Method and device for reducing vehicle detection error rate

Abstract

The present invention discloses a method and device for reducing the false recognition rate of vehicle detection. The method comprises first acquiring an original image beneath a power transmission and distribution line and detecting a vehicle location area from the original image; preprocessing the image of the vehicle location area; calculating the Gaussian distribution score and correlation coefficient score of each pixel point in the vehicle location area in the preprocessed image; using the Gaussian distribution scores and correlation coefficient scores of all pixels in the vehicle location area, calculating the Gaussian distribution score and correlation coefficient score of the vehicle location area; and determining whether a vehicle exists in the detected vehicle location area based on the Gaussian distribution score and correlation coefficient score of the vehicle location area. The present invention utilizes the combined Gaussian distribution score and correlation coefficient score within the vehicle location area to determine whether the vehicle location area is a vehicle or background, and can accurately determine whether the detected vehicle location area is a real vehicle or background.

Description

Technical Field

[0001] The present disclosure relates to a method and device for reducing the false recognition rate of vehicle detection. Background Art

[0002] Power transmission and distribution lines are crucial for power grids. Overhead transmission and distribution lines are widely distributed and exposed to the elements for long periods of time, subject to interference from various factors and the ever-changing nature of nature. Construction work under these lines can easily damage them. Early warning is a crucial measure to prevent the impact of construction on these lines. Construction sites often involve vehicles such as tower cranes, cement trucks, and excavators. Detecting the areas beneath transmission and distribution lines and generating alerts when construction vehicles are present can effectively prevent damage to these lines. Cameras mounted on transmission towers can capture images of the area beneath the lines. After image processing, the presence and type of vehicles beneath the lines can be identified.

[0003] Due to the complex environment and changing lighting conditions in nature, a certain percentage of misidentifications can occur, misidentifying background objects as vehicles and causing false alarms. Therefore, designing an accurate vehicle detection method that can reduce misidentifications remains a technical challenge. Summary of the Invention

[0004] In order to overcome the deficiencies of the above-mentioned prior art, the present disclosure provides a method and device for reducing the false recognition rate of vehicle detection, which utilizes the combined Gaussian distribution score and correlation coefficient score in the vehicle position area to determine whether the vehicle position area is a vehicle or background. It can quickly and accurately determine whether the detected vehicle position area is a real vehicle or background, thereby reducing the probability of false alarms.

[0005] The technical solution adopted in this disclosure is:

[0006] A method for reducing vehicle detection error rate, the method comprising the following steps:

[0007] Obtaining an original image under the power transmission and distribution line, and detecting the vehicle location area from the original image;

[0008] Preprocessing the image of the area where the vehicle is located;

[0009] Calculate the Gaussian distribution score and correlation coefficient score of each pixel in the vehicle location area in the preprocessed image;

[0010] Calculate the Gaussian distribution score and correlation coefficient score of the vehicle location area using the Gaussian distribution scores and correlation coefficient scores of all pixels in the vehicle location area;

[0011] According to the Gaussian distribution score and correlation coefficient score of the vehicle location area, it is determined whether there is a vehicle in the detected vehicle location area.

[0012] Furthermore, the trained deep learning model is used to detect the vehicle location area from the original image to obtain the vehicle location information and type information.

[0013] Furthermore, the step of preprocessing the image of the area where the vehicle is located includes:

[0014] In the original image, for each pixel in the area where the vehicle is located, calculate the mean and standard deviation of the data in the square area with each pixel as the center point;

[0015] The obtained mean and standard deviation of the data in the square area with each pixel as the center point are used to calculate the new pixel value of each pixel, and the new pixel value is used as the processed image pixel value of each pixel to obtain the preprocessed image.

[0016] Furthermore, the Gaussian distribution score of each pixel is calculated as follows:

[0017] In the pre-processed image, for each pixel in the vehicle's location area, calculate the Gaussian distribution mean and standard deviation of the corresponding pixel values obtained in the images of several consecutive time periods before the current time;

[0018] The Gaussian distribution score of each pixel point is calculated by using the Gaussian distribution mean and standard deviation of the pixel values corresponding to the image obtained in several consecutive time periods before the current time.

[0019] Furthermore, the correlation coefficient score of each pixel is calculated as follows:

[0020] In the pre-processed image, for each pixel point in the vehicle location area, calculate the pixel change in the square area centered on each pixel point;

[0021] The correlation coefficient score of each pixel is calculated using the obtained pixel change in the square area centered on each pixel.

[0022] Furthermore, the Gaussian distribution score and the correlation coefficient score of the vehicle location area are calculated as follows:

[0023] Calculate the sum of the Gaussian distribution scores of all pixels in the vehicle's location area and average them to obtain the Gaussian distribution score of the vehicle's location area;

[0024] The sum of the correlation coefficient scores of all pixels in the vehicle location area is calculated and averaged to obtain the correlation coefficient score of the vehicle location area.

[0025] Furthermore, the step of determining whether there is a vehicle in the detected vehicle location area includes:

[0026] Setting thresholds 、 、 , weight coefficient 、 ;

[0027] Give threshold Gaussian distribution score of the area where the vehicle is located The difference distribution weight , give the threshold Correlation coefficient score with the vehicle's location area The difference distribution weight , linearly combine the two and calculate the total score of the vehicle's location area ;

[0028] Determine the total score of the vehicle's location area With threshold size;

[0029] like , then there is a vehicle in the area where the detected vehicle is located, otherwise the area is the background and there is no vehicle.

[0030] A device for reducing the vehicle detection error rate, the device is used to implement the method for reducing the vehicle detection error rate as described above, the device comprising:

[0031] an image acquisition module, mounted on a transmission tower, configured to acquire images below the transmission and distribution lines and transmit the images to a processor;

[0032] The processor is configured to receive the original image uploaded by the image acquisition module, use the deep learning model to detect the vehicle location area from the original image, preprocess the image of the vehicle location area, calculate the Gaussian distribution score and the correlation coefficient score of each pixel point in the vehicle location area in the preprocessed image, calculate the sum of the Gaussian scores of all pixels in the vehicle location area and average them to obtain the Gaussian score of the vehicle location area, calculate the sum of the correlation coefficient scores of all pixels in the vehicle location area and average them to obtain the correlation coefficient score of the vehicle location area, and determine whether there is a vehicle in the detected vehicle location area based on the Gaussian distribution score and the correlation coefficient score of the vehicle location area.

[0033] The beneficial effects of the present disclosure are:

[0034] (1) The present disclosure utilizes the combined Gaussian distribution score and correlation coefficient score within the vehicle location area to determine whether the vehicle location area is a vehicle or background. This can quickly and accurately determine whether the detected vehicle location area is a real vehicle or background, thereby reducing the probability of false alarms.

[0035] (2) The present disclosure uses each pixel point in the vehicle position area to calculate the mean and standard deviation of the square area centered on it, and calculates the new pixel value of the pixel point. By obtaining the new pixel value of the pixel point, the Gaussian distribution score and correlation coefficient score of each pixel point are calculated, and the average value of the Gaussian distribution score and correlation coefficient score of all pixels in the vehicle position area is used as the Gaussian score and correlation coefficient score of the vehicle position area. A threshold is set, and the Gaussian score and correlation coefficient score of the vehicle position area are linearly combined to calculate the final score. The final score is compared with the threshold to determine whether the detected vehicle position area is a real vehicle or background, thereby greatly reducing the misidentification phenomenon in vehicle detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings, which constitute a part of the present disclosure, are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and their descriptions are used to explain the present application and do not constitute an improper limitation to the present disclosure.

[0037] Figure 1 It is a flow chart of a method for reducing the false recognition rate of vehicle detection. DETAILED DESCRIPTION

[0038] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used in the present disclosure have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0041] One or more embodiments provide a method for reducing the false recognition rate of vehicle detection, such as Figure 1 As shown, the method includes the following steps:

[0042] S101, vehicle detection.

[0043] In step 101, the specific implementation process of vehicle detection is as follows:

[0044] S101-1, obtaining an original image below a power transmission and distribution line.

[0045] The camera installed on the transmission tower collects the original images below the transmission and distribution lines through the camera.

[0046] S101-2, using a deep learning model to detect the vehicle location area from the original image obtained in step S101-1.

[0047] In this embodiment, a pre-trained deep learning model is used to detect the vehicle location area in the image to obtain the vehicle location information and type information.

[0048] S102, remove misidentification.

[0049] In step 102, the specific implementation process of misidentification removal is as follows:

[0050] S102-1, pre-processing the image of the area where the vehicle is located.

[0051] The specific implementation of step 102-1 is as follows:

[0052] (1-1) For the original image, for each pixel in the area where the vehicle is located, calculate the value of the pixel as the center point and the side length as The mean and standard deviation of the data within the square area.

[0053] (1-2) Process the image according to the mean and standard deviation obtained in step (1-1) to obtain a preprocessed image.

[0054] In this embodiment, the following calculation is performed for each pixel in the vehicle position area:

[0055]

[0056] in, is the image row where the pixel is located, is the image column where the pixel is located, is the pixel value of the current pixel, is a square centered on the pixel (the side length of the square is , choose appropriately according to the image size , is a natural number), is the standard deviation of the data within the square centered on the pixel, is the pixel value of the preprocessed image after processing, if If it is greater than 255, it is set to 255. If it is less than 0, it is set to 0.

[0057] S102-2, calculating the Gaussian distribution score of each pixel in the vehicle location area in the pre-processed image And find the average .

[0058] In step 102-2, the Gaussian distribution score of each pixel is calculated as follows:

[0059] (2-1) In the pre-processed image, for each pixel in the vehicle location area, calculate the Gaussian distribution mean of the pixel corresponding to the image obtained in the n consecutive time periods before the current time. and standard deviation , the mean of the Gaussian distribution and standard deviation The calculation formula is as follows:

[0060]

[0061]

[0062] Where T is the current time, is the pixel value of the corresponding pixel point on the preprocessed image at time t.

[0063] (2-2) Using the pixels corresponding to the images obtained in the n consecutive time periods before the current time and standard deviation , calculate the Gaussian distribution score of each pixel and its average value .

[0064] According to the pixel points corresponding to the image obtained in the n consecutive time periods before the current time and standard deviation , calculate the Gaussian distribution score of each pixel in the location area of a single vehicle , The calculation formula is:

[0065]

[0066] Then, calculate the sum of the Gaussian distribution scores of each pixel in the location area of a single vehicle ,Will

[0067] Compared with the number of pixels N in the area where a single vehicle is located, the average value of the Gaussian distribution score is obtained , the mean of the Gaussian distribution scores The calculation formula is:

[0068]

[0069] Among them, N is the number of pixels in the location area of a single vehicle, It is the sum of the Gaussian distribution scores of each pixel in the location area of a single vehicle.

[0070] In this embodiment, the average value of the Gaussian distribution scores of all pixels in the vehicle location area is calculated. As the Gaussian distribution score of the vehicle location area. The higher it is, the higher the probability that the area is not background.

[0071] S102-3, calculate the correlation coefficient of each pixel in the vehicle's location area to obtain a score And find the average .

[0072] In step 102-3, the correlation coefficient score is calculated as follows:

[0073] (3-1) In the pre-processed image, for each pixel in the area where the vehicle is located, calculate the pixel with a side length of The pixel change in the square area is calculated to calculate the correlation coefficient score of each pixel .

[0074] Correlation coefficient score The calculation formula is:

[0075]

[0076] in,

[0077]

[0078]

[0079] Where, is pixel pixel; is the pixel change in the square area; is a natural number; is the Gaussian distribution mean of the pixel points; is the average value of the Gaussian distribution of pixels in the square area.

[0080] (2) Calculate the average value of the relevant system scores of all pixels in the vehicle's location area ,average value The calculation formula is:

[0081]

[0082] Among them, N is the number of pixels in the location area of a single vehicle, It is the sum of the system scores for each pixel in the area where a single vehicle is located.

[0083] In this embodiment, the average value of the relevant system scores of all pixels in the vehicle location area is calculated. As the correlation system score of the vehicle's location area. The correlation coefficient score of the vehicle's location area The higher it is, the higher the probability that the area is background.

[0084] S103, scoring based on the Gaussian distribution of the vehicle's location area and correlation coefficient scores To determine whether the detected vehicle location area is the background.

[0085] In this embodiment, the Gaussian score of the vehicle position area is and correlation coefficient scores Perform linear combination to calculate the total score of the vehicle location area and set a threshold. If the total score is greater than the threshold, it is background, otherwise it is a vehicle.

[0086] The process of determining whether the detected vehicle location area is background in step 103 is as follows:

[0087] S103-1, setting threshold 、 、 , weight coefficient 、 ;

[0088] S103-2, calculate the total score ;

[0089] S103-3, Judgment and size;

[0090] S103-4, if , then there is a vehicle in the area, otherwise the area is the background and there is no vehicle.

[0091] The method for reducing the vehicle detection error rate proposed in this embodiment is to first detect the vehicle position information and type information in the image, calculate the mean and standard deviation of the square area centered on each pixel point in the vehicle position area, obtain a new pixel value of the pixel point by calculation, calculate the Gaussian distribution mean and standard deviation of the image corresponding to the pixel in the continuous time period before the current time, calculate the Gaussian distribution score of the pixel point, calculate the Gaussian distribution score of the vehicle position area, calculate the average value of the Gaussian distribution mean and standard deviation of the square area centered on each pixel point in the vehicle position, calculate the correlation coefficient score of each pixel point, calculate the correlation coefficient score of the vehicle position area, linearly combine the Gaussian distribution score and the correlation coefficient score of the vehicle position area to obtain a total score of the vehicle position area, and judge whether the vehicle position area is a vehicle or background based on the total score of the vehicle position area, thereby greatly reducing the misrecognition phenomenon in vehicle detection and improving vehicle detection efficiency.

[0092] One or more embodiments provide experimental verification of the above-mentioned method for reducing the vehicle detection error rate. The process of the experimental verification is as follows:

[0093] The database stores 1,000 images taken at the same location at different times. 40% of the images contain vehicles, and 60% do not. There are actually 700 vehicles, and 728 vehicles were detected, of which 28 were false positives. After applying the above method to reduce the false recognition rate of vehicle detection, 26 falsely detected vehicles were successfully removed, greatly reducing the false recognition phenomenon. The above method can effectively reduce the false recognition rate of vehicles.

[0094] One or more embodiments further provide a device for reducing the vehicle detection error rate, the device being used to implement the above-mentioned method for reducing the vehicle detection error rate, the device comprising:

[0095] an image acquisition module, mounted on a transmission tower, configured to acquire images below the transmission and distribution lines and transmit the images to a processor;

[0096] The processor is configured to receive the original image uploaded by the image acquisition module, use the deep learning model to detect the vehicle location area from the original image, preprocess the image of the vehicle location area, calculate the Gaussian distribution score and correlation coefficient score of each pixel in the vehicle location area in the preprocessed image, and determine whether there is a vehicle in the detected vehicle location area based on the average value of the Gaussian distribution score and the correlation coefficient score.

[0097] From the above description, it can be seen that one or more of the above embodiments achieve the following technical effects:

[0098] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A method for reducing the false recognition rate of vehicle detection, characterized in that: The method comprises the following steps: Obtaining an original image under the power transmission and distribution line, and detecting the vehicle location area from the original image; Preprocessing the image of the area where the vehicle is located; Calculate the Gaussian distribution score and correlation coefficient score of each pixel in the vehicle location area in the preprocessed image; Calculate the Gaussian distribution score and correlation coefficient score of the vehicle location area using the Gaussian distribution scores and correlation coefficient scores of all pixels in the vehicle location area; According to the Gaussian distribution score and correlation coefficient score of the vehicle location area, it is determined whether there is a vehicle in the detected vehicle location area; The step of determining whether there is a vehicle in the detected vehicle location area includes: Setting thresholds 、 、 , weight coefficient 、 ; Give threshold Gaussian distribution score of the area where the vehicle is located The difference distribution weight , give the threshold Correlation coefficient score with the vehicle's location area The difference distribution weight , linearly combine the two and calculate the total score of the vehicle's location area ; Determine the total score of the vehicle's location area With threshold size; like , then there is a vehicle in the area where the detected vehicle is located, otherwise the area is the background and there is no vehicle.

2. The method for reducing the vehicle detection error rate according to claim 1, wherein: The trained deep learning model is used to detect the vehicle location area from the original image to obtain the vehicle location information and type information.

3. The method for reducing the vehicle detection error rate according to claim 1, wherein: The step of pre-processing the image of the vehicle location area includes: In the original image, for each pixel in the area where the vehicle is located, calculate the mean and standard deviation of the data in the square area with each pixel as the center point; The obtained mean and standard deviation of the data in the square area with each pixel as the center point are used to calculate the new pixel value of each pixel, and the new pixel value is used as the processed image pixel value of each pixel to obtain the preprocessed image.

4. The method for reducing the vehicle detection error rate according to claim 1, wherein: The calculation method of the Gaussian distribution score of each pixel is: In the pre-processed image, for each pixel in the vehicle's location area, calculate the Gaussian distribution mean and standard deviation of the corresponding pixel values obtained in the images of several consecutive time periods before the current time; The Gaussian distribution score of each pixel point is calculated by using the Gaussian distribution mean and standard deviation of the pixel values corresponding to the image obtained in several consecutive time periods before the current time.

5. The method for reducing the vehicle detection error rate according to claim 1, wherein: The calculation method of the correlation coefficient score of each pixel is: In the pre-processed image, for each pixel point in the vehicle location area, calculate the pixel change in the square area centered on each pixel point; The correlation coefficient score of each pixel is calculated using the obtained pixel change in the square area centered on each pixel.

6. The method for reducing the vehicle detection error rate according to claim 1, wherein: The calculation method of the Gaussian distribution score and the correlation coefficient score of the vehicle location area is: Calculate the sum of the Gaussian distribution scores of all pixels in the vehicle's location area and average them to obtain the Gaussian distribution score of the vehicle's location area; The sum of the correlation coefficient scores of all pixels in the vehicle location area is calculated and averaged to obtain the correlation coefficient score of the vehicle location area.

7. A device for reducing the vehicle detection error rate, the device being used to implement the method for reducing the vehicle detection error rate according to any one of claims 1 to 6, the device comprising: an image acquisition module, mounted on a transmission tower, configured to acquire images below the transmission and distribution lines and transmit the images to a processor; The processor is configured to receive an original image uploaded by the image acquisition module, detect a vehicle location area from the original image using a deep learning model, preprocess the image of the vehicle location area, calculate a Gaussian distribution score and a correlation coefficient score for each pixel in the vehicle location area in the preprocessed image, calculate the sum and average of the Gaussian scores of all pixels in the vehicle location area to obtain a Gaussian score for the vehicle location area, calculate the sum and average of the correlation coefficient scores of all pixels in the vehicle location area to obtain a correlation coefficient score for the vehicle location area, and determine whether there is a vehicle in the detected vehicle location area based on the Gaussian distribution score and the correlation coefficient score of the vehicle location area; The step of determining whether there is a vehicle in the detected vehicle location area includes: Setting thresholds 、 、 , weight coefficient 、 ; Give threshold Gaussian distribution score of the area where the vehicle is located The difference distribution weight , give the threshold Correlation coefficient score with the vehicle's location area The difference distribution weight , linearly combine the two and calculate the total score of the vehicle's location area ; Determine the total score of the vehicle's location area With threshold size; like , then there is a vehicle in the area where the detected vehicle is located, otherwise the area is the background and there is no vehicle.

Images