A target warning method for cooling infrared detector images

Abstract

The present invention relates to a target warning method, and in particular to a target warning method of a refrigerated infrared detector image; comprising: S1, constructing a target warning background library; S2, calculating target restoration parameters, and establishing a target tracking and restoration model; S3, filtering out an outbreak warning point through a center point threshold and a threshold near the center point; S4, filtering the outbreak warning point in combination with the intensity ratio relationship near the center point to obtain an alarm point; S5, updating the alarm point to a historical alarm point tracking library, and listing it in a pre-alarm queue; S6, judging whether the alarm point is a newly-exploded alarm point; S7, checking the newly-exploded alarm point and the pre-alarm queue for duplicates, and confirming whether the newly-exploded alarm point is a re-explosion of a historically-exploded alarm point; S8, if the number of recurrences recorded by a counter exceeds 20 times, an alarm is given by a refrigerated infrared detector. The present invention can be applied to the alarm of any target; it can not only reduce the false alarm rate, but also improve the accuracy of the alarm.

Description

Technical Field

[0001] The invention relates to a target warning method, in particular to a target warning method of a refrigeration type infrared detector image. Background Art

[0002] As a sensor, infrared detectors are widely used in military, security and other fields to monitor and identify potential targets. However, infrared detectors have a series of inherent characteristics when collecting images:

[0003] 1. The images collected by the infrared detector have low contrast, unclear details and high noise.

[0004] (2) Due to the poor quality of infrared images, infrared warning systems face the challenge of high false alarm rates in practical applications.

[0005] In order to reduce image noise and false alarm rate, infrared warning filtering technology came into being. However, most of the existing technologies are aimed at specific targets and perform identification filtering based on the special characteristics of specific targets, such as the flight attitude and other information of the flying target. However, in actual use, these special parameters of the target may not be fully predicted, and when these parameter information of the target is obtained, the target may already be at a relatively close distance, which seriously reduces the effect of early warning and the effectiveness of the warning.

[0006] In addition, even if the infrared detector can detect the target, due to the limitations of the infrared image itself, even if the target can be extracted from the infrared image, the number of targets is often far greater than the actual number of targets, making it difficult to accurately identify the real target and issue an alarm. Summary of the invention

[0007] The purpose of the present invention is to solve the technical problem that the existing infrared warning filtering technology is difficult to use for different targets and cannot guarantee the early warning effect and the accuracy of the warning, and to provide a target warning method for a refrigerated infrared detector image.

[0008] To solve the above technical problems, the technical solutions provided by the present invention are as follows:

[0009] A target warning method for a refrigerated infrared detector image comprises the following steps:

[0010] S1, using a cooling infrared detector to collect images of the target area, and using the image of the previous second as the background image library;

[0011] All target warning points in each frame of the background image library are assigned an ID number to build a target warning background library;

[0012] S2, calculating the target restoration parameters based on the positional relationship between any two target warning points in the target warning background library in the continuous frame images, and establishing a target tracking and restoration model;

[0013] S3, set the center point threshold and the threshold near the center point; filter out the outbreak alarm point through the center point threshold and the threshold near the center point, and record it as P_filter(x,y);

[0014] S4, establishing an intensity ratio relationship near the center point; filtering the outbreak alarm points obtained in step S3 based on the intensity ratio relationship near the center point to obtain an alarm point, and recording it as T_ini(x,y);

[0015] S5. Establish a pre-alarm queue and a historical alarm point tracking library; update the alarm point obtained in step S4 to the historical alarm point tracking library and add it to the pre-alarm queue;

[0016] S6, according to the position of the alarm point in the current frame image, restore the position of the alarm point in the previous 50 frames of images with the help of the target restoration parameters and the target tracking restoration model obtained in step S2, and determine whether the alarm point is a new outbreak alarm point;

[0017] If it is a new alarm point, execute step S7;

[0018] If it is not a newly-erupted alarm point, the counter will record the number of recurrences of the alarm point;

[0019] S7, checking the newly erupted alarm points in step S6 with the alarm points in the pre-alarm queue to confirm whether the newly erupted alarm points are the re-eruption of the historically erupted alarm points;

[0020] If it is confirmed that the alarm point of the historical outbreak has re-emerged, the counter will record the number of recurrences of the alarm point of the historical outbreak;

[0021] If it is confirmed to be a newly-emerged alarm point, it will be updated to the historical alarm point tracking library and included in the pre-alarm queue, and the counter will record the number of recurrences of the newly-emerged alarm point;

[0022] S8. Repeat the above steps. If the number of recurrences recorded by the counter exceeds the recurrence threshold, the refrigerated infrared detector will issue an alarm.

[0023] Furthermore, step S1 is specifically as follows:

[0024] The refrigerated infrared detector collects images in the target area at a sampling frequency of 50 Hz, and uses the 50 frames of images in the previous second as the background image library;

[0025] All target alarm points in each frame of the background image library are assigned an ID number. If the distance between two alarm points in two adjacent frames of images does not exceed four pixels, the two alarm points are considered to be the same alarm point and use the same ID number to construct the target alarm background library.

[0026] Furthermore, in step S2, the specific method of calculating the target restoration parameter based on the positional relationship between any two target warning points in the target warning background library in the continuous frame image is:

[0027] In the target warning background library, select any two target warning points in two consecutive frames of images;

[0028] The positions of the two target warning points in the previous frame image are recorded as and , and their positions in the subsequent frame images are recorded as ;

[0029] Combine the positions of the two target warning points in two consecutive frames of images to calculate the target restoration parameters θ, Δx and Δy;

[0030] ; Where θ is the rotation angle of the field of view;

[0031] ; In the formula, Δx is the horizontal movement of the field of view;

[0032] ; Where Δy is the vertical movement of the field of view.

[0033] Furthermore, in step S2, the target restoration tracking model is:

[0034] ;

[0035] ;

[0036] ;

[0037] .

[0038] Furthermore, step S3 is specifically as follows:

[0039] Set the center point threshold and the threshold near the center point;

[0040] The intensity of the center point of the target warning point in the current frame and the intensity of the center point of the target warning point in the previous frame image are recorded as S(x, y) and S_pre(x, y) respectively, and the intensity difference between two adjacent frames is recorded as Diff(x, y);

[0041] The brightness mean value within the 3×3 pixel range with the target alarm point as the center point in the current frame and the brightness mean value within the 3×3 pixel range with the target alarm point as the center point in the previous frame image are recorded as S3_3(x, y) and S_pre3_3(x, y) respectively; the brightness mean difference between two adjacent frames is recorded as Diff3×3(x, y);

[0042] If Diff(x,y) is greater than the center point threshold and Diff3×3(x,y) is greater than the threshold near the center point, the target alarm point is recorded as an outbreak alarm point and is recorded as P_filter(x,y).

[0043] Further, step S4 is specifically as follows:

[0044] The brightness mean value within the 5×5 pixel range centered on the burst alarm point in the current frame and the brightness mean value within the 5×5 pixel range centered on the burst alarm point in the previous frame are recorded as S5×5(x,y) and S5_5(x,y) respectively, and the brightness mean difference between two adjacent frames is recorded as Diff_center_5×5(x,y);

[0045] The difference between the brightness of the center point of the burst alarm point in the previous frame of the current frame and the average brightness within a 5×5 pixel range centered on the burst alarm point is recorded as Diff_pre_center_5×5(x,y);

[0046] Get the difference between the average brightness value within the 3×3 pixel range centered on the burst alarm point and the average brightness value within the 5×5 pixel range of the current frame, recorded as Diff_3×3_5×5(x,y);

[0047] Get the difference between the average brightness value in the 3×3 pixel range centered on the burst alarm point and the average brightness value in the 5×5 pixel range in the previous frame, recorded as Diff_pre_3×3_5×5(x,y);

[0048] In the outbreak alarm point P_filter(x,y), if the outbreak alarm point meets:

[0049] ;

[0050] and ; then the outbreak alarm point is recorded as the alarm point and is recorded as T_ini(x,y).

[0051] Further, in step S6, the method of determining whether the alarm point is a new alarm point is as follows:

[0052] According to the target restoration parameters and target tracking restoration model in step S2, the position of the warning point in the current frame in the previous frame image is estimated and used as the estimated warning point and recorded as ;

[0053] Determine whether there is an alarm point within a 7×7 pixel range centered on the estimated alarm point in the previous frame image;

[0054] If there is an alarm point, the alarm point is taken as a historical alarm point, and steps S3 and S4 are executed to determine whether it is an outbreak alarm point;

[0055] If it is an outbreak alarm point, confirm that the alarm point in the previous frame is a new outbreak alarm point;

[0056] If it is not an outbreak alarm point, continue to extrapolate to the previous frame;

[0057] If there is no alarm point, the estimated alarm point is used as the historical alarm point, and steps S3 and S4 are executed to determine whether it is an outbreak alarm point;

[0058] If it is an outbreak alarm point, confirm that the alarm point in the previous frame is a new outbreak alarm point;

[0059] If it is not an outbreak alarm point, continue to extrapolate to the previous frame;

[0060] If no historical alarm point of the alarm point can be found in the images of the last 50 frames, and the estimated alarm point always meets the outbreak condition, then the alarm point is considered to be a new outbreak alarm point.

[0061] Furthermore, in step S5, it is also necessary to remove the old alarm points in the pre-alarm queue that no longer alarm in the current frame, and the specific method is as follows:

[0062] Calculate the target restoration parameters θ, Δx and Δy according to the position of the alarm point in the current frame and the previous frame image;

[0063] The position of the warning point in the next frame image is calculated by target restoration parameters θ, Δx and Δy;

[0064] If there is no alarm point corresponding to the alarm point in the current frame in the next frame image, it is considered that the alarm point no longer alarms and is removed from the pre-alarm queue.

[0065] Furthermore, in step S8, if the alarm point disappears and does not appear within 200 frames, it is removed from the pre-alarm queue;

[0066] If it appears again within 200 frames, continue to execute step S3.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] 1. The target warning method of the refrigerated infrared detector image provided by the present invention can be applied to the warning of any target by identifying the target in the continuous frame image and the relative movement characteristics of the infrared detector; and the target warning point is filtered through the center point threshold and the threshold near the center point, as well as the intensity ratio relationship near the center point to filter out false alarms; at the same time, the alarm point is confirmed with the help of the target restoration parameter and the target tracking and restoration model; and an alarm is issued after the alarm point recurs 20 times, which can not only reduce the false alarm rate, but also improve the accuracy of the alarm.

[0069] 2. The target warning method of the cooled infrared detector image provided by the present invention makes good use of the spatial characteristics of the continuous background target, effectively improves the accuracy of the target warning, and greatly reduces the false alarm rate; and can retain the characteristics of the original image, while the original image remains unchanged, while improving the warning capability of the image.

[0070] 3. The target warning method of the cooled infrared detector image provided by the present invention has low computational complexity and spatial complexity, is very easy to implement through software and hardware, and is suitable for application in various devices, especially in resource-constrained embedded devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a schematic diagram of a flow chart of an embodiment of the present invention;

[0072] Figure 2 is a schematic diagram of two consecutive frames of images moving in the horizontal and vertical directions according to an embodiment of the present invention;

[0073] Figure 3 is a schematic diagram of two consecutive frames of images rotating in the field of view in an embodiment of the present invention;

[0074] Figure 4 It is a schematic diagram of two consecutive frames of images rotating around a certain point in the field of view in an embodiment of the present invention;

[0075] Figure 5 This is the original picture of the scene test of the lens stabilization simulation source timed switch when the indoor simulation source is turned on;

[0076] Figure 6 This is the full alarm diagram of the lens stable simulation source timed switch scene test when the indoor simulation source is turned on;

[0077] Figure 7 This is the full-filter alarm diagram of the lens stable simulation source timed switch scene test when the indoor simulation source is turned on;

[0078] Figure 8 This is the original picture of the scene test of the lens stabilization simulation source timed switch when the indoor simulation source is turned off;

[0079] Fig. 9 It is the full alarm diagram of the scene test of lens stabilization simulation source timed switch when the indoor simulation source is turned off;

[0080] Fig.10 This is the full-filter alarm diagram of the lens stable simulation source timed switch scene test when the indoor simulation source is turned off;

[0081] Fig.11 The original image of the indoor lens rotating clockwise to simulate the source timing switch scene test;

[0082] Fig.12 This is the full alarm diagram of the indoor lens clockwise rotating simulation source timing switch scenario test;

[0083] Fig.13 This is the full-filter alarm diagram of the indoor lens clockwise rotating simulation source timing switch scenario test;

[0084] Fig.14 The original image of the indoor lens rotating counterclockwise to simulate the source timing switch scene test;

[0085] Fig.15 This is the full alarm diagram of the indoor lens counterclockwise rotation simulation source timing switch scenario test;

[0086] Fig.16 This is the full-filter alarm diagram of the indoor lens counterclockwise rotation simulation source timing switch scenario test;

[0087] Fig.17 The original image of the outdoor fireworks source igniting the rising target scene test;

[0088] Fig.18 A full warning diagram for the outdoor fireworks source igniting the rising target scenario test;

[0089] Fig.19 Fully filtered warning image for the outdoor fireworks source igniting the rising target scene test;

[0090] Fig. 20 This is the original image of the outdoor fireworks source explosion landing target scene test;

[0091] Fig.21 This is the full warning image for the outdoor fireworks source explosion and landing target scenario test;

[0092] Fig. 22 This is a fully filtered warning image for the outdoor fireworks source explosion and landing target scene test;

[0093] Fig.23 This is the original image of the double-target scene test of fireworks source explosion and car source;

[0094] Fig.24 This is the full warning image for the double-target scene test of fireworks source explosion and car source;

[0095] Fig.25 This is the fully filtered warning image for the double-target scene test of fireworks source explosion and car source;

[0096] Fig.26 This is the original image of the double-target scene test of falling fireworks and car sources;

[0097] Fig. 27 This is the full warning image for the double-target scene test of falling fireworks and car sources;

[0098] Fig.28 This is the fully filtered warning image for the dual-target scene test of falling fireworks source and car source. DETAILED DESCRIPTION

[0099] The technical solutions in the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0100] A target warning method for a refrigerated infrared detector image comprises the following steps:

[0101] S1, the refrigerated infrared detector collects images in the target area at a sampling frequency of 50 Hz, and uses the images of 50 frames in the first second as the background image library;

[0102] All target alarm points in each frame of the background image library are assigned an ID number. If the distance between two alarm points in two adjacent frames of images does not exceed four pixels, the two alarm points are considered to be the same alarm point and use the same ID number to construct the original target alarm background library.

[0103] S2. In the target warning background library, select any two target warning points in two consecutive frames of images;

[0104] The positions of the two target warning points in the previous frame image are recorded as and , and their positions in the subsequent frame images are recorded as ;

[0105] Combine the positions of the two target warning points in two consecutive frames of images to calculate the target restoration parameters θ, Δx and Δy;

[0106] ; Where θ is the rotation angle of the field of view (i.e., the infrared detector);

[0107] ; In the formula, Δx is the horizontal movement of the field of view;

[0108] ; Where Δy is the vertical movement of the field of view.

[0109] The target restoration tracking model is established through the target restoration parameters θ, Δx and Δy:

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] S3, set the center point threshold and the threshold near the center point, the center point threshold is 100, and the threshold near the center point is 25;

[0115] The intensity of the center point of the target warning point in the current frame and the intensity of the center point of the target warning point in the previous frame image are recorded as S(x, y) and S_pre(x, y) respectively, and the intensity difference between two adjacent frames is recorded as Diff(x, y);

[0116] The brightness mean value within the 3×3 pixel range with the target alarm point as the center point in the current frame and the brightness mean value within the 3×3 pixel range with the target alarm point as the center point in the previous frame image are recorded as S3_3(x, y) and S_pre3_3(x, y) respectively; the brightness mean difference between two adjacent frames is recorded as Diff3×3(x, y);

[0117] If Diff(x,y) is greater than the center point threshold, and Diff3×3(x,y) is greater than the threshold near the center point, the target alarm point is recorded as an outbreak alarm point and recorded as P_filter(x,y);

[0118] S4, record the brightness mean value within the 5×5 pixel range centered on the burst alarm point in the current frame and the brightness mean value within the 5×5 pixel range centered on the burst alarm point in the previous frame as S5×5(x,y) and S5_5(x,y), respectively, and record the brightness mean difference between two adjacent frames as Diff_center_5×5(x,y);

[0119] The difference between the brightness of the center point of the burst alarm point in the previous frame of the current frame and the average brightness within a 5×5 pixel range centered on the burst alarm point is recorded as Diff_pre_center_5×5(x,y);

[0120] Get the difference between the average brightness value within the 3×3 pixel range centered on the burst alarm point and the average brightness value within the 5×5 pixel range of the current frame, recorded as Diff_3×3_5×5(x,y);

[0121] Get the difference between the average brightness value in the 3×3 pixel range centered on the burst alarm point and the average brightness value in the 5×5 pixel range in the previous frame, recorded as Diff_pre_3×3_5×5(x,y);

[0122] In the outbreak alarm point P_filter(x,y), if the outbreak alarm point meets:

[0123] ;

[0124] and ; then the outbreak alarm point is recorded as the alarm point and is recorded as T_ini(x,y);

[0125] S5. Establish a pre-alarm queue and a historical alarm point tracking library; update the alarm point obtained in step S4 to the historical alarm point tracking library and add it to the pre-alarm queue;

[0126] S6, according to the position of the alarm point in the current frame image, restore the position of the alarm point in the previous 50 frames of images with the help of the target restoration parameters and the target tracking restoration model obtained in step S2, and determine whether the alarm point is a new outbreak alarm point;

[0127] The method for determining whether the alarm point is a new outbreak alarm point is as follows:

[0128] According to the target restoration parameters and target tracking restoration model in step S2, the position of the warning point in the current frame in the previous frame image is estimated and used as the estimated warning point and recorded as ;

[0129] Determine whether there is an alarm point within a 7×7 pixel range centered on the estimated alarm point in the previous frame image;

[0130] If there is an alarm point, the alarm point is taken as a historical alarm point, and steps S3 and S4 are executed to determine whether it is an outbreak alarm point;

[0131] If it is an outbreak alarm point, then confirm that the alarm point in the previous frame is a new outbreak alarm point, the confirmation of the alarm point is completed, and continue to confirm other alarm points in the current frame. After all alarm points are confirmed, execute step S7;

[0132] If it is not an outbreak alarm point, continue to extrapolate to the previous frame;

[0133] If there is no alarm point, the estimated alarm point is used as the historical alarm point, and steps S3 and S4 are executed to determine whether it is an outbreak alarm point;

[0134] If it is an outbreak alarm point, the alarm point in the previous frame is confirmed to be a new outbreak alarm point, the confirmation of the alarm point is completed, and other alarm points in the current frame are confirmed. After all alarm points are confirmed, step S7 is executed;

[0135] If it is not an outbreak alarm point, continue to extrapolate to the previous frame;

[0136] If no historical alarm point of the alarm point is found in the images of the recent 50 frames, and the estimated alarm point always meets the outbreak condition, the alarm point is considered to be a new outbreak alarm point, and the confirmation of the alarm point is completed. Continue to confirm other alarm points in the current frame. After all alarm points are confirmed, execute step S7;

[0137] Otherwise, it is considered that it is not a new alarm point, and the counter records the number of recurrences of the alarm point;

[0138] S7, checking the newly erupted alarm points in step S6 with the alarm points in the pre-alarm queue to confirm whether the newly erupted alarm points are the re-eruption of the historically erupted alarm points;

[0139] If it is confirmed that the alarm point of the historical outbreak has re-emerged, the counter will record the number of recurrences of the alarm point of the historical outbreak;

[0140] If it is confirmed to be a newly-emerged alarm point, it will be updated to the historical alarm point tracking library and included in the pre-alarm queue, and the counter will record the number of recurrences of the newly-emerged alarm point;

[0141] S8, repeat the above steps, if the number of recurrences recorded by the counter exceeds the recurrence threshold, the recurrence threshold is 20 times, the refrigerated infrared detector will issue an alarm;

[0142] If the alarm point disappears and does not appear within 200 frames, it is removed from the pre-alarm queue; if it appears again within 200 frames, continue to step S3.

[0143] In step S5, it is also necessary to remove the old alarm points in the pre-alarm queue that no longer alarm in the current frame, and the specific method is as follows:

[0144] Calculate the target restoration parameters θ, Δx and Δy according to the position of the alarm point in the current frame and the previous frame image;

[0145] The position of the warning point in the next frame image is calculated by target restoration parameters θ, Δx and Δy;

[0146] If there is no alarm point corresponding to the alarm point in the current frame in the next frame image, it is considered that the alarm point no longer alarms and is removed from the pre-alarm queue.

[0147] Specific examples are given below.

[0148] The specific steps of the algorithm of the present invention are:

[0149] Step 1: Record the image background target;

[0150] The background target of the image is recorded frame by frame, recorded as Tb(frame), which is used as the original basis for calculating the lens tremor parameters of the historical trajectory. The same target between the previous and next frames is judged by the continuous frame direct target movement of no more than 4 pixels, and the same target is identified by the same ID number. The sampling frequency of the image is 50Hz. The background target record of the previous second, that is, the previous 50 frames, can be retained.

[0151] Step 2, filter the alarm point T_ini(x,y);

[0152] Filter out the alarm points T_ini(x,y) that meet the independent outbreak conditions from the target alarm points in the image frame by frame;

[0153] Step 2.1, filtering the outbreak alarm point according to the center point threshold and the threshold near the center point;

[0154] The intensity of the center point of the alarm point in the current frame and the intensity of the center point of the alarm point in the previous frame are recorded as S(x, y) and S_pre(x, y) respectively, and the intensity difference between two adjacent frames is recorded as Diff(x, y);

[0155] The brightness mean value within the 3×3 pixel range with the alarm point as the center point in the current frame and the brightness mean value within the 3×3 pixel range with the alarm point as the center point in the previous frame are recorded as S3_3(x, y) and S_pre3_3(x, y) respectively; the brightness mean difference between two adjacent frames is recorded as Diff3×3(x, y);

[0156] Set the center point threshold to 100 and the threshold near the center point to 25;

[0157] If Diff(x,y) is greater than the center point threshold and Diff3×3(x,y) is greater than the threshold near the center point, the target alarm point is recorded as an outbreak alarm point and is recorded as P_filter(x,y).

[0158] Step 2.2, further filter the outbreak alarm points according to the intensity ratio relationship near the center point;

[0159] The average brightness value within the 5×5 pixel range centered on the burst alarm point in the current frame and the average brightness value within the 5×5 pixel range centered on the burst alarm point in the previous frame are recorded as S5×5(x,y) and S5_5(x,y) respectively, and the brightness mean difference between two adjacent frames is recorded as Diff_center_5×5(x,y);

[0160] The difference between the brightness of the center point of the burst alarm point in the previous frame of the current frame and the average brightness within a 5×5 pixel range centered on the burst alarm point is recorded as Diff_pre_center_5×5(x,y);

[0161] Get the difference between the average brightness value within the 3×3 pixel range centered on the burst alarm point and the average brightness value within the 5×5 pixel range of the current frame, recorded as Diff_3×3_5×5(x,y);

[0162] Get the difference between the average brightness value in the 3×3 pixel range centered on the burst alarm point and the average brightness value in the 5×5 pixel range in the previous frame, recorded as Diff_pre_3×3_5×5(x,y);

[0163] In the outbreak alarm point P_filter(x,y), if the outbreak alarm point meets:

[0164] ;

[0165] and ; then filter out the alarm point T_ini(x,y).

[0166] Step 3: Calculate target tracking restoration parameters;

[0167] by Figure 2 and Figure 3 Take the target tracking back-calculation derivation process as an example.

[0168] Assume that the field of view of the current frame is A′, the field of view of the previous frame is A, and the current target In the previous frame, the target is P(x,y), Δx represents the horizontal movement of the field of view, Δy represents the vertical movement of the field of view, and θ represents the rotation angle of the field of view. If the field of view (lens) does not rotate, Figure 2 It can be seen that the relationship between the target points in the two frames of images is as follows:

[0169] ;

[0170] ;

[0171] If the lens rotates about a certain point, causing the field of view to rotate, the rotation state is as follows Figure 3As shown in the figure, angle θ represents the angle of lens rotation, angle β represents the angle of target rotation, and θ and β constitute the overall rotation angle of the target, which we represent with α. The relationship between α, β, and θ is:

[0172] β+θ=α

[0173] Let γ represent the radial movement, then

[0174] ;

[0175] ;

[0176] ;

[0177] ;

[0178] Expand the formula for the sum and difference of two angles:

[0179] ;

[0180] ;

[0181] If formula (3) (4) is replaced by the same amount, formula (7) (8) can be converted to:

[0182] ;

[0183] ;

[0184] If the lens rotates around a point in the field of view, as follows Figure 4 As shown. (xr, yr) is the central rotation point, (x, y) is the current target point, then we can get:

[0185] ;

[0186] ;

[0187] Removing the brackets gives the following:

[0188] ;

[0189] ;

[0190] Observation available is a constant related to the rotation angle of the rotation center. Denoted as Δx, Denoted as Δy, (13)(14) is transformed into:

[0191] ;

[0192] ;

[0193] And so on to the previous frame, the formula is:

[0194] ;

[0195] ;

[0196] Substituting (17) and (18) for x' and y' in (15) and (16), we can obtain:

[0197] ;

[0198] ;

[0199] ;

[0200] ;

[0201] ;

[0202] ;

[0203] It can be seen from equations (19) and (20) that the position of the historical target point can be determined by the movement rotation angle and the horizontal and vertical movement. As long as the target's motion trajectory is determined and the movement rotation angle and the horizontal and vertical movement are further determined, the coordinates of the historical target can be calculated. In this way, most of the historical alarm points can be excluded, thereby reducing the false alarm rate.

[0204] From equations (19) and (20), we can see that the trajectory of a certain historical point can be estimated by the rotation angle and translation. If a series of background alarm trajectories can be obtained, the rotation and translation parameters of the scene can be calculated. The background alarm point and the real alarm point exist in the same scene, so the rotation and translation parameters of the background alarm point are also applicable to the real target. The background alarm point is often a static and continuous alarm point in the field of view, which is not difficult to obtain. We can obtain the rotation and translation parameters of the scene by back-calculating multiple background target points.

[0205] Assume that there are two stable target points in the scene, which are represented in the current frame as , , which are represented as (x1, y1) and (x2, y2) in the previous frame. The relationship between them is expressed by formula (19) (20):

[0206] ;

[0207] ;

[0208] ;

[0209] ;

[0210] Formula (21)-Formula (23), Formula (22)-Formula (24) can be obtained:

[0211] ;

[0212] ;

[0213] Formula (25) × (y1-y2) gives

[0214] ;

[0215] Formula (26) × (x1-x2) gives

[0216] ;

[0217] Formula (27)-Formula (28) gives:

[0218] ;

[0219] ;

[0220] ;

[0221] Substituting the obtained θ into equation (21), we get Δx

[0222] ;

[0223] Substituting the obtained θ into equation (22), we get Δy

[0224] ;

[0225] The field of view parameters obtained by equations (29), (30), and (31) can be used to calculate the alarm position of the alarm point in the current frame image in the previous frame image, and the trajectory of the alarm point can be determined. Similarly, the displacement relationship θ, Δx, Δy sequence between the current frame image and the nearly 50 frames of images can be further calculated to obtain the target tracking (determine the alarm point as an old target, further exclude the alarm point, and eliminate false alarms).

[0226] Step 4: Update the historical alarm point tracking library.

[0227] According to the displacement relationship θ, Δx, Δy between the current frame image and the previous frame image calculated in step 3, the position of the alarm point of the current frame image in the previous frame image is inferred using formulas (22) and (24), and combined with the alarm point record of the previous frame, it is determined whether the alarm point of the current frame is an outbreak alarm point. If so, the current position of the alarm point is updated to the alarm record and the tracking is continued. If not, the alarm point is recorded in the pre-alarm queue and continues to be tracked. After it continues to alarm stably, an alarm is issued. In this process, the old alarm points in the alarm queue that no longer alarm in the current frame need to be removed. To remove the old alarm points, the displacement relationship θ, Δx, Δy between the current frame image and the previous frame image calculated in step 3 needs to be used to infer the position of the old point in the new image. If the corresponding alarm point of the old point cannot be found in the new image, it is considered that the old point no longer alarms and is removed from the alarm queue.

[0228] Step 5: Confirm that the alarm point is a new outbreak alarm point.

[0229] To confirm that the alarm point is a new alarm point, it is necessary to search and eliminate it in the latest 50 frames of images.

[0230] The basic alarm point sequence for new outbreak judgment is the alarm points that meet the judgment in step 2. According to step 3, θ, Δx, Δy are calculated, and the alarm position of the alarm point of the current frame in the previous frame is calculated, recorded as (xp1, yp1). In the alarm point sequence of the previous frame, search for an alarm point in the 7×7 range near (xp1, yp1). If an alarm point exists, use the coordinates of this alarm point as a reference point, and use the outbreak condition judgment method of step 2 to determine whether it meets the outbreak condition. If it meets, this point is recorded as a outbreak point, and the confirmation of this new alarm point is completed. If it does not meet, continue to extrapolate to the previous frame of the previous frame; if no alarm point is found in the 7×7 range near (xp1, yp1) in the alarm point sequence of the previous frame, then use the outbreak condition judgment method of step 2 at the back calculation point to determine whether it meets the outbreak condition. If it meets, the outbreak count BN (BreakNumber) is incremented by 1. If the point cannot find a historical point in the last 50 images and the extrapolated point always meets the outbreak condition, then the point is considered to be a outbreak point. Complete the judgment of all new outbreaks that meet the outbreak point in sequence.

[0231] Step 6: Add the outbreak alarm point to the alarm queue.

[0232] Adding the alarm sequence point generated in step 5 to the alarm queue requires further duplication checking with previously generated alarm points. After comparison with historical alarm points, confirm whether it is a new alarm point or a re-emergence of a historical alarm point. If it is a historical alarm point, it will not be added repeatedly, and its status only needs to be changed to recurrence, and the number of continuous recurrences will start to increase; if it is a new alarm point, it will be added to the alarm queue, the status will be marked as online, and the number of recurrences will increase.

[0233] Step 7: Alarm is triggered.

[0234] The alarm does not start as soon as the alarm target is found and confirmed. Since there may be other interfering heat sources in the field of view, continuous tracking and confirmation are required. This algorithm starts the alarm after 20 frames of tracking and confirmation. In addition, since there may be obstructions such as clouds, buildings, and trees in the field of view, the alarm target is not immediately cleared from the alarm target queue after it disappears. Instead, it is cleared if it does not appear after a period of time by using a continuous counting decrement method. The continuous decrement method uses a counting decrement of 200 times. If it appears within 200 frames, it continues to be tracked.

[0235] Experimental results of the algorithm of the present invention:

[0236] In order to test the performance of the algorithm invented in this paper, we first used a cooling image sensor to collect test image samples of several heat sources in different scenes, and then performed intuitive sensory effects of image target warning filtering on these scenes (green marks indicate all warning points obtained through preliminary judgment by the image sensor, and red marks indicate the final warning points after filtering).

[0237] (1) Indoor, lens stabilization, simulated source timed switching scenario test, such as Figures 5 to 10 shown.

[0238] like Figure 5 As shown in the figure, from the original image of the indoor lens stable analog source timed switch scene map, it can be seen that this is a complex indoor scene. There are many heat source interference items caused by the ceiling lights, working equipment, and other reflections in the scene. This can be seen in the full alarm map. The bright spots circled in the green frame in the full alarm map are the alarm targets obtained by the initial judgment of the image sensor. The number of targets that have been initially judged when the analog source is turned on and off has reached more than 40. If all these initially judged targets are used as alarm targets, huge false alarms will be caused. After filtering by this algorithm, all false alarms have been filtered out. The image with the analog source turned on only retains the analog source alarm; the image with the analog source not turned on has no alarm after filtering out all false alarms. It can be seen here that this algorithm effectively achieves the effect of filtering out false alarms.

[0239] (2) Indoor, lens rotation, simulated source scene test, such as Figures 11 to 16 shown.

[0240] like Fig.11 As shown in the figure, from the original picture of the indoor lens shaking simulation source timed switch scene diagram, it can be seen that this is Figure 5 Similar indoor scenes, except Figure 5 The sub-scene with similar objects adds the rotation of the lens. Because the lens is mounted on the aircraft in the usage scenario, the shaking of the aircraft is inevitable, which causes the lens to shake. It is very necessary to add the shaking of the lens to the test. Fig.12 , Fig.15This is an experiment conducted for the clockwise and counterclockwise rotation of the lens. From the bright spots circled in the green box in the full alarm image, it can be seen that there are still many alarm targets obtained through the initial judgment of the image sensor, and the number of targets reaches more than 70. If all these initially judged targets are used as alarm targets, a large number of false alarms will be caused. After filtering by this algorithm, all false alarms have been filtered out. Fig.13 , Fig.16 As shown in Figure 2, only the analog source alarm is retained for both rotations. It can be seen that this algorithm can effectively filter out false alarms when the camera is shaking.

[0241] (3) Outdoor, fireworks source target scene test, such as Figures 17 to 22 shown.

[0242] Outdoors is the real scene of equipment use, so it is very necessary to test the effectiveness of the algorithm in outdoor scenes, such as Fig.17 and Fig. 20 As shown in the figure, it can be seen from the original picture that this group of experiments is conducted on outdoor scenes. The scenes involve high-rise buildings, stationary vehicles, etc. The target appears in these scenes, which is more in line with the usage scenario. At the same time, these scenes may also generate interference items that interfere with the target. Fig.18 and Fig.21 As shown in the figure, from the bright spots circled in the green box in the full warning image, it can be seen that there are still many warning targets obtained through the initial judgment of the image sensor, and the number of targets reaches more than 30. Fig.19 and Fig. 22 As shown in the figure, after filtering by this algorithm, all false alarms have been filtered out, leaving only the simulated source alarm. From the rise to the fall of the fireworks, this algorithm has achieved good target recognition and tracking. It can be seen that this algorithm also has a good effect of filtering out false alarms in outdoor scenes.

[0243] (4) Outdoor fireworks source and car source dual-target scene test, such as Figure 23 to Figure 28 shown.

[0244] like Fig.23 and Fig.26 As shown in the figure, it can be seen that the background scene of this experiment is the same as that of Experiment 3. Compared with Scene 3, this scene adds a car source target driving on the road; Fig.24 and Fig. 27 As shown in the figure, there are more than 30 full warning targets in the background of fireworks explosion and falling in the two groups of pictures. After filtering by this algorithm, Fig.25 As shown in Figure 2, the moving car source (upper) and fireworks source (lower) targets are detected. Fig.28 As shown in the figure, the moving car is blocked and only the fireworks source is detected, while the other false alarm targets are filtered out. It can be seen that the false alarm filtering and target detection are effectively achieved through this algorithm.

[0245] From the comprehensive test results, the number of targets initially judged by the image sensor is much larger than the actual number of targets, and the number of false alarms is quite large. After being processed by this algorithm, a large number of false alarm information is effectively filtered out, which improves the reliability of the alarm. In the experiment, it is very effective for both indoor and outdoor scenes, and can adapt to the interference caused by lens shaking. It has strong scene adaptability, and the algorithm is low in complexity, which is easy to implement in resource-constrained embedded devices.

[0246] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A target warning method for cooling infrared detector images, characterized in that: The steps include: S1, using a cooling infrared detector to collect images of the target area, and using the image of the previous second as the background image library; All target warning points in each frame of the background image library are assigned an ID number to build a target warning background library; S2, calculating the target restoration parameters based on the positional relationship between any two target warning points in the target warning background library in the continuous frame images, and establishing a target tracking and restoration model; S3, setting the center point threshold and the threshold near the center point; Filter out the outbreak alarm points through the center point threshold and the threshold near the center point, and record it as P_filter(x,y); S4, record the brightness mean value within the 5×5 pixel range centered on the burst alarm point in the current frame and the brightness mean value within the 5×5 pixel range centered on the burst alarm point in the previous frame as S5×5(x,y) and S5_5(x,y), respectively, and record the brightness mean difference between two adjacent frames as Diff_center_5×5(x,y); The difference between the brightness of the center point of the burst alarm point in the previous frame of the current frame and the average brightness within a 5×5 pixel range centered on the burst alarm point is recorded as Diff_pre_center_5×5(x,y); Get the difference between the average brightness value within the 3×3 pixel range centered on the burst alarm point and the average brightness value within the 5×5 pixel range of the current frame, recorded as Diff_3×3_5×5(x,y); Get the difference between the average brightness value in the 3×3 pixel range centered on the burst alarm point and the average brightness value in the 5×5 pixel range in the previous frame, recorded as Diff_pre_3×3_5×5(x,y); In the outbreak alarm point P_filter(x,y), if the outbreak alarm point meets: Diff_center_5×5(x,y)>2Diff_pre_center_5×5(x,y); and Diff_3×3_5×5(x,y)>2Diff_pre_3×3_5×5(x,y); then the outbreak alarm point is recorded as the alarm point and is recorded as T_ini(x,y); S5. Establish a pre-alarm queue and a historical alarm point tracking library; update the alarm point obtained in step S4 to the historical alarm point tracking library and add it to the pre-alarm queue; S6, according to the position of the alarm point in the current frame image, restore the position of the alarm point in the previous 50 frames of images with the help of the target restoration parameters and the target tracking restoration model obtained in step S2, and determine whether the alarm point is a new outbreak alarm point; If it is a new alarm point, execute step S7; If it is not a newly-erupted alarm point, the counter will record the number of recurrences of the alarm point; S7, checking the newly erupted alarm points in step S6 with the alarm points in the pre-alarm queue to confirm whether the newly erupted alarm points are the re-eruption of the historically erupted alarm points; If it is confirmed that the alarm point of the historical outbreak has re-emerged, the counter will record the number of recurrences of the alarm point of the historical outbreak; If it is confirmed to be a newly-emerged alarm point, it will be updated to the historical alarm point tracking library and included in the pre-alarm queue, and the counter will record the number of recurrences of the newly-emerged alarm point; S8. Repeat the above steps. If the number of recurrences recorded by the counter exceeds the recurrence threshold, the refrigerated infrared detector will issue an alarm.

2. The target warning method for cooling infrared detector images according to claim 1 is characterized in that: Step S1 is specifically as follows: The refrigerated infrared detector collects images in the target area at a sampling frequency of 50 Hz, and uses the 50 frames of images in the previous second as the background image library; All target alarm points in each frame of the background image library are assigned an ID number. If the distance between two alarm points in two adjacent frames of images does not exceed four pixels, the two alarm points are considered to be the same alarm point and use the same ID number to construct the target alarm background library.

3. The target warning method for cooling infrared detector images according to claim 1 is characterized in that: In step S2, the specific method of calculating the target restoration parameter based on the positional relationship between any two target warning points in the target warning background library in the continuous frame image is: In the target warning background library, select any two target warning points in two consecutive frames of images; The positions of the two target warning points in the previous frame image are recorded as (x1′, y1′) and (x2′, y2′), and their positions in the subsequent frame image are recorded as (x1, y1) and (x2, y2). Combine the positions of the two target warning points in two consecutive frames of images to calculate the target restoration parameters θ, Δx and Δy; ; Where θ is the rotation angle of the field of view; Δx=x1′cosθ-y1′sinθ-x1; where Δx is the horizontal movement of the field of view; Δy=x1′sinθ+ y1′cosθ-y1; where Δy is the vertical movement of the field of view.

4. The target warning method for cooling infrared detector images according to claim 3 is characterized in that: In step S2, the target restoration tracking model is: x1=x1′cosθ-y1′sinθ-Δx; y1=x1′sinθ+y1′cosθ-Δy; x2=x2′cosθ-y2′sinθ-Δx; y2=x2′sinθ+y2′cosθ-Δy.

5. The target warning method for cooling infrared detector images according to claim 1 is characterized in that: Step S3 is specifically as follows: Set the center point threshold and the threshold near the center point; The intensity of the center point of the target warning point in the current frame and the intensity of the center point of the target warning point in the previous frame image are recorded as S(x, y) and S_pre(x, y) respectively, and the intensity difference between two adjacent frames is recorded as Diff(x, y); The brightness mean value within the 3×3 pixel range with the target alarm point as the center point in the current frame and the brightness mean value within the 3×3 pixel range with the target alarm point as the center point in the previous frame image are recorded as S3_3(x, y) and S_pre3_3(x, y) respectively; the brightness mean difference between two adjacent frames is recorded as Diff3×3(x, y); If Diff(x,y) is greater than the center point threshold and Diff3×3(x,y) is greater than the threshold near the center point, the target alarm point is recorded as an outbreak alarm point and is recorded as P_filter(x,y).

6. The target warning method for cooling infrared detector images according to claim 1 is characterized in that: In step S6, whether the alarm point is a new alarm point is determined by: According to the target restoration parameters and the target tracking restoration model in step S2, the position of the warning point in the current frame in the previous frame image is estimated, and it is used as the estimated warning point and recorded as (xp1, yp1); Determine whether there is an alarm point within a 7×7 pixel range centered on the estimated alarm point in the previous frame image; If there is an alarm point, the alarm point is taken as a historical alarm point, and steps S3 and S4 are executed to determine whether it is an outbreak alarm point; If it is an outbreak alarm point, confirm that the alarm point in the previous frame is a new outbreak alarm point; If it is not an outbreak alarm point, continue to extrapolate to the previous frame; If there is no alarm point, the estimated alarm point is used as the historical alarm point, and steps S3 and S4 are executed to determine whether it is an outbreak alarm point; If it is an outbreak alarm point, confirm that the alarm point in the previous frame is a new outbreak alarm point; If it is not an outbreak alarm point, continue to extrapolate to the previous frame; If no historical alarm point of the alarm point can be found in the images of the last 50 frames, and the estimated alarm point always meets the outbreak condition, then the alarm point is considered to be a new outbreak alarm point.

7. The target warning method for cooling type infrared detector images according to claim 1 is characterized in that: In step S5, it is also necessary to remove the old alarm points in the pre-alarm queue that no longer alarm in the current frame, and the specific method is as follows: Calculate the target restoration parameters θ, Δx and Δy according to the position of the alarm point in the current frame and the previous frame image; The position of the warning point in the next frame image is calculated by target restoration parameters θ, Δx and Δy; If there is no alarm point corresponding to the alarm point in the current frame in the next frame image, it is considered that the alarm point no longer alarms and is removed from the pre-alarm queue.

8. The target warning method for refrigerated infrared detector images according to claim 1 is characterized in that: In step S8, if the alarm point disappears and does not appear within 200 frames, it is removed from the pre-alarm queue; If it appears again within 200 frames, continue to execute step S3.

Images