Intelligent security object recognition method and system

Abstract

The application is suitable for the field of intelligent security technology, and particularly relates to an object recognition method and system for intelligent security, which comprises the following steps: constructing an intelligent security monitoring network; obtaining real-time monitoring images through the intelligent security monitoring network, performing pedestrian detection on the real-time monitoring images, and generating infrared interference detection images; performing pixel comparison on the infrared interference detection images, and dividing dynamic areas and static areas; performing clustering analysis on the pixels in the dynamic areas, determining the gradient direction of the pixel color change, repairing the pixel color based on the gradient direction, performing pedestrian detection, and outputting the detection result. Through the analysis on the pixels in the real-time monitoring images, the dynamic areas and the static areas are divided, the pixel distribution characteristics in the dynamic areas are determined, it is determined whether there is an interference area, the pixels in the interference area are repaired, and then secondary pedestrian detection is performed, so that the detection accuracy is greatly improved, the infrared interference is effectively avoided, and the night vision detection capability is improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent security technology, and in particular relates to an object recognition method and system for intelligent security. Background Technology

[0002] Intelligent security is an advanced security system that combines modern information technology, network communication technology, and various sensor technologies. Through intelligent devices such as high-definition cameras, access control systems, alarm devices, and environmental sensors, it uses artificial intelligence algorithms to analyze and process collected data, enabling real-time monitoring and management of personnel, vehicles, and other important assets. Intelligent security not only provides 24 / 7 uninterrupted security protection but also automatically alarms and notifies relevant personnel to take action in case of abnormal situations, greatly improving security and response efficiency.

[0003] In current intelligent security identification processes, images are captured at night using infrared imaging to identify pedestrians. However, at night, intruders can use infrared interference to alter the color of the human body in the infrared image, causing pedestrian identification to fail and achieving invisibility in the infrared image. Therefore, existing intelligent security systems are prone to infrared interference, affecting the effectiveness of intelligent security. Summary of the Invention

[0004] The purpose of this invention is to provide an object recognition method for intelligent security, which aims to solve the problem that existing intelligent security systems are prone to infrared interference, thus affecting the effectiveness of intelligent security.

[0005] This invention is implemented as follows: an object recognition method for intelligent security, the method comprising: An intelligent security monitoring network is constructed, which consists of multiple security monitoring devices, with different security monitoring devices corresponding to different monitoring areas; Real-time monitoring images are acquired through an intelligent security monitoring network. Pedestrian detection is performed on the real-time monitoring images to identify the areas where pedestrians are located in the real-time monitoring images. The areas where pedestrians are located are marked, and infrared interference detection images are generated. The infrared interference detection images are pixels from other areas in the real-time monitoring images that do not contain the areas where pedestrians are located. Pixel comparison is performed on the infrared interference detection images. Pixels that change in adjacent infrared interference detection images are defined as dynamic pixels, and the remaining pixels are defined as static pixels, thus dividing the dynamic region into the static region. Cluster analysis is performed on pixels in the dynamic region to determine the gradient direction of pixel color changes. The pixel colors are then repaired based on the gradient direction. Pedestrian detection is performed on the repaired infrared interference detection image, and the detection results are output.

[0006] Preferably, the steps of acquiring real-time monitoring images through an intelligent security monitoring network, performing pedestrian detection on the real-time monitoring images, identifying the areas where pedestrians are located in the real-time monitoring images, marking the areas where pedestrians are located, and generating infrared interference detection images specifically include: Real-time monitoring images are acquired through an intelligent security monitoring network, and the monitoring area corresponding to the real-time monitoring image and the time of image acquisition are recorded. The trained pedestrian detection model is invoked to monitor pedestrians in real-time surveillance images, and the pedestrian detection results are output. Based on the position of each pedestrian in the pedestrian detection results, the real-time monitoring image is cropped according to the pedestrian's position, retaining pixels other than the pedestrians, to obtain an infrared interference detection image.

[0007] Preferably, the step of performing pixel comparison on the infrared interference detection images, defining pixels that change in adjacent infrared interference detection images as dynamic pixels and the remaining pixels as static pixels, and dividing the dynamic and static regions specifically includes: Retrieve two sets of infrared interference detection images corresponding to adjacent acquisition time points, and extract the color information of corresponding pixels in adjacent infrared interference detection images; The color information of pixels at the same location is compared and the difference between the two is measured. When the difference is greater than a preset value, the current pixel is determined to be a dynamic pixel and the remaining pixels are static pixels. By merging dynamic pixels, a connected dynamic pixel region is obtained, which is the dynamic region, and the remaining part is the static region.

[0008] Preferably, the steps of performing cluster analysis on pixels in the dynamic region to determine the gradient direction of pixel color changes, repairing pixel colors based on the gradient direction, performing pedestrian detection on the repaired infrared interference detection image, and outputting the detection results specifically include: Cluster analysis is performed on the pixels in the dynamic region to classify the pixels in each dynamic region according to the color values ​​they contain. Pixels of the same type are connected to construct pixel bands corresponding to multiple pixel types. The direction of temperature change is determined based on the positional distribution of the pixel bands, and the existence of interference areas is determined. The pixel information of pixels around the interference area is statistically analyzed, and the color information of pixels within the interference area is adjusted to obtain a repaired infrared interference detection image. Pedestrian detection is then performed, and the detection results are output.

[0009] Preferably, if a pedestrian is detected in the infrared interference detection image, an alarm is issued.

[0010] Another object of the present invention is to provide an intelligent security object recognition system, the system comprising: The network construction module is used to build an intelligent security monitoring network, which consists of multiple security monitoring devices, with different security monitoring devices corresponding to different monitoring areas; The image extraction module is used to acquire real-time monitoring images through the intelligent security monitoring network, perform pedestrian detection on the real-time monitoring images, identify the area where pedestrians are located in the real-time monitoring images, mark the area where pedestrians are located, and generate an infrared interference detection image. The infrared interference detection image consists of pixels from other areas in the real-time monitoring image that do not contain the area where pedestrians are located. The region segmentation module is used to compare pixels in infrared interference detection images. It defines pixels that change in adjacent infrared interference detection images as dynamic pixels and the remaining pixels as static pixels, thus dividing the region into dynamic and static regions. The repair detection module is used to perform cluster analysis on pixels in dynamic regions, determine the gradient direction of pixel color changes, repair pixel colors based on the gradient direction, perform pedestrian detection on the repaired infrared interference detection image, and output the detection results.

[0011] Preferably, the image extraction module includes: The image acquisition unit is used to acquire real-time monitoring images through the intelligent security monitoring network and record the monitoring area corresponding to the real-time monitoring image and the time of image acquisition. The pedestrian detection unit is used to retrieve the trained pedestrian detection model to monitor pedestrians in real-time surveillance images and output pedestrian detection results. The image cropping unit is used to crop the real-time monitoring image based on the position of each pedestrian in the pedestrian detection results, retaining pixels other than the pedestrians, to obtain an infrared interference detection image.

[0012] Preferably, the region division module includes: The pixel information extraction unit is used to retrieve two sets of infrared interference detection images corresponding to adjacent acquisition time points and extract the color information of corresponding pixels in adjacent infrared interference detection images. The first pixel classification unit is used to compare the color information of pixels at the same position and measure the difference between them. When the difference is greater than a preset value, the current pixel is determined to be a dynamic pixel and the remaining pixels are static pixels. The pixel partitioning unit is used to merge dynamic pixels to obtain a connected dynamic pixel region, which is the dynamic region, and the remaining part is the static region.

[0013] Preferably, the repair detection module includes: The second pixel classification unit is used to perform cluster analysis on pixels in the dynamic region and classify pixels in each dynamic region according to the color values ​​contained in the pixels. The interference area identification unit is used to connect pixels of the same type to construct pixel bands corresponding to multiple pixel types, determine the direction of temperature change based on the positional distribution of the pixel bands, and determine whether there is an interference area. The detection output unit is used to statistically analyze the pixel information of pixels around the interference area, adjust the color information of pixels within the interference area, obtain the repaired infrared interference detection image, perform pedestrian detection, and output the detection results.

[0014] Preferably, if a pedestrian is detected in the infrared interference detection image, an alarm is issued.

[0015] This invention provides an intelligent security object recognition method that analyzes pixels in real-time monitoring images to divide them into dynamic and static regions. Based on the pixel distribution characteristics in the dynamic region, it determines whether there is an interference area, repairs the pixels in the interference area, and performs secondary pedestrian detection, which greatly improves the detection accuracy, effectively avoids infrared interference, and enhances night vision detection capabilities. Attached Figure Description

[0016] Figure 1 A flowchart illustrating an object recognition method for intelligent security provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the steps for generating an infrared interference detection image provided in an embodiment of the present invention; Figure 3 The flowchart illustrates the steps of dividing an infrared interference detection image into dynamic and static regions by comparing pixels in adjacent infrared interference detection images, defining pixels that change in adjacent infrared interference detection images as dynamic pixels and the remaining pixels as static pixels, as provided in this embodiment of the invention. Figure 4 A flowchart illustrating the steps for repairing and detecting interference areas provided in this embodiment of the invention; Figure 5 This invention provides an architecture diagram of an intelligent security object recognition system. Figure 6 An architecture diagram of an image extraction module for detection provided in an embodiment of the present invention; Figure 7 An architecture diagram of a region partitioning module provided in an embodiment of the present invention; Figure 8 This is an architecture diagram of a repair detection module provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] like Figure 1 The diagram shows a flowchart of an object recognition method for intelligent security provided by an embodiment of the present invention. The method includes: S100, Construct an intelligent security monitoring network, which consists of multiple security monitoring devices, with different security monitoring devices corresponding to different monitoring areas.

[0019] In this step, an intelligent security monitoring network is constructed by setting up multiple security monitoring devices in the areas that need to be monitored. The monitoring devices within the intelligent security monitoring network are interconnected, with each security monitoring device corresponding to a monitoring area. Overlapping monitoring areas are allowed. The security monitoring devices have infrared capabilities and can collect infrared images.

[0020] The S200 acquires real-time monitoring images through an intelligent security monitoring network, performs pedestrian detection on the real-time monitoring images, identifies the areas where pedestrians are located in the real-time monitoring images, marks the areas where pedestrians are located, and generates an infrared interference detection image. The infrared interference detection image consists of pixels from other areas in the real-time monitoring images that do not contain the areas where pedestrians are located.

[0021] In this step, real-time monitoring images are acquired through an intelligent security monitoring network. These real-time monitoring images are infrared monitoring images collected at night and are continuously generated. For example, a security monitoring device with a frame rate of 30 frames per second generates real-time monitoring images at a rate of 30 frames per second. The images are numbered 1-30 according to time sequence. The real-time monitoring image that needs to be identified is retrieved, and a preset pedestrian detection model is used to monitor the pedestrians contained within it. The pedestrian detection model determines the position of the pedestrians in the real-time monitoring image, and the positions of the pedestrians are removed, thus obtaining the infrared interference detection image.

[0022] S300 performs pixel comparison on the infrared interference detection image, defines the pixels that change in adjacent infrared interference detection images as dynamic pixels and the remaining pixels as static pixels, and divides the dynamic region into the static region.

[0023] In this step, the infrared interference detection image is compared pixel by pixel. The infrared interference detection image corresponding to the previous set of real-time monitoring images is retrieved. The two have the same size. The pixels at the same position in the two sets of infrared interference detection images are compared. Based on the color information of the two, it is determined whether the pixels in the current infrared interference detection image are dynamic pixels. Accordingly, the pixels in the infrared interference detection image are divided into static pixels and dynamic pixels. The dynamic pixels are connected to form a closed dynamic region, and the remaining part is the static region.

[0024] The S400 performs cluster analysis on pixels in the dynamic region to determine the gradient direction of pixel color changes, repairs pixel colors based on the gradient direction, performs pedestrian detection on the repaired infrared interference detection image, and outputs the detection results.

[0025] In this step, cluster analysis is performed on the pixels in the dynamic area. Infrared interference is achieved by using an infrared light source to emit infrared light of the corresponding frequency band, thereby changing the color distribution in the real-time monitoring image and altering the image to evade pedestrian detection model recognition and achieve the purpose of avoiding night vision monitoring. After obtaining the infrared interference detection image, a closed dynamic area is taken as a detection object, and the pixels are further classified to determine the distribution position of pixels under different color categories. Based on the pixel distribution position of different colors, it is determined whether the color is interfered with by the infrared light source. For the positions with interference, repair is performed to obtain the repaired infrared interference detection image. The pedestrian detection model is then used for further detection, and the detection results are output.

[0026] like Figure 2 As shown, in a preferred embodiment of the present invention, the steps of acquiring real-time monitoring images through an intelligent security monitoring network, performing pedestrian detection on the real-time monitoring images, identifying the areas where pedestrians are located in the real-time monitoring images, marking the areas where pedestrians are located, and generating infrared interference detection images specifically include: S201 acquires real-time monitoring images through an intelligent security monitoring network and records the monitoring area corresponding to the real-time monitoring image and the time of image acquisition.

[0027] In this step, real-time monitoring images are acquired through an intelligent security monitoring network. To reduce the amount of data processing, frame extraction can be performed. That is, for real-time monitoring images, a portion of the real-time monitoring images are extracted for detection. For example, real-time monitoring images with odd numbers are used for detection. When acquiring real-time monitoring images, the corresponding monitoring area and relevant monitoring information, such as the image acquisition time, are recorded.

[0028] S202: Retrieve the trained pedestrian detection model to monitor pedestrians in the real-time surveillance image and output the pedestrian detection results.

[0029] In this step, the trained pedestrian detection model is retrieved to monitor pedestrians in the real-time surveillance image. The pedestrian detection model can be the YOLO detection model (You Only Look Once). By training and validating it, the accuracy of its detection is ensured. The pedestrian detection results are output, and the position of each pedestrian is determined in the pedestrian detection results.

[0030] S203, based on the position of each pedestrian in the pedestrian detection results, the real-time monitoring image is cropped according to the position of the pedestrian, and the pixels other than the pedestrians are retained to obtain the infrared interference detection image.

[0031] In this step, the position of each pedestrian in the pedestrian detection results is extracted. The area corresponding to each pedestrian contains multiple pixels. Since the pedestrian in this area has been detected, the pixels in the corresponding area are deleted by cropping and then filled with preset pixels. Pixels outside the pedestrian area are retained, thus obtaining the infrared interference detection image. The infrared interference detection image has the same size as the original real-time monitoring image.

[0032] like Figure 3 As shown, in a preferred embodiment of the present invention, the step of performing pixel comparison on the infrared interference detection image, defining pixels that change in adjacent infrared interference detection images as dynamic pixels and the remaining pixels as static pixels, and dividing the dynamic region into the static region, specifically includes: S301, retrieve two sets of infrared interference detection images corresponding to adjacent acquisition time points, and extract the color information of corresponding pixels in adjacent infrared interference detection images.

[0033] In this step, two sets of infrared interference detection images corresponding to adjacent acquisition time points are retrieved. For example, if the real-time monitoring images obtained in chronological order are numbered 1, 2, and 3, and the currently processed real-time monitoring image is numbered 2, then the infrared interference detection image corresponding to the current real-time monitoring image and the infrared interference detection image corresponding to the previous real-time monitoring image are retrieved simultaneously. That is, the infrared interference detection images corresponding to the real-time monitoring images numbered 1 and 2 are retrieved at the same time. 2 is the real-time monitoring image currently being processed. The color information of each pixel in the two sets of infrared interference detection images is statistically analyzed. This color information is used to determine whether there is a change in the image. The infrared interference detection images are processed in grayscale, and the pixel is represented by its grayscale value. The color information at this time is the grayscale value.

[0034] S302 compares the color information of pixels at the same position and measures the difference between them. When the difference is greater than a preset value, the current pixel is determined to be a dynamic pixel, and the remaining pixels are static pixels.

[0035] In this step, the color information of pixels at the same location is compared. For example, if a is the previous infrared interference detection image and b is the current infrared interference detection image, the color information of pixels at the same location is extracted, and the difference between grayscale values ​​is calculated. If the difference in grayscale values ​​is greater than a preset value, the current pixel is determined to be a dynamic pixel; otherwise, it is a static pixel. Alternatively, an infrared interference detection image is used for comparison when no one is passing by, i.e., when the image is unchanged. By comparing these images, dynamic pixels and static pixels are determined.

[0036] S303 merges the dynamic pixels to obtain a connected dynamic pixel region, which is the dynamic region, while the remaining part is the static region.

[0037] In this step, dynamic pixels are merged. Dynamic pixels represent areas in the image that are changing. By merging, dynamic pixels with a pixel distance less than a preset value are merged together. For example, if the distance between two adjacent dynamic pixels is less than 10 pixels, they are connected and merged into the same dynamic area. Areas outside the dynamic area are static areas.

[0038] like Figure 4 As shown, in a preferred embodiment of the present invention, the steps of performing cluster analysis on pixels in the dynamic region to determine the gradient direction of pixel color changes, repairing pixel colors based on the gradient direction, performing pedestrian detection on the repaired infrared interference detection image, and outputting the detection results specifically include: S401 performs cluster analysis on the pixels in the dynamic region, classifying the pixels in each dynamic region according to the color values ​​contained in the pixels.

[0039] In this step, cluster analysis is performed on the pixels in the dynamic region. Before this, the dynamic region is screened, and dynamic regions with an area smaller than a preset value are removed. If the number of pixels in a dynamic region is less than N, it is removed and considered a static region. The grayscale value of each pixel in the dynamic region is extracted and statistically analyzed to determine the range of pixel grayscale value distribution. For example, if 90% of the pixel grayscale values ​​are distributed between 0 and 150, this range is divided into multiple equal-width intervals, such as: 0-20, 20-40, 40-60, 60-80, 80-100, 100-120, 120-140, 140-150, then each range is a pixel type.

[0040] S402: Connect pixels of the same type to construct pixel bands corresponding to multiple pixel types. Determine the direction of temperature change based on the positional distribution of the pixel bands and determine whether there is an interference zone.

[0041] In this step, pixels of the same type are connected. Since the infrared interference source is caused by light, the intensity of the light source gradually decreases from the center to the periphery. At the interference source, multiple concentric pixel bands will be formed. The temperature change is determined according to the distribution of the pixel bands. For example, the lower the gray value, the higher the temperature of the area, and vice versa. This determines the direction of temperature change. If it is finally determined that there is a temperature change area in the dynamic area that gradually decreases from the center to the periphery, then the area is classified as the interference area.

[0042] S403: Collect pixel information of pixels around the interference area, adjust the color information of pixels within the interference area to obtain the repaired infrared interference detection image, perform pedestrian detection, and output the detection results.

[0043] In this step, the pixel information of pixels around the interference area is statistically analyzed, and the color information of multiple pixels located around the interference area is randomly selected. These multiple pixels are then randomly copied and overlaid onto each pixel within the interference area, thereby achieving random repair of the infrared interference detection image and reducing the impact of the interference area on pedestrian recognition. Alternatively, a preset pattern can be used for filling. The repaired infrared interference detection image is then imported into the pedestrian detection model for detection, and the detection results are output. If a pedestrian is detected in the infrared interference detection image, an alarm is issued.

[0044] like Figure 5 As shown, an intelligent security object recognition system is provided in an embodiment of the present invention. The system includes: The network construction module 100 is used to build an intelligent security monitoring network, which consists of multiple security monitoring devices, with different security monitoring devices corresponding to different monitoring areas.

[0045] In this system, the network construction module 100 constructs an intelligent security monitoring network. Multiple security monitoring devices are set up in the areas that need to be monitored, thereby constructing an intelligent security monitoring network. The monitoring devices within the intelligent security monitoring network are interconnected, with each security monitoring device corresponding to a monitoring area. Overlapping monitoring areas are allowed. The security monitoring devices have infrared functionality and can collect infrared images.

[0046] The image extraction module 200 is used to acquire real-time monitoring images through an intelligent security monitoring network, perform pedestrian detection on the real-time monitoring images, identify the areas where pedestrians are located in the real-time monitoring images, mark the areas where pedestrians are located, and generate an infrared interference detection image. The infrared interference detection image consists of pixels from other areas in the real-time monitoring images that do not contain the areas where pedestrians are located.

[0047] In this system, the image extraction module 200 acquires real-time monitoring images through an intelligent security monitoring network. These real-time monitoring images are infrared monitoring images collected at night. The real-time monitoring images are continuously generated images. For example, a security monitoring device with a frame rate of 30 frames per second generates real-time monitoring images at a rate of 30 frames per second. According to the time sequence, the images are numbered 1-30. The current real-time monitoring image to be identified is retrieved, and a preset pedestrian detection model is used to monitor the pedestrians contained within it. The pedestrian detection model determines the position of the pedestrian in the real-time monitoring image, and the position of the pedestrian is removed, thus obtaining the infrared interference detection image.

[0048] The region division module 300 is used to compare pixels in infrared interference detection images, define pixels that change in adjacent infrared interference detection images as dynamic pixels, and the remaining pixels as static pixels, and divide dynamic regions into static regions.

[0049] In this system, the region division module 300 performs pixel comparison on the infrared interference detection image, retrieves the infrared interference detection image corresponding to the previous set of real-time monitoring images, and the two have the same size. The pixels at the same position in the two sets of infrared interference detection images are compared, and the color information of the two is used to determine whether the pixels in the current infrared interference detection image are dynamic pixels. Based on this, the pixels in the infrared interference detection image are divided into static pixels and dynamic pixels. The dynamic pixels are connected to form a closed dynamic region, and the remaining part is the static region.

[0050] The repair detection module 400 is used to perform cluster analysis on pixels in the dynamic region, determine the gradient direction of pixel color changes, repair pixel colors based on the gradient direction, perform pedestrian detection on the repaired infrared interference detection image, and output the detection results.

[0051] In this system, the repair and detection module 400 performs cluster analysis on pixels in the dynamic area. Infrared interference is achieved by using an infrared lamp source to emit infrared light of the corresponding frequency band, thereby changing the color distribution in the real-time monitoring image and altering the image to evade pedestrian detection model recognition and achieve the purpose of avoiding night vision monitoring. After obtaining the infrared interference detection image, the pixels are further classified into a closed dynamic area as a detection object to determine the distribution position of pixels under different color categories. Based on the pixel distribution position of different colors, it is determined whether the color is interfered with by the infrared light source. For the positions with interference, repair is performed to obtain the repaired infrared interference detection image. The pedestrian detection model is then used for further detection, and the detection results are output.

[0052] like Figure 6 As shown, in a preferred embodiment of the present invention, the image extraction module 200 includes: The image acquisition unit 201 is used to acquire real-time monitoring images through the intelligent security monitoring network and record the monitoring area corresponding to the real-time monitoring image and the time of image acquisition.

[0053] In this module, the image acquisition unit 201 acquires real-time monitoring images through the intelligent security monitoring network. In order to reduce the amount of data processing, frame extraction processing can be performed. That is, for real-time monitoring images, a portion of the real-time monitoring images are extracted for detection. For example, real-time monitoring images with odd numbers are used for detection. When acquiring real-time monitoring images, the corresponding monitoring area and corresponding monitoring information, such as the image acquisition time, are recorded.

[0054] The pedestrian detection unit 202 is used to retrieve the trained pedestrian detection model to monitor pedestrians in real-time monitoring images and output pedestrian detection results.

[0055] In this module, the pedestrian detection unit 202 retrieves the trained pedestrian detection model to monitor pedestrians in the real-time monitoring image. The pedestrian detection model can be the YOLO detection model (You Only Look Once). By training and validating it, the accuracy of its detection is ensured. The pedestrian detection results are output, and the position of each pedestrian is determined in the pedestrian detection results.

[0056] The image cropping unit 203 is used to crop the real-time monitoring image based on the position of each pedestrian in the pedestrian detection result, retaining pixels other than the pedestrians, to obtain an infrared interference detection image.

[0057] In this module, the image cropping unit 203 extracts the position of each pedestrian in the pedestrian detection result. The area corresponding to each pedestrian contains multiple pixels. Since the pedestrian in this area has been detected, the pixels in the corresponding area are deleted by cropping and then filled with preset pixels. The pixels outside the pedestrian area are retained to obtain the infrared interference detection image. The infrared interference detection image has the same size as the original real-time monitoring image.

[0058] like Figure 7 As shown, in a preferred embodiment of the present invention, the region division module 300 includes: The pixel information extraction unit 301 is used to retrieve two sets of infrared interference detection images corresponding to adjacent acquisition time points and extract the color information of corresponding pixels in the adjacent infrared interference detection images.

[0059] In this module, the pixel information extraction unit 301 retrieves two sets of infrared interference detection images corresponding to adjacent acquisition time points. For example, if the real-time monitoring images obtained in chronological order are numbered 1, 2, and 3, and the currently processed real-time monitoring image is numbered 2, then the infrared interference detection image corresponding to the current real-time monitoring image and the infrared interference detection image corresponding to the previous real-time monitoring image are retrieved simultaneously. That is, the infrared interference detection images corresponding to the real-time monitoring images numbered 1 and 2 are retrieved at the same time. 2 is the real-time monitoring image currently being processed. The color information of each pixel in the two sets of infrared interference detection images is statistically analyzed. This color information is used to determine whether there is a change in the image. The infrared interference detection images are processed in grayscale, and the pixel is represented by its grayscale value. The color information at this time is the grayscale value.

[0060] The first pixel classification unit 302 is used to compare the color information of pixels at the same position and measure the difference between them. When the difference is greater than a preset value, the current pixel is determined to be a dynamic pixel and the remaining pixels are static pixels.

[0061] In this module, the first pixel classification unit 302 compares the color information of pixels at the same position. For example, if a is the previous infrared interference detection image and b is the current infrared interference detection image, the color information of pixels at the same position is extracted, and the difference between gray values ​​is calculated. If the difference between gray values ​​is greater than a preset value, the current pixel is determined to be a dynamic pixel; otherwise, it is a static pixel. Alternatively, the infrared interference detection image is used for comparison when no one is passing by, i.e., when the image is unchanged. By comparing the images, dynamic pixels and static pixels are determined.

[0062] Pixel partitioning unit 303 is used to merge dynamic pixels to obtain a connected dynamic pixel region, which is the dynamic region, and the remaining part is the static region.

[0063] In this module, the pixel partitioning unit 303 performs merging processing on dynamic pixels. Dynamic pixels represent areas in the image that change. By merging, dynamic pixels with a pixel distance lower than a preset value are merged together. For example, if the distance between two adjacent dynamic pixels is less than 10 pixels, they are connected and merged into the same dynamic area. Areas outside the dynamic area are static areas.

[0064] like Figure 8 As shown, in a preferred embodiment of the present invention, the repair detection module 400 includes: The second pixel classification unit 401 is used to perform cluster analysis on pixels in the dynamic region and classify pixels in each dynamic region according to the color values ​​contained in the pixels.

[0065] In this module, the second pixel classification unit 400 performs cluster analysis on the pixels in the dynamic region. Before this, the dynamic region is screened, and dynamic regions with an area smaller than a preset value are removed. If the number of pixels in a dynamic region is less than N, it is removed and regarded as a static region. The gray value of each pixel in the dynamic region is extracted and statistically analyzed to determine the range of pixel gray value distribution. For example, if 90% of the pixel gray values ​​are distributed between 0 and 150, this range is divided into multiple equal-width intervals, such as: 0-20, 20-40, 40-60, 60-80, 80-100, 100-120, 120-140, 140-150, then each range is a pixel type.

[0066] The interference area identification unit 402 is used to connect pixels of the same type to construct pixel bands corresponding to multiple pixel types, determine the direction of temperature change based on the positional distribution of the pixel bands, and determine whether there is an interference area.

[0067] In this module, the interference area identification unit 402 connects pixels of the same type. Since the infrared interference source is caused by light, the intensity of the light source gradually decreases from the center to the periphery. At the interference source, multiple concentric pixel bands will be formed. The temperature change is determined according to the distribution position of the pixel bands. For example, the lower the gray value, the higher the temperature of the area, and vice versa. This determines the direction of temperature change. If it is finally determined that there is a temperature change area in the dynamic area that gradually decreases from the center to the periphery, then the area is classified as an interference area.

[0068] The detection output unit 403 is used to collect pixel information of pixels around the interference area, adjust the color information of pixels within the interference area, obtain the repaired infrared interference detection image, perform pedestrian detection, and output the detection result.

[0069] In this module, the detection output unit 403 counts the pixel information of pixels around the interference area, randomly selects the color information of multiple pixels located around the interference area, and randomly copies these multiple pixels to cover each pixel in the interference area, thereby realizing random repair of the infrared interference detection image and reducing the impact of the interference area on pedestrian recognition. Alternatively, a preset pattern can be used for filling. Then, the repaired infrared interference detection image is imported into the pedestrian detection model for detection to output the detection result. If a pedestrian is detected in the infrared interference detection image, an alarm is issued.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An object recognition method for intelligent security, characterized in that, The method includes: An intelligent security monitoring network is constructed, which consists of multiple security monitoring devices, with different security monitoring devices corresponding to different monitoring areas; Real-time monitoring images are acquired through an intelligent security monitoring network. Pedestrian detection is performed on the real-time monitoring images to identify the areas where pedestrians are located in the real-time monitoring images. The areas where pedestrians are located are marked, and an infrared interference detection image is generated. The infrared interference detection image consists of pixels from other areas in the real-time monitoring images that do not contain the areas where pedestrians are located. Pixel comparison is performed on the infrared interference detection images. Pixels that change in adjacent infrared interference detection images are defined as dynamic pixels, and the remaining pixels are defined as static pixels, thus dividing the dynamic region into dynamic and static regions. Cluster analysis is performed on pixels in the dynamic region to determine the gradient direction of pixel color changes. The pixel colors are then repaired based on the gradient direction. Pedestrian detection is performed on the repaired infrared interference detection image, and the detection results are output.

2. The object recognition method for intelligent security according to claim 1, characterized in that, The steps of acquiring real-time monitoring images through an intelligent security monitoring network, performing pedestrian detection on the real-time monitoring images, identifying the areas where pedestrians are located in the real-time monitoring images, marking the pedestrian areas, and generating infrared interference detection images specifically include: Real-time monitoring images are acquired through an intelligent security monitoring network, and the monitoring area corresponding to the real-time monitoring image and the time of image acquisition are recorded. The trained pedestrian detection model is invoked to monitor pedestrians in real-time surveillance images, and the pedestrian detection results are output. Based on the position of each pedestrian in the pedestrian detection results, the real-time monitoring image is cropped according to the pedestrian's position, retaining pixels other than the pedestrians, to obtain an infrared interference detection image.

3. The object recognition method for intelligent security according to claim 1, characterized in that, The step of comparing pixels in the infrared interference detection images, defining pixels that change in adjacent infrared interference detection images as dynamic pixels and the remaining pixels as static pixels, and dividing the dynamic and static regions, specifically includes: Retrieve two sets of infrared interference detection images corresponding to adjacent acquisition time points, and extract the color information of corresponding pixels in adjacent infrared interference detection images; The color information of pixels at the same location is compared and the difference between the two is measured. When the difference is greater than a preset value, the current pixel is determined to be a dynamic pixel and the remaining pixels are static pixels. By merging dynamic pixels, a connected dynamic pixel region is obtained, which is the dynamic region, and the remaining part is the static region.

4. The object recognition method for intelligent security according to claim 1, characterized in that, The steps of performing cluster analysis on pixels in the dynamic region to determine the gradient direction of pixel color changes, repairing pixel colors based on the gradient direction, performing pedestrian detection on the repaired infrared interference detection image, and outputting the detection results specifically include: Cluster analysis is performed on the pixels in the dynamic region to classify the pixels in each dynamic region according to the color values ​​they contain. Pixels of the same type are connected to construct pixel bands corresponding to multiple pixel types. The direction of temperature change is determined based on the positional distribution of the pixel bands, and the existence of interference areas is determined. The pixel information of pixels around the interference area is statistically analyzed, and the color information of pixels within the interference area is adjusted to obtain a repaired infrared interference detection image. Pedestrian detection is then performed, and the detection results are output.

5. The object recognition method for intelligent security according to claim 4, characterized in that, If a pedestrian is detected in the infrared interference detection image, an alarm will be issued.

6. An intelligent security object recognition system, characterized in that, The system includes: The network construction module is used to build an intelligent security monitoring network, which consists of multiple security monitoring devices, with different security monitoring devices corresponding to different monitoring areas; The image extraction module is used to acquire real-time monitoring images through the intelligent security monitoring network, perform pedestrian detection on the real-time monitoring images, identify the area where pedestrians are located in the real-time monitoring images, mark the area where pedestrians are located, and generate an infrared interference detection image. The infrared interference detection image consists of pixels from other areas in the real-time monitoring image that do not contain the area where pedestrians are located. The region segmentation module is used to compare pixels in infrared interference detection images. It defines pixels that change in adjacent infrared interference detection images as dynamic pixels and the remaining pixels as static pixels, thus dividing the region into dynamic and static regions. The repair detection module is used to perform cluster analysis on pixels in dynamic regions, determine the gradient direction of pixel color changes, repair pixel colors based on the gradient direction, perform pedestrian detection on the repaired infrared interference detection image, and output the detection results.

7. The intelligent security object recognition system according to claim 6, characterized in that, The image extraction module includes: The image acquisition unit is used to acquire real-time monitoring images through the intelligent security monitoring network and record the monitoring area corresponding to the real-time monitoring image and the time of image acquisition. The pedestrian detection unit is used to retrieve the trained pedestrian detection model to monitor pedestrians in real-time surveillance images and output pedestrian detection results. The image cropping unit is used to crop the real-time monitoring image based on the position of each pedestrian in the pedestrian detection results, retaining pixels other than the pedestrians, to obtain an infrared interference detection image.

8. The intelligent security object recognition system according to claim 6, characterized in that, The region division module includes: The pixel information extraction unit is used to retrieve two sets of infrared interference detection images corresponding to adjacent acquisition time points and extract the color information of corresponding pixels in adjacent infrared interference detection images. The first pixel classification unit is used to compare the color information of pixels at the same position and measure the difference between them. When the difference is greater than a preset value, the current pixel is determined to be a dynamic pixel and the remaining pixels are static pixels. The pixel partitioning unit is used to merge dynamic pixels to obtain a connected dynamic pixel region, which is the dynamic region, and the remaining part is the static region.

9. The intelligent security object recognition system according to claim 6, characterized in that, The repair detection module includes: The second pixel classification unit is used to perform cluster analysis on pixels in the dynamic region and classify pixels in each dynamic region according to the color values ​​contained in the pixels. The interference area identification unit is used to connect pixels of the same type to construct pixel bands corresponding to multiple pixel types, determine the direction of temperature change based on the positional distribution of the pixel bands, and determine whether there is an interference area. The detection output unit is used to statistically analyze the pixel information of pixels around the interference area, adjust the color information of pixels within the interference area, obtain the repaired infrared interference detection image, perform pedestrian detection, and output the detection results.

10. The intelligent security object recognition system according to claim 9, characterized in that, If a pedestrian is detected in the infrared interference detection image, an alarm will be issued.

Images