A method for detecting small targets at long distances

Abstract

The application discloses a long-distance small target detection method of fire point. A plurality of infrared cameras are arranged in a scene to make the shooting visual field fully cover the scene and collect images in real time. Whether a suspicious small target S of fire point exists in the scene is judged according to the local intensity and the local gradient characteristics of the target and the background by processing the images. When it is judged that the suspicious small target S of fire point exists, n frames of images of the target S are continuously collected, the relative change distance of the target S in the n frames of images is calculated according to the center of the target S in the n frames of images. Finally, whether the suspicious small target S of fire point is a small target of fire point is judged by calculating the maximum value and the average value of the gray scale in an M*M area with the center of the target S in the n frames of images and the variance of the M*M area calculated according to the average value of the gray scale. The application can discover the small target of fire point in time, has high accuracy and low false alarm rate.

Description

Technical Field

[0001] This invention belongs to the field of infrared imaging processing technology, specifically a method for detecting small fire points at long distances. Based on the radiation characteristics of infrared imaging systems, it is particularly applicable to forest fire prevention scenarios to detect whether there is a fire within the field of view, and has the ability to detect small fire points. Background Technology

[0002] Infrared thermal imaging has advantages such as strong penetration and high clarity at night. It can not only penetrate smoke, but also image in environments with poor visibility or low light or even at night. It can achieve 24-hour uninterrupted monitoring and has been widely used in security, perimeter, military, border and coastal defense and other fields.

[0003] Currently, frequent forest fires not only damage the ecological environment but can also cause serious property losses to nearby residents. Due to the unique nature of the forest environment, it is difficult to detect small fires before large-scale fires occur, and even forest rangers find it difficult to approach deep into the forest. Infrared cameras can not only work 24 hours a day to detect fires, but also have a long operating range and are less affected by interference, making them more suitable for detecting smoke and fire in forest fire prevention, especially in the early stages of fires, and even providing fire warnings before they start. Although infrared cameras are widely used in forest fire prevention, false detections and missed detections can still occur, especially at long distances. Small fire targets are easily missed, causing missed opportunities for rescue and resulting in incalculable losses. This patent proposes a method for detecting small fire targets at long distances in forest fire prevention based on infrared imaging technology. This method can not only accurately detect targets but also reduce the probability of false detections.

[0004] Infrared small target detection is typically a challenging task for computer vision. In infrared images, small targets exhibit an isotropic Gaussian intensity function shape. In terms of intensity, the target's brightness value is greater than that of its neighboring pixels in the infrared image. For a two-dimensional Gaussian function, almost all gradients point towards its center. Similarly, the corresponding gradient of an infrared small target also points roughly towards its center. These two properties are respectively called local intensity characteristics and local gradient characteristics. In infrared images, since the grayscale values ​​of the background are almost identical, their intensity values ​​are also similar, while the target's intensity value is greater than the background's intensity value. Therefore, a local intensity threshold can effectively suppress uniform backgrounds. Similarly, for backgrounds with strong edges, their gradient directions are usually consistent, and these gradients differ significantly in distribution from the target's gradient. Therefore, by combining these two properties—local intensity characteristics and local gradient characteristics—background clutter can be effectively suppressed, and high-temperature targets can be extracted. Summary of the Invention

[0005] To address the problems existing in the detection of small fire points in current forest fire prevention scenarios, this invention proposes a long-distance small fire point detection method to quickly detect whether a fire has occurred in the forest, facilitating timely response measures.

[0006] The present invention provides a method for detecting small targets at long distances from fire points, the specific steps of which are as follows:

[0007] Step 1: Deploy infrared cameras in the scene that needs to be monitored and collect data in real time.

[0008] Step 2: Process the acquired image to determine the target and background regions, and further calculate the local intensity characteristics Q of the target and background regions. p and local gradient properties T p Determine if there are any small targets in the image that can be ignited.

[0009] Step 3: Determine if there are small fire targets in the image.

[0010] When the local intensity feature Q p Greater than a given threshold Q Y And the local gradient characteristic T p If the gradient directions all point to the center of a certain target, then the local gradient characteristic T will be... p The target pointed to by the gradient direction is marked as S, which is a small target suspected of being a fire point; at the same time, the current time is recorded as the time t when the target is discovered.

[0011] Step 4: After detecting small fire targets in Step 3, the infrared camera focuses on each small fire target and continuously acquires n frames of infrared images, labeled as N1, N2, ..., N. n .

[0012] Step 5: Perform steps 2 and 3 above on the acquired n frames of infrared images respectively to obtain N frames marked with suspicious fire points S. n Frame image.

[0013] Step 6: Record N1, N2, ..., N n The center coordinates of the suspected fire point small target S in the frame image are Z1, Z2, ..., Zn. n .

[0014] Step 7: Calculate the change in center coordinates L of the suspected small target fire point S within n consecutive frames of images. 21 =|Z2-Z1|、L 31 =|Z3-Z1|、…、L n1 =|Z n -Z1|.

[0015] Step 8: Calculate the maximum gray value of the suspicious small target S within an M*M region centered at coordinate Z1 in each of the n consecutive images, and label them as M1, M2, ..., M... n Calculate the mean gray value of an M*M region, labeled J1, J2, ..., J... n Based on the mean J1, J2, ..., J n Calculate the variance of an M*M region in multiple images, denoted as V.

[0016] Step 9, when the distance L changes 21 L 31 …、L n1 All are less than the given threshold L Y Maximum values ​​M1, M2, ..., M n All are greater than the maximum value given threshold M Y The variance V is greater than a given variance threshold V. Y At this point, the suspicious small target S is identified as a small fire point, and a fire warning is issued to remind staff to check the scene for any fire.

[0017] The advantages of this invention are:

[0018] 1) The method for detecting small fire points at long distances in this invention utilizes the advantages of infrared cameras, which have a long range and can work continuously for 24 hours, to detect small fire points in forest fire prevention scenarios. It can not only intuitively display real-time infrared images of forest scenarios, but also detect small fire points in a timely manner, and has high accuracy and low false alarm rate.

[0019] 2) Compared with traditional AI-based fire identification methods, the long-distance small target fire detection method of this invention has a longer operating distance and can identify small target fire points smaller than 5*5 pixels. At the same time, it does not require AI calculation or visible light cameras, but only infrared cameras, which greatly reduces equipment costs.

[0020] 3) The method for detecting small fire points at long distances in this invention can detect small fire points at long distances in a timely manner, and then take timely measures to minimize the losses. Attached Figure Description

[0021] Figure 1 This is a flowchart of the long-distance fire point small target detection method of the present invention. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings.

[0023] This invention relates to a method for detecting small fire targets at long distances. Based on an infrared thermal imaging system, it can identify small fire targets smaller than 5*5 pixels in forest fire prevention areas. Figure 1 As shown, the specific steps are as follows:

[0024] Step 1: Camera Setup

[0025] Infrared thermal imaging cameras are selected based on the customer's requirements for scene image clarity and cost control. For example, if high clarity is required and the cost allows, a resolution of 1920*1280 can be selected; if the clarity requirement is not high and the cost is limited, a resolution of 384*288 can be selected; if the clarity and cost are moderate, a resolution of 640*512 can be selected.

[0026] The infrared thermal imaging camera can be selected as either cooled or uncooled, and features temperature measurement, real-time grayscale value analysis, and real-time uploading of hazard signal alarms.

[0027] The infrared thermal imaging camera is fixed on the bracket. N infrared cameras are set up in the scene according to the required field of view of the forest scene to ensure that the field of view of N cameras covers the entire scene area.

[0028] Step 2: Obtain target and background features in the image

[0029] N infrared thermal imaging cameras acquire scene image data within their shooting range in real time. The image grayscale values ​​are compared with an empirically set grayscale threshold; grayscale values ​​greater than the threshold are considered target areas, while those less than the threshold are considered background areas. Furthermore, the local intensity characteristics Q of the target and background areas in each image are calculated. p and local gradient properties T p .

[0030] Step 3: Based on the local strength characteristic Q obtained in Step 2 p and local gradient properties T p Determine if there are any small targets in the image that can be ignited.

[0031] When the local intensity feature Q p Greater than a given threshold Q Y And the local gradient characteristic T p If the gradient directions all point to the center of a certain target, then the local gradient characteristic T will be... p The target pointed to by the gradient direction is marked as S, which is a small target suspected of being a fire point; at the same time, the current time is recorded as the time t when the target is discovered.

[0032] Step 4: After detecting small fire targets in Step 3, the infrared camera focuses on each small fire target and continuously acquires 25-30 frames of infrared images, labeled as N1, N2, ..., N n .

[0033] Step 5: Perform steps 2 and 3 above on the acquired n frames of infrared images respectively to obtain N frames marked with suspicious fire points S.n Frame image.

[0034] Step 6: Record N1, N2, ..., N n The center coordinates of the suspected fire point small target S in the frame image are Z1, Z2, ..., Zn. n .

[0035] Step 7: Calculate the change in center coordinates L of the suspected small target fire point S within n consecutive frames of images. 21 =|Z2-Z1|、L 31 =|Z3-Z1|、…、L n1 =|Z n -Z1|.

[0036] Step 8: Calculate the maximum gray value of the suspicious small target S within an M*M region centered at coordinate Z1 in each of the n consecutive images, and label them as M1, M2, ..., M... n M is 80, but the specific value depends on the resolution. Calculate the mean grayscale value of the M*M region, labeled J1, J2, ..., J... n Based on the mean J1, J2, ..., J n Calculate the variance of an M*M region in multiple images, denoted as V.

[0037] Step 9, when the distance L changes 21 L 31 …、L n1 All are less than the given threshold L Y Maximum values ​​M1, M2, ..., M n All are greater than the maximum value given threshold M Y The variance V is greater than a given variance threshold V. Y At this point, the suspicious small target S is identified as a small fire point, and the system issues a fire warning, reminding staff to check on-site whether a fire has occurred.

Claims

1. A method for detecting small targets at long distances, characterized in that: The specific steps are as follows: Step 1: Deploy infrared cameras in the scene that needs to be monitored and collect data in real time; Step 2: Process the acquired image to determine the target and background regions, and further calculate the local intensity characteristics of the target and background regions. and local gradient properties ; Step 3: Determine if there are small fire targets in the image; When local intensity characteristics Greater than a given threshold and local gradient characteristics If the gradient directions all point towards the center of a certain target, then the local gradient characteristics will be... The target pointed to by the gradient direction is marked as S, which is a suspected small fire point; at the same time, the current time is recorded as the time t when the target is discovered. Step 4: After detecting small fire targets in Step 3, the infrared camera focuses on each small fire target and continuously acquires n frames of infrared images, which are then marked as follows: , … ; Step 5: Process the collected data. The infrared images were processed using steps 2 and 3 described above to obtain small targets S marked with suspicious fire points. Frame image; Step 6: Record , … The center coordinates of the small, suspicious fire point target S in the frame image are respectively , … ; Step 7: Calculate the continuous values ​​separately. Changes in the center coordinates of the suspicious small target fire point S within the frame image | |、 | |、…、 | |; Step 8: Calculate the suspicious small target S in continuous The image is centered at coordinates Centered The maximum grayscale value within a large or small region is marked as , … calculate The average gray level of the large and small regions is denoted as , … According to the mean , … Calculate multiple images The variance of the large and small regions is denoted as ; Step 9, when the distance changes , …、 All are less than the given threshold maximum value , … All are greater than the maximum value given threshold The variance is Greater than a given variance threshold At this point, the suspicious small target S is identified as a small fire point, and a fire warning is issued to remind staff to check the scene for any fire.

2. The method for detecting small targets at long distances from fire points as described in claim 1, characterized in that: The infrared thermal imaging camera can be either cooled or uncooled, and features temperature measurement, real-time grayscale analysis, and real-time uploading of hazard signal alarms.

3. The method for detecting small targets at long distances from fire points as described in claim 1, characterized in that: The infrared thermal imaging camera is fixed on the bracket. N infrared cameras are set up in the scene according to the required field of view of the scene area to ensure that the field of view of N cameras covers the entire scene area.

4. The method for detecting small targets at long distances from fire points as described in claim 1, characterized in that: In step two, the grayscale values ​​of the acquired image are compared with a grayscale threshold. Grayscale values ​​greater than the threshold are considered target areas, while grayscale values ​​less than the threshold are considered background areas.

5. The method for detecting small targets at long distances from fire points as described in claim 1, characterized in that: In step 4, it is advisable to acquire 25 to 30 frames of images.

Images