Anti-interference one-way channel monitoring system and method based on infrared image recognition

Abstract

The invention relates to the technical field of infrared image data recognition, in particular to an anti-interference one-way channel monitoring system and method based on infrared image recognition, according to the method, when a car is free of people and objects and a door A and a door B are closed, infrared images are shot as reference images, and real-time infrared images are shot after the door A and the door B are opened and closed once during operation; preprocessing the two images to obtain a gray scale reference image and a gray scale image, dividing the gray scale image into sub-regions, setting an importance weight and a minimum significant area, comparing the two gray scale images, dividing foreground and background pixels, forming a binary foreground mask, identifying a connected domain, calculating the area of the connected domain and the overlapping area of the connected domain and the sub-region, and obtaining a binary foreground mask; if all the overlapping areas do not exceed the minimum significant area, the areas of the connected domains in all the sub-areas are multiplied by weights and summed to obtain scores, all the scores are added to obtain a scene total score, the scene total score is compared with a threshold value, and an alarm is given if the sum exceeds the threshold value; and if the overlapping area exceeding the minimum significant area exists, directly alarming. According to the invention, the monitoring alarm result is accurate, and false alarms are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infrared image data recognition, in particular to an anti-interference one-way channel monitoring system and method based on infrared image recognition. BACKGROUND

[0002] Currently, specific access places usually set one-way outbound channels to guide passenger flow and prevent external personnel from entering the paid area or restricted area without authorization. These channels are often designed as a closed "cabin type" structure with two doors, similar to an elevator car, and the operation process is, for example: when a person exits, enters the car through door A, door A closes, then door B opens, the person leaves, and door B closes to complete a pass.

[0003] To detect whether there are still personnel remaining in the car or whether personnel attempt to break in reversely after door B is closed, the prior art often installs a transmission light curtain sensor at the bottom of the car. However, this solution has obvious defects: if an object is placed or pasted on the side wall of the car instead of the ground, the light curtain cannot detect it; if a person uses auxiliary tools to lift his feet off the ground or climbs the side wall, the light curtain cannot detect his presence either. This may cause the person to illegally enter the station before door A opens after door B closes, posing a serious safety hazard.

[0004] In addition, if image recognition technology is used, a direct visible light image recognition scheme is easily disturbed by the external environment of the car, leading to misjudgment. SUMMARY

[0005] To solve the technical problems in the background art, the present application provides an anti-interference one-way channel monitoring system and method based on infrared image recognition.

[0006] The technical solution of the present application is as follows:

[0007] The anti-interference one-way channel monitoring method based on infrared image recognition is applicable to a scenario where door A and door B are respectively arranged on opposite faces in a closed space to achieve one-way passage, and the closed space is defined as a car, comprising the following steps:

[0008] S1, when there is no personnel and object in the car, and door A and door B are both in a closed state, an infrared image of the inside of the car is taken as a reference image.

[0009] S2, during operation, after door A and door B are opened and closed once, an infrared image of the inside of the car is taken when door B is completely closed and door A has not been opened again.

[0010] S3, respectively, the reference image, the real-time infrared image is pretreated to obtain the gray reference image and the gray image, the gray image is divided into a plurality of sub-regions with different attention priorities, and an importance weight and a minimum significant area are set for each sub-region;

[0011] The gray image is compared and analyzed with the gray reference image, the foreground pixels and the background pixels are divided, and all the foreground pixels constitute a binary foreground mask; the binary foreground mask is morphologically processed, the connected domains are recognized, and the areas thereof are calculated, and the overlapping areas of the connected domains and each sub-region where the connected domains are located are calculated;

[0012] If all the overlapping areas do not exceed the minimum significant area, the areas of the connected domains in all the sub-regions where the connected domains are located are multiplied by the importance weight of the sub-region and added to obtain the weighted change contribution score of the connected domain;

[0013] The weighted change contribution scores of all the connected domains in the gray image are added to obtain the scene weighted change total score, and the scene weighted change total score is compared with the final change alarm threshold value, if the scene weighted change total score exceeds the final change alarm threshold value, it is judged that there is personnel or article retention, and an alarm is issued;

[0014] If there is an overlapping area exceeding the minimum significant area, an alarm is issued.

[0015] After the alarm is processed, the reference image is forcibly updated, when there is no personnel and article in the car, and doors A and B are in the closed state, an infrared image of the inside of the car is shot as the reference image;

[0016] The temperature of the car is obtained in real time, if the temperature change value exceeds the temperature difference threshold value in a preset unit time, the reference image environment is updated, when there is no personnel and article in the car, and doors A and B are in the closed state, an infrared image of the inside of the car is shot as the reference image;

[0017] If there is no alarm in a preset time domain, and the temperature change value does not exceed the temperature difference threshold value, the reference image time domain is updated in the next time domain, when there is no personnel or article in and out of the car in a preset unit time, and there is no personnel and article in the car, and doors A and B are in the closed state, an infrared image of the inside of the car is shot as the reference image.

[0018] In S3, the gray image is compared and analyzed with the gray reference image, and the foreground pixels and the background pixels are divided, and the specific operation is as follows: the gray value difference degree of the corresponding pixel positions of the gray image and the gray reference image is compared by using the Euclidean distance, if the gray value difference degree exceeds a preset threshold value, the pixel is determined as a foreground pixel; otherwise, it is a background pixel, and all the foreground pixels constitute an initial binary foreground mask.

[0019] The gray-scale image is divided into a plurality of sub-regions with different attention priorities in S3. The specific operation is to set different non-overlapping circular rings with different inner diameters as sub-regions with different attention priorities, with the center point of the car in the gray-scale image as the center.

[0020] The preprocessing in S3 includes converting the gray-scale reference image and the real-time infrared image into a single-channel gray-scale image and applying adaptive histogram equalization.

[0021] The morphological processing in S3 includes performing an opening operation on the binary foreground mask and then performing a closing operation.

[0022] The anti-interference one-way channel monitoring system based on infrared image recognition is used to implement the anti-interference one-way channel monitoring method based on infrared image recognition, and includes:

[0023] The infrared camera is arranged at the center of the top of the car and is used to shoot the reference image and the real-time infrared image.

[0024] The image analysis module is used to preprocess the real-time infrared image to obtain a gray-scale image, divide the gray-scale image into a plurality of sub-regions with different attention priorities, and set an importance weight and a minimum significant area for each sub-region. The reference image and the gray-scale image are compared and analyzed to determine foreground pixels and background pixels, and all foreground pixels form an initial binary foreground mask. The binary foreground mask is subjected to morphological processing, connected domains are identified, and the areas of the connected domains are calculated. If the minimum significant area threshold is not exceeded, the areas of the connected domains in all sub-regions are multiplied by the importance weight of the sub-regions and summed to obtain a weighted change contribution score of the connected domain. The weighted contribution scores of all connected domains in the gray-scale image are summed to obtain a scene weighted change total score. The total score is compared with a final change alarm threshold to determine whether there is a person or object left behind.

[0025] The alarm module issues an alarm if there is a person or object left behind according to the result determined by the image analysis module.

[0026] The temperature sensor is used to obtain the temperature inside the car in real time.

[0027] The infrared camera adopts a fisheye lens. Further, other wide-angle lenses can also be used to ensure that all inner walls and floor areas of the car are shot without dead angles.

[0028] The monitoring system further includes an infrared light transmission filter installed in front of the lens of the infrared camera, which only allows infrared light of a specific waveband to pass into the camera sensor.

[0029] If the material forming the closed space is a light-transmitting material, an infrared filter film is pasted or integrated on the surface of the material, and the filter film can effectively block the infrared light from the outside of the car into the inside of the car.

[0030] The monitoring system further comprises at least one infrared light source installed in the inside of the car, which actively emits infrared light to illuminate the entire inside space of the car.

[0031] The present application has the following advantages:

[0032] The present application can detect targets exceeding the minimum significant area, and can also calculate the total score of the scene weighted change and compare it with the final change alarm threshold when all the overlapping areas do not exceed the minimum significant area. This double-triggering logic can ensure accurate identification of personnel and objects, reduce false alarms, and improve the accuracy of monitoring by combining the forced updating of the reference image, the environmental updating of the reference image, and the time-domain updating of the reference image.

[0033] Through the fisheye infrared camera and the infrared light source inside the car, the four walls and the floor of the car can be monitored without dead angles, effectively detecting personnel who try to evade the bottom sensor by climbing, hanging, etc., and objects placed or pasted on the side wall. The camera only receives the infrared light reflected by the objects inside the car, effectively shielding the interference of visible light and infrared light outside the car, ensuring the accuracy of image recognition in a transparent glass environment.

[0034] The detection blind area of the traditional light curtain sensor is effectively compensated, preventing illegal intrusion of personnel and improving the safety management level of the one-way passage. BRIEF DESCRIPTION OF DRAWINGS

[0035] In the drawings:

[0036] Figure 1 A schematic diagram of dividing a grayscale image into multiple sub-regions. DETAILED DESCRIPTION

[0037] The technical scheme of the present application is as follows:

[0038] The anti-interference one-way passage monitoring method based on infrared image recognition is suitable for a scene where doors A and B are respectively arranged on opposite faces of a closed space to realize one-way passage, and the closed space is defined as a car, comprising the following steps:

[0039] S1. When there is no personnel and object inside the car, and doors A and B are both in the closed state, an infrared image of the inside of the car is taken as a reference image.

[0040] S2. During operation, after doors A and B are opened and closed once respectively, an infrared image of the real-time inside of the car is taken when door B is completely closed and door A has not been opened again.

[0041] A complete running process of the experiment of the present application is as follows: a. Door A is opened, and door B is closed, allowing personnel to enter the car. b. After the personnel enter the car, door A is closed, and preferably, it is confirmed that the passenger has entered at this time through a traditional sensor or preliminary image detection. c. Door B is opened, allowing personnel to leave the car to the outside of the station. e. After the personnel leave the car, door B is closed.

[0042] The precise image acquisition opportunity is selected at the moment when door B is closed and door A has not been opened again for image acquisition and comparison. At this time, the car forms a completely closed and stable space, which eliminates the influence of door movement and external environment on image consistency, ensuring the reliability and accuracy of image comparison.

[0043] S3, the reference image and the real-time infrared image are pre-processed respectively to obtain a gray reference image and a gray image, the gray image is divided into a plurality of sub-regions with different attention priorities, and an importance weight and a minimum significant area are set for each sub-region;

[0044] The gray image and the gray reference image are compared and analyzed to divide foreground pixels and background pixels, and all foreground pixels constitute a binary foreground mask; the binary foreground mask is subjected to morphological processing to identify connected domains and calculate their areas, and the overlapping areas of the connected domains and each connected domain in the sub-region are calculated;

[0045] If all overlapping areas do not exceed the minimum significant area, the area of the connected domain in all sub-regions is multiplied by the pre-set importance weight of the sub-region and added to obtain the weighted change contribution score of the connected domain;

[0046] The weighted change contribution scores of all connected domains in the gray image are added to obtain a scene weighted change total score, and the scene weighted change total score is compared with a final change alarm threshold value, and if it exceeds the final change alarm threshold value, it is judged that there is personnel or article retention, and an alarm is issued;

[0047] If there is an overlapping area that exceeds the minimum significant area, an alarm is issued.

[0048] If the scene weighted change total score is compared with the final change alarm threshold value and does not exceed the final change alarm threshold value, door A is opened to allow the next person to enter the car.

[0049] In this step, the reference image and the real-time infrared image are converted into gray images suitable for subsequent comparison and analysis, and the contrast of the images is enhanced to more accurately identify the changed area.

[0050] Reference images and real-time infrared images may be in YUV format, converted to a standard color space such as BGR, and then typically converted to single-channel grayscale images to reduce computational complexity and focus on brightness variation information.

[0051] Specifically, the grayscale reference image and real-time infrared image are converted into a single-channel grayscale reference image and a grayscale image, and adaptive histogram equalization is applied.

[0052] When contrast is limited, adaptive histogram equalization is used to locally enhance the contrast of the grayscale reference image and the grayscale image, which improves the distinction between the foreground and background under non-uniform lighting conditions in subsequent steps. The preprocessed image is easier to perform subsequent comparative analysis, thus improving the accuracy and robustness of the algorithm.

[0053] The grayscale image is divided into multiple sub-regions with different attention priorities. An importance weight and a minimum salient area are set for each sub-region to allow for differentiated processing of different regions. Specifically, non-overlapping annexes with different inner diameters are set as sub-regions with different attention priorities, with the center point of the car in the grayscale image as the center.

[0054] Furthermore, the car floor may be square or irregularly shaped. Therefore, setting the sub-area to a circular shape would be detrimental to recognizing people and objects. The car floor is typically the area where alarms occur most frequently. Therefore, the sub-area with the highest priority can be set to the shape of the car floor, such as... Figure 1 As shown, different colors represent different sub-regions. Except for the sub-region with the highest attention priority, other sub-regions are set as annular rings of different radii with the center point of the car in the grayscale image as the center. The inner ring of the sub-region adjacent to the sub-region with the highest attention priority is adapted to the shape of the sub-region with the highest attention priority.

[0055] Based on the spatial layout and focus of the car, an importance weight is assigned to each sub-region to reflect its importance in change detection. For example, the sub-regions closer to the center can be assigned a higher weight than those farther from the center. This is because an anomaly occurring in the center of the car is of higher importance than an anomaly occurring in a sub-region farther from the center, and an anomaly occurring in the center of the car is more likely to affect the overall operation of the car.

[0056] Setting a minimum salient area helps filter out areas of excessively small variation, reducing false alarms. For example, if an anomaly smaller than the minimum salient area occurs inside the car due to windblown sand or dust from passengers' shoes, no alarm needs to be triggered. Therefore, setting the minimum salient area helps filter out noise and minute changes, reducing the possibility of false alarms. Different minimum salient areas are set for different sub-regions.

[0057] The grayscale image is compared with the grayscale reference image, and foreground pixels and background pixels are divided to identify the change region in the image. Specifically, the Euclidean distance is used to compare the grayscale value difference of the corresponding pixel positions of the grayscale image and the grayscale reference image. If the grayscale value difference exceeds a preset threshold, the pixel is determined as a foreground pixel; otherwise, it is a background pixel. All foreground pixels form an initial binary foreground mask.

[0058] The comparison analysis and foreground pixel division can accurately identify the change region in the image, providing a basis for subsequent connected component identification and area calculation.

[0059] The binary foreground mask is subjected to morphological processing to remove noise and fill holes, and then the connected components are identified and their areas are calculated. Therefore, the binary foreground mask is first subjected to an opening operation, and then a closing operation. The opening operation and the closing operation can remove small noise points and fill small holes, improving the accuracy of connected component identification.

[0060] Since the image comparison is performed under specific, closed and stable lighting conditions, the complexity of the image comparison algorithm can be reduced, and the processing efficiency and identification robustness can be improved.

[0061] The accuracy of connected component identification is improved, and the area calculation is more accurate. Therefore, the overlapping area of the connected component and each sub-region where the connected component is located is calculated to determine whether an alarm is needed.

[0062] After the alarm is processed, the reference image is forcibly updated. When there is no person or object inside the car, and doors A and B are both closed, an infrared image of the inside of the car is taken as the reference image.

[0063] The temperature of the car is obtained in real time. If the temperature change value exceeds the temperature difference threshold within a preset unit of time, the reference image environment is updated. When there is no person or object inside the car, and doors A and B are both closed, an infrared image of the inside of the car is taken as the reference image.

[0064] If there is no alarm in the preset time domain, and the temperature change value does not exceed the temperature difference threshold, the reference image time domain is updated in the next time domain. When there is no person or object entering or leaving the car within a preset unit of time, and there is no person or object inside the car, and doors A and B are both closed, an infrared image of the inside of the car is taken as the reference image.

[0065] The alarm situation may still exist after being reviewed by personnel, so the reference image needs to be forcibly updated. There may also be situations where the car manager changes the structure of the car as needed, such as adding disinfection devices or posting promotional posters. When the car structure change is completed, the reference image also needs to be forcibly updated.

[0066] Since the infrared camera is used in the system of the present application, the influence of infrared radiation on imaging needs to be considered. The temperature has the greatest impact on infrared radiation. Therefore, when the temperature change value exceeds the temperature difference threshold, the reference image environment is updated. For example, the weather outside the car changes dramatically, which affects the temperature inside the car, causing the temperature inside the car to rise or fall by 5 degrees within 1 hour. At this time, the real-time infrared image will inevitably change dramatically. If the reference image continues to be applied, it may cause false alarms. Therefore, the reference image environment needs to be updated to avoid false alarms.

[0067] In addition to the above-mentioned reference image update, the present application further provides an optimized operation of reference image update and selection.

[0068] According to the air temperature change coverage range of the local area, the temperature change coverage range is divided into several intervals according to the temperature step (such as 5℃). The corresponding end value or intermediate value of each interval is taken as the representative temperature of this interval. The reference image under the representative temperature is taken as the naming of this reference image.

[0069] During operation, when the real-time infrared image of the current car interior is taken, the real-time temperature in the car is collected at the same time through the temperature sensor. Based on the real-time temperature, it is determined which reference image to use for comparison.

[0070] For example, if the real-time temperature is exactly the same as a certain representative temperature, the reference image corresponding to the representative temperature is directly selected. However, in most cases, the real-time temperature cannot be exactly the same as a certain representative temperature. In this case, the reference image corresponding to the representative temperature of the interval in which the real-time temperature falls can be used. The reference image corresponding to the representative temperature with the smallest absolute value of the difference between the real-time temperature and the representative temperature can also be selected.

[0071] In addition, the obtained temperature data is also used to control the power of the infrared light source. The power of the infrared light source is set to change with time. For example, the temperature is low in the morning and evening, and the radiation value in the car is low. The infrared light source can be set to operate at high power. Therefore, the power of the infrared light source usually changes with time.

[0072] A temperature and infrared light source basic power change curve is established. According to the real-time temperature, the power of the infrared light source is adjusted according to the temperature and infrared light source basic power change curve. However, when the temperature change value exceeds the temperature difference threshold, the following scheme is used to adjust the power of the infrared light source:

[0073] The maximum infrared light source power and the minimum infrared light source power, the minimum temperature warning value and the maximum temperature warning value are set, so that the design purpose is to prevent the infrared light source power from being overloaded or invalid, when the temperature is the minimum temperature warning value, the infrared light source power is adjusted to the maximum infrared light source power, if the infrared light source power is further increased, the equipment may be damaged, when the temperature is greater than the minimum temperature warning value, the infrared light source power returns to the infrared light source power corresponding to the infrared light source basic power change curve; when the temperature is the maximum temperature warning value, and the infrared light source power corresponding to the temperature on the infrared light source basic power change curve at this time is less than the minimum infrared light source power, the infrared light source stops working to save energy consumption.

[0074] Preferably, in order to protect the infrared light source from being damaged by a large change in infrared light source power in a short time, the application also provides a way to adjust the infrared light source power: in a unit time, when the cooling value exceeds the temperature difference threshold, the infrared light source power in the next unit time is obtained by multiplying the current infrared light source power by the first power coefficient a1, until it rises to the infrared light source power corresponding to the infrared light source basic power change curve, specifically, the first power coefficient a1 is calculated as follows:

[0075]

[0076] This scheme is to avoid the infrared light source from being damaged due to rapid increase in power caused by rapid cooling, based on the same idea, considering that the environmental infrared radiation may increase rapidly when the temperature rises by more than the temperature difference threshold in a unit time, which may cause image overexposure and affect the accuracy of monitoring results, therefore, the infrared light source power in the next unit time is obtained by multiplying the current infrared light source power by the second power coefficient a2, until it decreases to the infrared light source power corresponding to the infrared light source basic power change curve, wherein the second power coefficient a2 is calculated as follows:

[0077]

[0078] During the operation of the car, there are a large number of preset time domains without alarm and temperature change value exceeding the temperature difference threshold, but if the reference image is not updated at all, the cumulative time may cause a large-scale outbreak of false alarms, therefore, the application designs a reference image time domain update to improve the overall process, for example: if the car is without alarm and the temperature change value does not exceed the temperature difference threshold within 24 hours, and if there is no personnel or goods in and out of the car for 0.5 hours within the next 24 hours, the reference image time domain update is performed.

[0079] The anti-interference one-way channel monitoring system based on infrared image recognition is used to realize the anti-interference one-way channel monitoring method based on infrared image recognition, which comprises:

[0080] An infrared camera is arranged at the center of the top of the car to capture a reference image and a real-time infrared image;

[0081] An image analysis module pre-processes the real-time infrared image to obtain a grayscale image, divides the grayscale image into a plurality of sub-regions with different attention priorities, and sets an importance weight and a minimum significant area for each sub-region; compares the reference image and the grayscale image to determine foreground pixels and background pixels, and all foreground pixels form an initial binary foreground mask; performs morphological processing on the binary foreground mask to identify connected domains, calculate the area of each connected domain, and calculate the overlapping area of each connected domain with each sub-region; if the minimum significant area threshold is not exceeded, the area of the connected domain in all sub-regions is multiplied by the importance weight of the sub-region and summed to obtain a weighted change contribution score of the connected domain; the weighted contribution scores of all connected domains in the grayscale image are summed to obtain a scene weighted change total score, which is compared with a final change alarm threshold to determine whether there is a person or object left behind;

[0082] An alarm module issues an alarm if a person or object is left behind according to the result determined by the image analysis module;

[0083] A temperature sensor acquires the temperature inside the car in real time.

[0084] The infrared camera uses a fisheye lens, and other wide-angle lenses can also be used to ensure that all interior walls and floor areas of the car are captured without dead angles.

[0085] Through the fisheye infrared camera and the infrared light source inside the car, the four walls and the floor of the car can be monitored without dead angles, effectively detecting persons who try to evade the bottom sensor by climbing, hanging, etc., and objects placed or pasted on the side walls.

[0086] The monitoring system further comprises an infrared light-transmitting filter installed in front of the lens of the infrared camera, which only allows infrared light of a specific waveband to pass into the camera sensor.

[0087] If the material forming the enclosed space is a light-transmitting material, an infrared filter film is pasted or integrated on its surface, which can effectively block the infrared light from the outside environment of the car from entering the inside of the car. The infrared filter film is attached to the transparent side wall, which cooperates with the infrared light-transmitting filter in front of the camera and the infrared light source inside the car to form a closed infrared environment. The camera only receives the infrared light reflected by the objects inside the car, effectively shielding the interference of visible light and infrared light from the outside of the car, ensuring the accuracy of image recognition in a transparent glass environment.

[0088] The infrared camera forms an image by detecting the infrared radiation emitted by an object. All objects with a temperature higher than absolute zero will emit infrared radiation, and the intensity of the radiation is related to the temperature of the object.

[0089] Considering the environment where the present application is located may be in low temperature condition, if the temperature of the target object is lower, the infrared radiation emitted by it is weaker, which may be difficult to be captured by the infrared camera, therefore the monitoring system further comprises at least one infrared light source installed in the car interior, which actively emits infrared light to illuminate the entire car interior space.

[0090] In addition, when it is completely dark or in low light environment, the natural infrared radiation emitted by the object may not be sufficient to be clearly captured by the infrared camera, at this time the infrared light source can also play an auxiliary role for the infrared camera to take pictures.

[0091] The infrared light source can also provide uniform infrared illumination, reduce image noise, improve image contrast and clarity, and clear images can better identify the interference situation in the car.

[0092] The present application effectively compensates for the detection blind area of the traditional light curtain sensor, prevents illegal intrusion of personnel, and improves the safety management level of the one-way channel.

Claims

1. A method for monitoring a non-interference one-way passage based on infrared image recognition, which is suitable for a scenario in which a door A and a door B are respectively arranged on opposite surfaces of a closed space to achieve one-way passage, and the closed space is defined as a car, characterized in that, The method comprises the following steps: S1, when there is no person and object in the car, and the doors A and B are both closed, an infrared image of the inside of the car is taken as a reference image; S2, during operation, after the doors A and B are opened and closed once respectively, when the door B is completely closed and the door A is not opened again, a real-time infrared image of the inside of the car is taken; S3, the reference image and the real-time infrared image are preprocessed to obtain a grayscale reference image and a grayscale image, the grayscale image is divided into a plurality of sub-regions with different attention priorities, and an importance weight and a minimum significant area are set for each sub-region; The grayscale image and the grayscale reference image are compared and analyzed to divide foreground pixels and background pixels, and all the foreground pixels form a binary foreground mask; The binary foreground mask is morphologically processed to identify connected domains, and the area of each connected domain and the overlapping area of the connected domain and each sub-region where the connected domain is located are calculated; If all the overlapping areas do not exceed the minimum significant area, the area of the connected domain in all the sub-regions is multiplied by the importance weight of the sub-region and added to obtain the weighted change contribution score of the connected domain; The weighted change contribution scores of all the connected domains in the grayscale image are added to obtain a scene weighted change total score, and the scene weighted change total score is compared with a final change alarm threshold value, and if the scene weighted change total score exceeds the final change alarm threshold value, it is determined that there is a person or object left, and an alarm is issued; If there is an overlapping area that exceeds the minimum significant area, an alarm is issued.

2. The anti-interference one-way channel monitoring method based on infrared image recognition according to claim 1, S3 further comprising: After the alarm is processed, the reference image is forcibly updated, and when there is no person and object in the car and the doors A and B are both closed, an infrared image of the inside of the car is taken as a reference image; The temperature of the car is obtained in real time, and if the temperature change value exceeds the temperature difference threshold value in a preset unit of time, the reference image is updated, and when there is no person and object in the car and the doors A and B are both closed, an infrared image of the inside of the car is taken as a reference image; If there is no alarm in a preset time domain, and the temperature change value does not exceed the temperature difference threshold value, the reference image is updated in the time domain in the next time domain, and when there is no person or object in and out of the car in a preset unit of time, and there is no person and object in the car, and the doors A and B are both closed, an infrared image of the inside of the car is taken as a reference image.

3. The anti-interference one-way channel monitoring method based on infrared image recognition according to claim 1, wherein in S3, the grayscale image and the grayscale reference image are compared and analyzed to divide foreground pixels and background pixels, and the specific operation is as follows: the difference degree of the grayscale values of the corresponding pixel positions of the grayscale image and the grayscale reference image is compared by using the Euclidean distance, and if the difference degree of the grayscale values exceeds a preset threshold value, the pixel is determined as a foreground pixel; otherwise, it is a background pixel, and all the foreground pixels form an initial binary foreground mask.

4. The anti-interference one-way passage monitoring method based on infrared image recognition of claim 1, wherein the step of dividing the gray image into a plurality of sub-regions with different attention priorities in S3 is specifically implemented by setting different non-overlapping circular rings with different inner diameters as the sub-regions with different attention priorities, with the center of the car as the center of the circle.

5. The anti-interference one-way passage monitoring method based on infrared image recognition according to claim 1, wherein the preprocessing in S3 comprises: The gray reference image and the real-time infrared image are converted into a single-channel gray image, and adaptive histogram equalization is applied.

6. The anti-interference one-way passage monitoring method based on infrared image recognition of claim 1, wherein the morphological processing in S3 is specifically implemented by first performing an opening operation on the binary foreground mask and then performing a closing operation.

7. The anti-interference one-way passage monitoring system based on infrared image recognition, characterized in that, The anti-interference one-way passage monitoring method based on infrared image recognition of any one of claims 1-6 is implemented by using a monitoring system comprising: an infrared camera arranged at the center of the top of the car to capture the reference image and the real-time infrared image; an image analysis module configured to pre-process the real-time infrared image to obtain a gray image, divide the gray image into a plurality of sub-regions with different attention priorities, set an importance weight and a minimum significant area for each sub-region, compare the reference image with the gray image to determine foreground pixels and background pixels, perform morphological processing on the binary foreground mask to identify connected domains and calculate their areas, calculate the overlapping area of each connected domain with each sub-region, multiply the area of the connected domain in all sub-regions by the importance weight of each sub-region and sum them up to obtain the weighted change contribution score of the connected domain, sum the weighted contribution scores of all connected domains in the gray image to obtain the scene weighted change total score, compare the total score with a final change alarm threshold to determine whether there is a person or object left behind, and output the result to the alarm module. an alarm module configured to output an alarm if there is a person or object left behind according to the result output by the image analysis module. a temperature sensor configured to obtain the temperature inside the car in real time.

8. The anti-tamper one-way passage monitoring system based on infrared image recognition according to claim 7, characterized in that, The infrared camera uses a fisheye lens.

9. The anti-tamper one-way passage monitoring system based on infrared image recognition according to claim 7, characterized in that, The monitoring system further comprises an infrared light transmission filter installed in front of the lens of the infrared camera.

10. The anti-tamper one-way passage monitoring system based on infrared image recognition according to claim 7, characterized in that, The monitoring system further comprises at least one infrared light source installed inside the car to actively emit infrared light.

Images