calculation is not complex, the data interaction operation in the cache and the comparison operation among the data are numerous, the mark processing can not be carried out in the local part and the complete parallel operation, the serial processing is needed, the operation efficiency is low, and the characteristic output is delayed. Because the resolution of the current image is higher and higher, and the frame frequency is faster and faster, when a continuous high-resolution image sequence (including a space or satellite remote sensing or infrared high-resolution image) rapidly enters, a target area is communicated in real time, and the difficulty in extracting features is high, so that the method is a bottleneck point in a real-time target detection processing system.
The traditional target feature extraction method adopts connected domain marking, then carries out secondary processing on the marked connected domain, and outputs after counting each feature of the connected domain. However, the above method flow needs to traverse the image region at least twice or even many times, which seriously affects the real-time performance of target detection.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a high-resolution image multi-target multi-feature real-time extraction method based on an FPGA (field programmable gate array), and solves the problems that the existing high-resolution high-frequency image target feature extraction method is poor in real-time performance, discrete points cannot be connected into a target area by traversing an image once after the image is subjected to threshold segmentation to generate discrete target points, and the features of the target area are output immediately. The method utilizes the semi-customization characteristics of the FPGA, has the advantage that internal resources can be freely allocated, adjusts the marking and calculating method, combines the connected domain marking and the feature extraction together, marks and counts the features while the image enters, immediately outputs the target region and related features after traversing the image once, realizes the real-time extraction of the target features, effectively reduces system resources, and can meet the requirements of a high-speed real-time target detection system.
The invention provides a high-resolution image multi-target multi-feature real-time extraction method based on an FPGA (field programmable gate array), which comprises the following steps of:
segmenting the received high-resolution original image to obtain a plurality of pixels, carrying out binary labeling on each pixel, and labeling binary information on the highest position of each pixel to obtain a plurality of pixels with binary information;
aiming at a plurality of pixels with binary information, establishing a mark table; according to the established mark table, performing connected domain marking and mark merging on the target, performing connected marking to form a target area, establishing a target characteristic set according to the target area, and extracting basic characteristic attributes of the target in the target characteristic set;
and after traversing the received original image once, outputting all target areas in the original image and the basic characteristic attributes of the corresponding targets.
As one improvement of the above technical solution, the received high resolution original image is segmented to obtain a plurality of pixels, each pixel in each pixel is subjected to binary labeling, and binary information is labeled on the highest position of each pixel to obtain a plurality of pixels with binary information; the specific process comprises the following steps:
receiving an input high-resolution original image, dividing the received original image into a target point which is higher than a threshold value and a background point which is lower than the threshold value according to a set threshold value to obtain a plurality of pixels;
carrying out binary labeling on the pixel of each point in each pixel; setting the highest pixel position of a target point as 1 and the highest pixel position of a background point as 0; and marking the binary information on the highest position of the pixel of each point to obtain a plurality of pixels with the binary information, thereby realizing the binarization marking of the input original image.
As one improvement of the above technical solution, a mark table is established for a plurality of pixels with binary information; the specific process comprises the following steps:
aiming at a plurality of pixels with binary information, according to the binary information of the highest bit of the pixel, carrying out region communication on a received original image according to the raster scanning direction, adopting an 8-neighborhood communication rule, setting a communication working window, wherein a pixel p in the communication working window is a foreground pixel, and marking the foreground pixel as f (p); the pixels a, b, c and d are background pixels, the corresponding labels of each background pixel are denoted as f (a), f (b), f (c) and f (d), and the labels are stored and stored in a pre-created empty label table.
As one improvement of the above technical solution, the target connected domain mark and the label are merged according to the established label table, the connected mark is performed to form a target region, a target feature set is established according to the target region, and the basic feature attribute of the target in the target feature set is extracted; the specific process comprises the following steps:
according to the established mark table, assuming that a pixel p is a foreground pixel, marking the pixel p as a foreground pixel, and marking the pixel p as a pixel to be marked as f (p); the pixels a, b, c and d are background pixels, and each background pixel is labeled as f (a), f (b), f (c) and f (d);
judging the pixels in the connected working window according to the relation between the background pixels and the foreground pixels in the connected working window, and distributing temporary labels;
the process of allocating and judging the temporary flag (p) is as follows:
in the first case, the neighborhood is the background pixel, p is the starting point of the new connected domain, and the label is defined as flag (i + 1);
in the second case, 1-4 marked points exist in the field, but the marks are the same, the current pixel mark inherits the neighborhood mark flag (i);
in the third case, the labels of two points in the neighborhood are different, and according to the 8-neighborhood communication criterion, the two different label positions are only two: a and c are different, and d and c are different;
at this time, flag (p) is set as a small label of two different labels; in the traversing process, only under the third condition, using an established mark table flag (x), wherein the mark table cache contains mark information of 2 lines which are the previous line and the current line;
when the labels conflict, determining the foreground pixel p as a small label, and synchronously updating the large label of the previous line as the small label;
when two independent communicated regions meet, the two temporary labels at the joint are merged and distributed according to the judgment process, the two regions are connected into a whole to obtain a complete communicated region, and the label table is updated;
after one traversal, the labels are uniformly updated to the same label at the bottom of the complete connected region after the regions are connected under different conditions;
and taking the area formed after the communication mark as a target area, wherein only the target pixel has a corresponding label in the target area. According to the numerical information of the binarization information of the highest bits of the target pixels, calculating and counting various feature information of the target area, and establishing a target feature set I (f) ((E) (f), S (f), N (f), e (f));
wherein E (f) is the intensity of the target; s (f) is the size of the target; n (f) is the number of targets; e (f) is a representative point of the target;
establishing a corresponding feature set for each target label, wherein the feature set a is cached in an external memory;
caching the target feature set corresponding to each label in an external memory according to the label sequence; synchronously establishing a one-to-one mapping label mapping table for the established target feature set, so that the label mapping table corresponds to the target feature set one to one;
and when no foreground pixel exists in the 8-connected region at any point of the bottom edge of one target region, completing the connection confirmation, outputting the target region, simultaneously reading a corresponding target feature set, and extracting the basic feature attribute of the real-time target in the target feature set.
As one improvement of the above technical solution, when there is no foreground pixel in an 8-connected region at any point of the bottom edge of a target region, completing the connection confirmation, outputting the target region, reading a corresponding target feature set, and extracting a basic feature attribute of a real-time target in the target feature set; the specific process comprises the following steps:
judging whether the target area is connected or not through the following formula;
wherein, N (f)connRepresenting the number of foreground pixels in any pixel 8 connected region at the bottom row of the target region;
is any pixel c at the bottom row of the area; fg is a foreground pixel; omegacIs a target region bottom edge RowbottomIs located at (x)c,yc) 8 connected regions of pixel c; omegac∈{p(x,y)||x-xc|≤1&|y-ycLess than or equal to 1 }; wherein p (x, y) is a pixel element p located at coordinates (x, y);
N(f)connwhen the value is equal to 0, the result shows that no foreground pixel exists in the 8-connected region of any point of the bottom edge of the target region and the connection is completed, and at the moment, the last pixel is used as the referenceAnd reading the label mapping table by the label, reading the feature set in the external cache according to the label mapping table, outputting the feature set, finishing the real-time output of the target area and the feature set, and further extracting the basic feature attribute of the target.
As an improvement of the above technical solution, the basic feature attributes of the target include: intensity, size, number of objects and object representative points.
Compared with the prior art, the invention has the beneficial effects that:
1. the real-time performance of target feature extraction is strong, after the image enters a segmented target point, the marking of a connected domain and the calculation, storage and output of a plurality of feature values of the target are completed while the pixel is traversed;
2. unlike the secondary or multiple traversals of other connected labeling algorithms, the feature extraction of the invention is completed with the primary traversal of the image, so that more resources are not required to be consumed to store information such as labels or intermediate values, and the resources are saved;
3. the traversal method is not limited by the size of the image, and can adapt to the image with any size;
4. the invention has independent module functions, can process a plurality of modules in parallel and improves the processing speed.
Drawings
FIG. 1 is a flow chart of the method for extracting multiple targets and multiple features of a high-resolution image in real time based on FPGA of the present invention;
FIG. 2 is a schematic diagram of a mark table created in the FPGA-based high-resolution image multi-target multi-feature real-time extraction method of the invention;
FIG. 3 is a schematic diagram of number table transfer updating in the FPGA-based high-resolution image multi-target multi-feature real-time extraction method of the invention.
Detailed Description
The invention will now be further described with reference to the accompanying drawings.
The invention provides a high-resolution image multi-target multi-feature real-time extraction method based on an FPGA (field programmable gate array), which is a method for connecting discrete targets generated after threshold segmentation into regions and extracting region features.
As shown in fig. 1, the method includes:
segmenting the received high-resolution original image to obtain a plurality of pixels, carrying out binary labeling on each pixel, and labeling binary information on the highest position of each pixel to obtain a plurality of pixels with binary information;
specifically, an input original image is received, a target point higher than a threshold value is taken as a background point lower than the threshold value according to a set threshold value, and the received original image is segmented to obtain a plurality of pixels;
carrying out binary labeling on the pixel of each point in each pixel; setting the highest pixel position of a target point as 1 and the highest pixel position of a background point as 0; and marking the binary information on the highest position of the pixel of each point to obtain a plurality of pixels with the binary information, thereby realizing the binarization marking of the input original image.
Aiming at a plurality of pixels with binary information, establishing a mark table; according to the established mark table, performing connected domain marking and mark merging on the target, performing connected marking to form a target area, establishing a target characteristic set according to the target area, and extracting basic characteristic attributes of the target in the target characteristic set; wherein the base feature attributes of the target include: intensity, size, number of objects and object representative points. And performing secondary calculation on the strength and the size of the target in the basic characteristic attribute of the target to further obtain the centroid and the area.
Establishing a mark table aiming at a plurality of pixels with binary information; the specific process comprises the following steps:
aiming at a plurality of pixels with binary information, according to the binary information of the highest bit of the pixel, carrying out region communication on a received original image according to the raster scanning direction, adopting an 8-neighborhood communication rule, setting a communication working window, wherein a pixel p in the communication working window is a foreground pixel, and marking the foreground pixel as f (p); the pixels a, b, c and d are background pixels, the corresponding labels of each background pixel are denoted as f (a), f (b), f (c) and f (d), and the labels are stored and stored in a pre-created empty label table.
Wherein the 8-neighborhood connected criterion, i.e. the labeling process for the current pixel, is only related to the pixel states at the top left, top right and left. The connection marking method of the invention creates a mark list on the basis of the traditional connection working window, as shown in figure 2.
The specific process of marking the connected domain and combining the label of the target according to the established label table, carrying out the connected marking to form a target area, establishing a target feature set according to the target area and extracting the basic feature attribute of the target in the target feature set comprises the following steps:
according to the established mark table, assuming that a pixel p is a foreground pixel, marking the pixel p as a foreground pixel, and marking the pixel p as a pixel to be marked as f (p); the pixels a, b, c and d are background pixels, and each background pixel is labeled as f (a), f (b), f (c) and f (d); as shown in FIG. 2, LprevAnd LcurrRespectively a previous line of mark list and a current line of mark list; the temporal labels of the pixels p depend on their labels at the top left (a), top (b), left (d) and top right (c).
The process of allocating and judging the temporary flag (p) is as follows:
in the first case, the neighborhood is the background pixel, p is the starting point of the new connected domain, and the label is defined as flag (i + 1);
in the second case, 1-4 marked points exist in the field, but the marks are the same, the current pixel mark inherits the neighborhood mark flag (i);
in the third case, the labels of two points in the neighborhood are different, and according to the 8-neighborhood communication criterion, the two different label positions are only two: a and c are different, and d and c are different;
at this time, flag (p) is set as a small label of two different labels; in the traversing process, only under the third condition, using an established mark table flag (x), wherein the mark table cache contains mark information of 2 lines which are the previous line and the current line;
when the labels conflict, determining the foreground pixel p as a small label, and synchronously updating the large label of the previous line as the small label;
when two independent communicated regions meet, the two temporary labels at the joint are merged and distributed according to the judgment process, the two regions are connected into a whole to obtain a complete communicated region, and the label table is updated; by establishing and updating the label table, the merging of equivalent labels is completed while judging whether to be communicated, and the communication judgment and the label are merged in one process.
With reference to fig. 3, after one traversal, the labels are uniformly updated to the same label at the bottom of the complete connected region after the regions are connected under different conditions;
taking an area formed after the communication mark as a target area, wherein only the target has a corresponding label, calculating and counting various feature information of the target area according to numerical information of binarization information of the target pixel except the highest bit, and establishing a target feature set I (f) { E (f), S (f), N (f), e (f) };
wherein E (f) is the intensity of the target; s (f) is the size of the target; n (f) is the number of targets; e (f) is a representative point of the target;
establishing a corresponding feature set for each target label, wherein the feature set a is cached in an external memory;
caching the target feature set corresponding to each label in an external memory (external cache) according to the label sequence; synchronously establishing a one-to-one mapping label mapping table for the established target feature set, so that the label mapping table corresponds to the target feature set one to one;
and when no foreground pixel exists in the 8-connected region at any point of the bottom edge of one target region, completing the connection confirmation, outputting the target region, simultaneously reading a corresponding target feature set, and extracting the basic feature attribute of the real-time target in the target feature set.
Specifically, whether the target area is connected is judged according to the following formula;
wherein, N (f)connRepresenting the number of foreground pixels in any pixel 8 connected region at the bottom row of the target region;
is any pixel c at the bottom row of the area; fg is a foreground pixel; omegacIs a target region bottom edge RowbottomIs located at (x)c,yc) 8 connected regions of pixel c; omegac∈{p(x,y)||x-xc|≤1&|y-ycLess than or equal to 1 }; wherein p (x, y) is a pixel element p located at coordinates (x, y);
N(f)connwhen the number of the foreground pixels is equal to 0, the foreground pixels do not exist in an 8-connected region of any point of the bottom edge of the target region and the connection is completed, at this time, a label mapping table is read according to the last label, a feature set in an external cache is read according to the label mapping table, the feature set is output, the real-time output of the target region and the feature set is completed, and further the basic feature attribute of the target is extracted.
And after traversing the received original image once, outputting all target areas in the original image and the basic characteristic attributes of the corresponding targets.
Example 1.
The invention provides a high-resolution image multi-target multi-feature real-time extraction method based on an FPGA (field programmable gate array), which comprises the following steps of:
step 1) image receiving and binaryzation labeling: receiving an input high-resolution original image, segmenting the received original image according to a set threshold value to obtain a plurality of pixels, and carrying out binary labeling on each pixel; setting the pixel higher than the threshold value as 1 as a target pixel; setting the pixel lower than the threshold value as 0, and taking the pixel as a background pixel; marking the binary result on the highest position of each pixel to finish the binary marking of the input original image; wherein, the high-resolution original image is a remote sensing or infrared high-resolution image from a space or satellite;
step 2) establishing a mark table: and according to the binarization information of the highest bit of the pixel, performing region communication on the pixel according to the raster scanning direction. The connected component labeling algorithm uses 8 neighborhood connected components criteria, i.e., the labeling process for the current pixel is only related to the pixel states at the top left, top right, and top left. The connection marking method of the invention creates a mark list on the basis of the traditional connection working window, as shown in figure 2.
According to the established mark table, assuming that a pixel p is a foreground pixel, marking the pixel p as a foreground pixel, and marking the pixel p as a pixel to be marked as f (p); the pixels a, b, c and d are background pixels, and each background pixel is labeled as f (a), f (b), f (c) and f (d); as shown in FIG. 2, LprevAnd LcurrRespectively a previous line of mark list and a current line of mark list; the temporal labels of the pixels p depend on their labels at the top left (a), top (b), left (d) and top right (c).
Step 3) merging the connected domain mark and the label: according to the mark table, the distribution states of different points in the neighborhood can be divided into three different working windows, and interpretation, marking and mark merging are respectively carried out according to the relation between the background and the foreground pixels in the working windows. The merging of equivalent labels is completed while judging whether to be communicated or not through the establishment and the update of the label table, and the communication judgment and the label are merged in one process.
In step 2), the pixel p is a pixel to be marked, LprevAnd LcurrRespectively a previous row of the mark table and a current row of the mark table. The temporal labels of the pixels p depend on their labels at the top left (a), top (b), left (d) and top right (c). According to the state of the number in the working window, the temporary mark flag (p) of p is established to be divided into three conditions:
1. the neighborhood is a background pixel, p is a new connected domain starting point, and the label of the new connected domain starting point is defined as flag (i + 1);
2. if 1-4 marked points exist in the field and the marks are the same, the current pixel mark inherits the neighborhood mark flag (i);
3. and the labels of the two points in the neighborhood are different, and according to the 8-neighborhood communication criterion, the positions of the two different labels are only two, namely a and c are different, or d and c are different. At this time, let p be the small label of the two labels; in the traversal process, only in this case, the label information table flag (x) established in step 2) is used, and the table caches the label information of the previous line and the current line which are 2 lines in total. When the labels conflict, the foreground pixels are determined as small labels, and simultaneously the large labels of the previous line are synchronously updated to the small labels.
As shown in fig. 3, after one traversal, the labels are uniformly updated to the same label at the bottom of the connected domain after the domains are connected under different conditions.
Step 4), establishing a target feature set: according to the actual requirement of target detection, an area formed after the communication marking is a target area, according to numerical information of binarization information of the target pixel except the highest bit, calculating and counting various characteristic information of the target area, and establishing a target characteristic set I (f) ((E) (f), S (f), N (f), e (f)) };
wherein E (f) is the intensity of the target; s (f) is the size of the target; n (f) is the number of targets; e (f) is a target area representative point;
specifically, the target intensity e (f): the target intensity is the sum of the total energy of the foreground pixels in the target area, and if e (i) is the energy of a single foreground pixel, the target area intensity is E (f) ═ Σp(x,y)＝fe (i); for each label, marking a foreground pixel, and accumulating the energy value of the foreground pixel into the corresponding E (f);
size of target s (f): the size of the object is the area surrounded by foreground pixels, and is defined by the upper left-hand coordinate p (x) of the areamin,ymin) And the lower right corner coordinate p (x)max,ymax) And (6) determining. For each label, each foreground pixel is communicated, the coordinates of the pixels are compared with the coordinates in the feature set, and the coordinates are updated according to the following principle;
p'(xmin)≤p(xmin) Then p (x)min)＝p'(xmin)
p'(ymin)≤p(ymin) Then p (y)min)＝p'(ymin)
p'(xmax)≥p(xmax) Then p (x)max)＝p'(xmax)
p'(ymax)≥p(ymax) Then p (y)max)＝p'(ymax)
After updating, the area locked by the two coordinates is the size of the target area.
Number of targets n (f): the number of targets is the number of foreground pixels marked by flag (i) contained in the connected region, and the calculation formula is as follows: a [ f ], | o (f) | { p (x, y) | f } |. For each label, accumulating and adding 1 to each connected foreground pixel, and obtaining the total number of foreground pixels in the target area after the connected domain is completed;
target region representative point e (f): the area after the target area is connected is large, and a representative point of the target area needs to be selected during subsequent detection processing, and the representative point can be usually the brightest point, the centroid and the like. The invention selects the brightest point as the representative point (the centroid can be calculated through the coordinates in the feature set). And for each label, judging the energy of the pixel and the energy of the feature set when each foreground pixel is communicated, and keeping the maximum energy value in the feature set.
The target feature set can contain various feature information such as the strength, the area, the size and the number of the target, target representative points, the shape, the mass center, the spatial relationship and the like according to actual requirements; the invention establishes a target basic feature set including (but not limited to) target intensity, size, quantity and representative points based on binary results. On the basis of the basic feature set, complex target feature information such as the shape, the mass center, the position relation, the relative speed and the like of the target can be further calculated;
step 5) feature set storage: caching the established target feature set in a high-speed external cache;
step 6) label mapping table establishment: synchronously establishing a label mapping table for the target feature sets established in the steps 4) and 5), so that the label mapping table and the target feature sets are in one-to-one correspondence;
step 7) feature set reading and updating: continuously reading the image, and repeating the steps 3) -6), and merging the labels of the connected domains according to the principle of the step 3) when the connected domains need to be merged. Reading a feature set from an external cache according to the label mapping table established in the step 6) to finish target labeling, label updating and feature extraction;
specifically, when two label regions in step 3) are linked together, the two labels are merged into one label. And (4) reading feature sets corresponding to the two labels from an external cache through a label mapping table, respectively calculating the content of each feature for each element of the feature sets according to the step 4), updating the result into the feature set corresponding to the small label, writing the updated feature set back to a small label storage area, completing the transmission and updating of the feature attributes, and updating the label mapping table. The large label corresponding to the storage area is discarded.
Step 8) feature output: when no foreground pixel exists in an 8-connected region at any point of the bottom edge of a target region, completing communication confirmation, outputting the target region, and simultaneously reading a target feature set and outputting the basic feature attribute of a real-time target;
judging whether the target area is connected or not through the following formula;
wherein, N (f)connRepresenting the number of foreground pixels in any pixel 8 connected region at the bottom row of the target region;
is any pixel c at the bottom row of the area; fg is a foreground pixel; omegacIs a target region bottom edge RowbottomIs located at (x)c,yc) 8 connected regions of pixel c; omegac∈{p(x,y)||x-xc|≤1&|y-ycLess than or equal to 1 }; wherein p (x, y) is a pixel element p located at coordinates (x, y);
N(f)connwhen the number of the foreground pixels is equal to 0, the foreground pixels do not exist in an 8-connected region of any point of the bottom edge of the target region and the connection is completed, at this time, a label mapping table is read according to the last label, a feature set in an external cache is read according to the label mapping table, the feature set is output, the real-time output of the target region and the feature set is completed, and further the basic feature attribute of the target is extracted.
And 9) repeating the steps 2) to 8), and finishing the extraction of all target marks of the whole image and the basic characteristic attributes of the targets.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and are not limited. Although the present invention has been described in detail with reference to the embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (6)
1. A high-resolution image multi-target multi-feature real-time extraction method based on FPGA comprises the following steps:
segmenting the received high-resolution original image to obtain a plurality of pixels, carrying out binary labeling on each pixel, and labeling binary information on the highest position of each pixel to obtain a plurality of pixels with binary information;
aiming at a plurality of pixels with binary information, establishing a mark table; according to the established mark table, performing connected domain marking and mark merging on the target, performing connected marking to form a target area, establishing a target characteristic set according to the target area, and extracting basic characteristic attributes of the target in the target characteristic set;
and after traversing the received original image once, outputting all target areas in the original image and the basic characteristic attributes of the corresponding targets.
2. The FPGA-based high-resolution image multi-target multi-feature real-time extraction method as recited in claim 1, wherein the received high-resolution original image is segmented to obtain a plurality of pixels, each pixel is subjected to binary labeling, binary information is labeled on the highest position of each pixel to obtain a plurality of pixels with binary information; the specific process comprises the following steps:
receiving an input high-resolution original image, dividing the received original image into a target point which is higher than a threshold value and a background point which is lower than the threshold value according to a set threshold value to obtain a plurality of pixels;
carrying out binary labeling on the pixel of each point in each pixel; setting the highest pixel position of a target point as 1 and the highest pixel position of a background point as 0; and marking the binary information on the highest position of the pixel of each point to obtain a plurality of pixels with the binary information, thereby realizing the binarization marking of the input original image.
3. The FPGA-based high-resolution image multi-target multi-feature real-time extraction method as claimed in claim 1, wherein a mark list is established for a plurality of pixels with binary information; the specific process comprises the following steps:
aiming at a plurality of pixels with binary information, according to the binary information of the highest bit of the pixel, carrying out region communication on a received original image according to the raster scanning direction, adopting an 8-neighborhood communication rule, setting a communication working window, wherein a pixel p in the communication working window is a foreground pixel, and marking the foreground pixel as f (p); the pixels a, b, c and d are background pixels, the corresponding labels of each background pixel are denoted as f (a), f (b), f (c) and f (d), and the labels are stored and stored in a pre-created empty label table.
4. The method for extracting the multiple targets and the multiple features of the high-resolution image based on the FPGA according to the claim 1, wherein the target connected domain mark and the mark are merged according to the established mark list to carry out the connected mark to form a target area, a target feature set is established according to the target area, and the basic feature attribute of the target in the target feature set is extracted; the specific process comprises the following steps:
according to the established mark table, assuming that a pixel p is a foreground pixel, marking the pixel p as a foreground pixel, and marking the pixel p as a pixel to be marked as f (p); the pixels a, b, c and d are background pixels, and each background pixel is labeled as f (a), f (b), f (c) and f (d);
judging the pixels in the connected working window according to the relation between the background pixels and the foreground pixels in the connected working window, and distributing temporary labels;
the process of allocating and judging the temporary flag (p) is as follows:
in the first case, the neighborhood is the background pixel, p is the starting point of the new connected domain, and the label is defined as flag (i + 1);
in the second case, 1-4 marked points exist in the field, but the marks are the same, the current pixel mark inherits the neighborhood mark flag (i);
in the third case, the labels of two points in the neighborhood are different, and according to the 8-neighborhood communication criterion, the two different label positions are only two: a and c are different, and d and c are different;
at this time, flag (p) is set as a small label of two different labels; in the traversing process, only under the third condition, using an established mark table flag (x), wherein the mark table cache contains mark information of 2 lines which are the previous line and the current line;
when the labels conflict, determining the foreground pixel p as a small label, and synchronously updating the large label of the previous line as the small label;
when two independent communicated regions meet, the two temporary labels at the joint are merged and distributed according to the judgment process, the two regions are connected into a whole to obtain a complete communicated region, and the label table is updated;
after one traversal, the labels are uniformly updated to the same label at the bottom of the complete connected region after the regions are connected under different conditions;
and taking the area formed after the communication mark as a target area, wherein only the target pixel has a corresponding label in the target area. According to the numerical information of the binarization information of the highest bits of the target pixels, calculating and counting various feature information of the target area, and establishing a target feature set I (f) ((E) (f), S (f), N (f), e (f));
wherein E (f) is the intensity of the target; s (f) is the size of the target; n (f) is the number of targets; e (f) is a representative point of the target;
establishing a corresponding feature set for each target label, wherein the feature set a is cached in an external memory;
caching the target feature set corresponding to each label in an external memory according to the label sequence; synchronously establishing a label mapping table in one-to-one mapping for the established target feature set, so that the label mapping table is in one-to-one correspondence with the target feature set;
and when no foreground pixel exists in the 8-connected region at any point of the bottom edge of one target region, completing the connection confirmation, outputting the target region, simultaneously reading a corresponding target feature set, and extracting the basic feature attribute of the real-time target in the target feature set.
5. The FPGA-based high-resolution image multi-target multi-feature real-time extraction method as recited in claim 4, wherein when no foreground pixel exists in an 8-way connection region at any point of the bottom edge of a target region, the completion of the connection confirmation is performed, the target region is output, a corresponding target feature set is read, and the basic feature attributes of real-time targets in the target feature set are extracted; the specific process comprises the following steps:
judging whether the target area is connected or not through the following formula;
wherein, N (f)connRepresenting the number of foreground pixels in any pixel 8 connected region at the bottom row of the target region;
for arbitrary pixels at the bottom row of the regionc; fg is a foreground pixel; omegacIs a target region bottom edge RowbottomIs located at (x)c,yc) 8 connected regions of pixel c; omegac∈{p(x,y)||x-xc|≤1&|y-ycLess than or equal to 1 }; wherein p (x, y) is a pixel element p located at coordinates (x, y);
N(f)connwhen the number of the foreground pixels is equal to 0, the foreground pixels do not exist in an 8-connected region of any point of the bottom edge of the target region and the connection is completed, at this time, a label mapping table is read according to the last label, a feature set in an external cache is read according to the label mapping table, the feature set is output, the real-time output of the target region and the feature set is completed, and further the basic feature attribute of the target is extracted.
6. The FPGA-based high-resolution image multi-target multi-feature real-time extraction method as claimed in claim 5, wherein the basic feature attributes of the target comprise: intensity, size, number of objects and object representative points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210110677.5A CN114419342A (en) | 2022-01-29 | 2022-01-29 | High-resolution image multi-target multi-feature real-time extraction method based on FPGA |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210110677.5A CN114419342A (en) | 2022-01-29 | 2022-01-29 | High-resolution image multi-target multi-feature real-time extraction method based on FPGA |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114419342A true CN114419342A (en) | 2022-04-29 |
Family
ID=81278363
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210110677.5A Pending CN114419342A (en) | 2022-01-29 | 2022-01-29 | High-resolution image multi-target multi-feature real-time extraction method based on FPGA |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114419342A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115778374A (en) * | 2022-10-14 | 2023-03-14 | 广东工业大学 | Fall detection method and system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103646232A (en) * | 2013-09-30 | 2014-03-19 | 华中科技大学 | Aircraft ground moving target infrared image identification device |
| CN107194946A (en) * | 2017-05-11 | 2017-09-22 | 昆明物理研究所 | A kind of infrared obvious object detection method based on FPGA |
| US20190066294A1 (en) * | 2017-08-31 | 2019-02-28 | Shenzhen United Imaging Healthcare Co., Ltd. | System and method for image segmentation |
| US20190147597A1 (en) * | 2017-11-14 | 2019-05-16 | Boe Technology Group Co., Ltd. | Maximum connected domain marking method, target tracking method, and augmented reality/virtual reality apparatus |
| CN110321888A (en) * | 2019-03-18 | 2019-10-11 | 华中科技大学 | A kind of spaceborne infrared small target detection method based on FPGA |
| CN110533678A (en) * | 2019-07-26 | 2019-12-03 | 杭州电子科技大学 | Image processing method based on the parallel connected domain screening characteristic target of FPGA frame |
| CN112509002A (en) * | 2020-11-27 | 2021-03-16 | 山东航天电子技术研究所 | Target detection tracking method based on connected domain marker |
-
2022
- 2022-01-29 CN CN202210110677.5A patent/CN114419342A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103646232A (en) * | 2013-09-30 | 2014-03-19 | 华中科技大学 | Aircraft ground moving target infrared image identification device |
| CN107194946A (en) * | 2017-05-11 | 2017-09-22 | 昆明物理研究所 | A kind of infrared obvious object detection method based on FPGA |
| US20190066294A1 (en) * | 2017-08-31 | 2019-02-28 | Shenzhen United Imaging Healthcare Co., Ltd. | System and method for image segmentation |
| US20190147597A1 (en) * | 2017-11-14 | 2019-05-16 | Boe Technology Group Co., Ltd. | Maximum connected domain marking method, target tracking method, and augmented reality/virtual reality apparatus |
| CN110321888A (en) * | 2019-03-18 | 2019-10-11 | 华中科技大学 | A kind of spaceborne infrared small target detection method based on FPGA |
| CN110533678A (en) * | 2019-07-26 | 2019-12-03 | 杭州电子科技大学 | Image processing method based on the parallel connected domain screening characteristic target of FPGA frame |
| CN112509002A (en) * | 2020-11-27 | 2021-03-16 | 山东航天电子技术研究所 | Target detection tracking method based on connected domain marker |
Non-Patent Citations (4)
| Title |
|---|
| FEDERICO BOLELLI, MICHELE CANCILLA, LORENZO BARALDI & COSTANTINO: "Toward reliable experiments on the performance of Connected Components Labeling algorithms", 《JOURNAL OF REAL-TIME IMAGE PROCESSING》 * |
| 周海，侯晴宇，卞春江，冯水春，刘一腾: "一种FPGA实现的复杂背景红外小目标检测网络", 《北京航空航天大学学报》 * |
| 王 凯，施隆照: "基于 FPGA的快速连通区域标记算法的设计与实现", 《计算机工程与应用》 * |
| 葛春平: "一种二值图像连通区域标记的简单快速算法", 《价值工程》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115778374A (en) * | 2022-10-14 | 2023-03-14 | 广东工业大学 | Fall detection method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108986119B (en) | Image segmentation method and device, computer equipment and readable storage medium | |
| CN108765371B (en) | Segmentation method of unconventional cells in pathological section | |
| CN110196053B (en) | FPGA-based real-time field robot vision navigation method and system | |
| CN113506318B (en) | Three-dimensional target perception method under vehicle-mounted edge scene | |
| CN103473785B (en) | A kind of fast multi-target dividing method based on three-valued image clustering | |
| CN107784657A (en) | A kind of unmanned aerial vehicle remote sensing image partition method based on color space classification | |
| WO2021082507A1 (en) | Semi-automated image segmentation and data annotation method, electronic device and storage medium | |
| CN107423734B (en) | SAR image ocean target rapid detection method and device | |
| CN111461036B (en) | Real-time pedestrian detection method using background modeling to enhance data | |
| CN115661173B (en) | Remote sensing image-based land parcel segmentation method | |
| CN101727654A (en) | Method realized by parallel pipeline for performing real-time marking and identification on connected domains of point targets | |
| Cao et al. | Traffic sign detection and recognition using multi-scale fusion and prime sample attention | |
| CN110210418A (en) | A kind of SAR image Aircraft Targets detection method based on information exchange and transfer learning | |
| CN114419342A (en) | High-resolution image multi-target multi-feature real-time extraction method based on FPGA | |
| CN104680531A (en) | Connection flux statistical information extraction method and VLSI structure | |
| Zhou et al. | Building segmentation from airborne VHR images using Mask R-CNN | |
| CN112487537A (en) | Building surface multistage optimization extraction method based on full convolution neural network | |
| CN110634142B (en) | Complex vehicle road image boundary optimization method | |
| CN113971723A (en) | Method, device, equipment and storage medium for constructing three-dimensional map in high-precision map | |
| CN103093241B (en) | Based on the remote sensing image nonuniformity cloud layer method of discrimination of homogeneity process | |
| CN113435266A (en) | FCOS intelligent target detection method based on extreme point feature enhancement | |
| CN115937719A (en) | Runway contour line obtaining method based on maximum area | |
| Song et al. | Image tracking and matching algorithm of semi-dense optical flow method | |
| CN115239963A (en) | Connected region marking flow processor applied to space target monitoring scene | |
| CN115100099A (en) | Point cloud data processing method, device, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |