技术领域Technical Field
本发明涉及环境检测领域，特别涉及一种基于反向投影的施工场地烟尘检测方法。The invention relates to the field of environmental detection, and in particular to a construction site smoke detection method based on reverse projection.
背景技术Background technique
传统的扬尘识别方法大多采用物理信号进行监测，如通过空气质量传感器检测空气中PM2.5、PM10的浓度反应空气中尘土含量，但此类方法识别距离有限，且比较依赖环境，容易受到环境干扰，从而导致烟尘检测距离短、准确率低。目前也有采用图像处理等手段如神经网络来实现扬尘图像的学习与识别，但该方法比较依赖数据集，然而由于训练集难以获取、扬尘面积难以划分，导致了烟尘检测准确率低的问题。Traditional dust identification methods mostly use physical signals for monitoring, such as using air quality sensors to detect the concentration of PM2.5 and PM10 in the air to reflect the dust content in the air. However, such methods have limited recognition distances and are more dependent on the environment and are easily disturbed by the environment, resulting in short smoke detection distances and low accuracy. Currently, image processing methods such as neural networks are also used to learn and identify dust images, but this method is more dependent on data sets. However, since it is difficult to obtain training sets and divide dust areas, it leads to low smoke detection accuracy.
发明内容Summary of the invention
本发明目的是为了解决现有烟尘检测方法还存在检测距离短、检测准确率低的问题，而提出了一种基于反向投影的施工场地烟尘检测方法。The purpose of the present invention is to solve the problems of short detection distance and low detection accuracy in existing smoke detection methods, and propose a construction site smoke detection method based on reverse projection.
一种基于反向投影的施工场地烟尘检测方法具体过程为：A method for detecting smoke and dust at construction sites based on reverse projection. The specific process is as follows:
步骤一、获取样本图像，从而获得样本图像的分布概率形式直方图；Step 1: Obtain a sample image, thereby obtaining a distribution probability histogram of the sample image;
步骤二、通过交并运算、差运算对步骤一获得的样本图像的分布概率形式直方图进行学习，获得学习后的分布概率形式直方图；Step 2: learning the distribution probability form histogram of the sample image obtained in step 1 through intersection and difference operations to obtain a learned distribution probability form histogram;
步骤三、获取待测图像，利用步骤二获得的学习后的分布概率形式二维分布直方图对待测图像进行反向投影，获得目标概率分布矩阵P，然后利用高斯滤波算子对目标概率分布矩阵P进行卷积，提取区域目标分布概率信息；Step 3: Obtain the image to be tested, and use the learned distribution probability form two-dimensional distribution histogram obtained in step 2 to reversely project the image to be tested to obtain the target probability distribution matrix P, and then use the Gaussian filter operator to convolve the target probability distribution matrix P to extract the regional target distribution probability information;
步骤四、将步骤三获得的区域目标分布概率信息与预设阈值比较，将待测图像中概率大于预设阈值的区域的图像掩膜置1，将待测图像中概率小于等于预设阈值区域的图像掩膜置0，获得初步识别结果；然后通过滑动窗口对初始识别结果中掩膜进行区域生长，获得掩膜检测结果；Step 4: compare the regional target distribution probability information obtained in step 3 with the preset threshold, set the image mask of the area in the image to be tested whose probability is greater than the preset threshold to 1, and set the image mask of the area in the image to be tested whose probability is less than or equal to the preset threshold to 0, and obtain a preliminary recognition result; then perform regional growth on the mask in the initial recognition result through a sliding window to obtain a mask detection result;
步骤五、利用步骤四获得的掩膜检测结果获得待测图像中的烟尘占比。Step 5: Use the mask detection result obtained in step 4 to obtain the smoke ratio in the image to be tested.
进一步地，所述步骤一中的获取样本图像，从而获得样本图像的分布概率形式直方图，具体为：Furthermore, the step 1 of obtaining a sample image, thereby obtaining a distribution probability histogram of the sample image, is specifically:
步骤一一、获取样本图像，并将样本图像转换为HSV数据格式，然后提取样本图像色相通道H、饱和度通道S；Step 11, obtain a sample image, convert the sample image into HSV data format, and then extract the hue channel H and saturation channel S of the sample image;
所述样本图像包括：正例样本图像Ht、反例样本图像Hf；The sample images include: a positive sample image Ht and a negative sample image Hf ;
所述正例样本图像为包含检测对象的样本图像；The positive sample image is a sample image containing the detection object;
所述反例样本图像为检测环境的样本图像；The counterexample sample image is a sample image of the detection environment;
步骤一二、将步骤一一提取出的样本图像色相通道H的范围缩放为0-360，以a为H通道单元间隔，作x轴；将饱和度通道S的范围缩放为0-255，以b为S通道间隔，作y轴，从而获得样本图像的HS通道二维分布直方图；Step 12: Scale the range of the hue channel H of the sample image extracted in step 11 to 0-360, with a as the H channel unit interval as the x-axis; scale the range of the saturation channel S to 0-255, with b as the S channel interval as the y-axis, so as to obtain a two-dimensional distribution histogram of the HS channel of the sample image;
步骤一三、将步骤一二获得到的HS通道二维分布直方图转化为概率分布形式，获得样本图像的分布概率形式直方图。Step 13: Convert the two-dimensional distribution histogram of the HS channel obtained in step 12 into a probability distribution form to obtain a distribution probability form histogram of the sample image.
进一步地，所述样本图像的分布概率形式直方图中每个位置的分布概率，通过以下公式获得：Furthermore, the distribution probability of each position in the distribution probability histogram of the sample image is obtained by the following formula:
其中，H1是HS通道二维分布直方图，(x,y)是二维分布直方图中的某点坐标，Max(H1)是HS通道二维分布直方图中的最大值，H′(x,y)是(x,y)处的分布概率，Min(H1)是HS通道二维分布直方图中的最小值，H1(x,y)是(x,y)处的是HS通道二维分布直方图。Where H1 is the two-dimensional distribution histogram of the HS channel, (x, y) is the coordinate of a point in the two-dimensional distribution histogram, Max(H1) is the maximum value in the two-dimensional distribution histogram of the HS channel, H′(x, y) is the distribution probability at (x, y), Min(H1) is the minimum value in the two-dimensional distribution histogram of the HS channel, and H1(x, y) is the two-dimensional distribution histogram of the HS channel at (x, y).
进一步地，所述步骤二中的通过交并运算、差运算对步骤一获得的样本图像的分布概率形式直方图进行学习，获得学习后的分布概率形式直方图，具体如下：Furthermore, in the step 2, the distribution probability form histogram of the sample image obtained in step 1 is learned through intersection and union operations and difference operations to obtain the learned distribution probability form histogram, which is specifically as follows:
步骤二一、对正例样本图像的分布概率形式直方图进行交集运算，如下式：Step 21: Perform an intersection operation on the distribution probability histogram of the positive sample image, as shown in the following formula:
其中，n是样本图像的分布概率形式直方图中的区块总数，j是区块标号，Ht1(j)是第t1个正例样本图像的分布概率形式直方图中的第j个区块，Ht2(j)是第t2个正例样本图像的分布概率形式直方图中的第j个区块；Where n is the total number of blocks in the distribution probability form histogram of the sample image, j is the block number, Ht1 (j) is the jth block in the distribution probability form histogram of thet1th positive sample image, and Ht2 (j) is the jth block in the distribution probability form histogram of thet2th positive sample image;
步骤二二、对正例样本图像的分布概率形式直方图进行并集运算，如下式：Step 22: Perform a union operation on the distribution probability histogram of the positive sample image, as shown in the following formula:
步骤二三、将正例样本图像的分布概率形式直方图与反例样本图像的分布概率形式直方图进行差运算，如下式：Step 2: Perform a difference operation on the distribution probability form histogram of the positive sample image and the distribution probability form histogram of the negative sample image, as shown in the following formula:
进一步地，所述步骤三中的高斯滤波算子，如下式：Furthermore, the Gaussian filter operator in step 3 is as follows:
其中，σ2是滤波方差，x'是滤波器的横轴尺寸，y'是滤波器的纵轴尺寸。Where σ2 is the filter variance, x' is the horizontal axis size of the filter, and y' is the vertical axis size of the filter.
进一步地，所述步骤四中的通过滑动窗口对初始识别结果中掩膜进行区域生长，获得掩膜检测结果，如下式：Furthermore, in step 4, the mask in the initial recognition result is subjected to region growth by means of a sliding window to obtain a mask detection result, which is as follows:
其中，M(x1,y1)通过阈值筛选的待测图像掩膜，为M的非，Model为掩膜检测结果，beta是超参数，x1、y1滑动窗口内像素的横纵坐标，ave是均值，std是方差。Among them, M(x1,y1) is the mask of the image to be tested that is filtered by the threshold. is the negation of M, Model is the mask detection result, beta is a hyperparameter, x1 and y1 are the horizontal and vertical coordinates of the pixels in the sliding window, ave is the mean, and std is the variance.
进一步地，further,
其中，Num为掩膜M内标记像素的数量，W为滑动窗口长度。Among them, Num is the number of marked pixels in the mask M, and W is the sliding window length.
进一步地，further,
进一步地，所述步骤五中的利用步骤四获得的掩膜检测结果获得待测图像中的烟尘占比，具体为：Furthermore, the smoke proportion in the image to be tested is obtained by using the mask detection result obtained in step 4 in step 5, specifically:
步骤五一、对步骤四获得的掩膜检测结果进行腐蚀与膨胀，去除检测噪点，获得去除检测噪点后的掩膜检测结果Model′，具体为：Step 51: Corrode and dilate the mask detection result obtained in step 4 to remove the detection noise points, and obtain the mask detection result Model′ after removing the detection noise points, which is specifically:
采用半径为4的线性球形算子，分别对步骤四获得的掩膜检测结果采取一次开运算与闭运算，填补图像掩膜孔洞同时清除离散点，获得Model′；A linear spherical operator with a radius of 4 is used to perform an opening operation and a closing operation on the mask detection result obtained in step 4, respectively, to fill the holes in the image mask and remove the discrete points, and obtain Model′;
步骤五二、利用去除检测噪点后的掩膜检测结果Model′获得待测图像中的烟尘占比。Step 52: Use the mask detection result Model′ after removing the detection noise to obtain the smoke and dust ratio in the image to be tested.
进一步地，所述步骤五二中的利用去除检测噪点后的掩膜检测结果Model′获得待测图像中的烟尘占比，如下式：Furthermore, in the step 52, the smoke proportion in the image to be tested is obtained by using the mask detection result Model′ after removing the detection noise, as shown in the following formula:
式中，row为待测图像的行数，col为待测图像的列数，Sum()是求和函数。In the formula, row is the number of rows of the image to be tested, col is the number of columns of the image to be tested, and Sum() is the summation function.
本发明的有益效果为：The beneficial effects of the present invention are:
本发明通过直方图反向投影方法，将识别图像转化为目标分布的概率形式，通过卷积提取局部分布概率来反应出烟尘存在概率，通过生长算法拓展烟尘检测覆盖面积，通过腐蚀膨胀对空洞、噪点进行消除，最终得到看烟尘覆盖掩膜。本发明的正例样本图像根据检测烟尘特点选取相近的烟尘样本图像；反例模板从检测现场采集不含烟尘的环境图像，无需使用大量的数据样本进行训练。本发明通过烟尘的分割与定量计算，获得了烟尘占比，从而烟尘占比量化了烟尘的扩散程度。本发明检测距离长，同时并不依赖数据集划分扬尘面积，从而提升了烟尘检测准确率。The present invention converts the recognition image into a probability form of target distribution through a histogram backprojection method, extracts the local distribution probability through convolution to reflect the probability of smoke existence, expands the smoke detection coverage area through a growth algorithm, eliminates voids and noise points through corrosion expansion, and finally obtains a smoke coverage mask. The positive sample image of the present invention selects similar smoke sample images according to the characteristics of the smoke detection; the negative example template collects environmental images without smoke from the detection site, and does not need to use a large amount of data samples for training. The present invention obtains the smoke proportion through the segmentation and quantitative calculation of the smoke, so that the smoke proportion quantifies the degree of smoke diffusion. The present invention has a long detection distance and does not rely on the data set to divide the dust area, thereby improving the accuracy of smoke detection.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
图1为本发明检测案例图像；FIG1 is an image of a detection case of the present invention;
图2为本发明检测过程中选用的正例模板；FIG2 is a positive example template selected in the detection process of the present invention;
图3为模板图提取HS通道后计算得到的二维直方图；FIG3 is a two-dimensional histogram calculated after extracting HS channels from the template image;
图4为检测图像经模板直方图反向投影得到的目标分布概率图像；FIG4 is a target distribution probability image obtained by back-projecting the detection image through the template histogram;
图5为掩膜原始范围、经生长后的范围、经腐蚀与膨胀操作后的范围图像；FIG5 is an image of the original range of the mask, the range after growth, and the range after corrosion and expansion operations;
图6为最终检测结果图像；Figure 6 is the final detection result image;
图7为本发明流程图。FIG. 7 is a flow chart of the present invention.
具体实施方式Detailed ways
具体实施方式一：如图7所示，本实施方式一种基于反向投影的施工场地烟尘检测方法具体过程为：Specific implementation method 1: As shown in FIG7 , the specific process of a construction site smoke detection method based on reverse projection in this implementation method is as follows:
步骤一、获取样本图像，从而获得样本图像的分布概率形式直方图：Step 1: Get the sample image to obtain the distribution probability histogram of the sample image:
步骤一一、根据检测的环境条件、检测的烟尘类型获取样本图像，将样本图像转换为HSV数据格式，然后提取样本图像色相通道H、饱和度通道S；Step 11: Obtain a sample image according to the detected environmental conditions and the detected smoke type, convert the sample image into HSV data format, and then extract the hue channel H and saturation channel S of the sample image;
所述样本图像包括：正例样本图像Ht、反例样本图像Hf；The sample images include: a positive sample image Ht and a negative sample image Hf ;
所述正例样本图像为包含检测对象的样本图像，如图2所示；The positive sample image is a sample image containing a detection object, as shown in FIG2 ;
所述反例样本图像为检测环境的样本图像；The counterexample sample image is a sample image of the detection environment;
步骤一二、将步骤一一提取出的样本图像色相通道H的范围缩放为0-360，以2为H通道单元间隔，作x轴；将饱和度通道S的范围缩放为0-255，以1为S通道间隔，作y轴，从而获得样本图像的HS通道二维分布直方图，如图3所示；Step 12: Scale the range of the hue channel H of the sample image extracted in step 1 to 0-360, with 2 as the H channel unit interval, as the x-axis; scale the range of the saturation channel S to 0-255, with 1 as the S channel interval, as the y-axis, so as to obtain a two-dimensional distribution histogram of the HS channel of the sample image, as shown in Figure 3;
步骤一三、将步骤一二获得到的HS通道二维分布直方图转化为概率分布形式，获得样本图像的分布概率形式直方图，如图4所示；Step 13: convert the HS channel two-dimensional distribution histogram obtained in step 12 into a probability distribution form to obtain a distribution probability form histogram of the sample image, as shown in FIG4 ;
每个位置的分布概率通过以下公式获得：The distribution probability of each position is obtained by the following formula:
其中，H1是HS通道二维分布直方图，(x,y)是二维分布直方图中的某点坐标，Max(H1)是HS通道二维分布直方图中的最大值，H′(x,y)是(x,y)处的分布概率，Min(H1)是HS通道二维分布直方图中的最小值，H1(x,y)是(x,y)处的是HS通道二维分布直方图。Where H1 is the two-dimensional distribution histogram of the HS channel, (x, y) is the coordinate of a point in the two-dimensional distribution histogram, Max(H1) is the maximum value in the two-dimensional distribution histogram of the HS channel, H′(x, y) is the distribution probability at (x, y), Min(H1) is the minimum value in the two-dimensional distribution histogram of the HS channel, and H1(x, y) is the two-dimensional distribution histogram of the HS channel at (x, y).
步骤二、通过交并运算、差运算对步骤一获得的样本图像的分布概率形式直方图进行学习，获得学习后的分布概率形式直方图，具体如下：Step 2: The distribution probability form histogram of the sample image obtained in step 1 is learned through intersection and difference operations to obtain the distribution probability form histogram after learning, which is as follows:
步骤二一、对正例样本图像的分布概率形式直方图进行交集运算，以提取正例样本间的共同特征，如公式(2)所示：Step 2: Perform an intersection operation on the distribution probability form histogram of the positive sample image to extract the common features between the positive sample images, as shown in formula (2):
其中，j是区块标号，n是直方图区块总数，Ht1(j)是第t1个正例样本图像的分布概率形式直方图中的第j个区块，Ht2(j)是第t2个正例样本图像的分布概率形式直方图中的第j个区块；通过交集运算能够提高检测的精度，减少因环境中相近的颜色造成的干扰。Among them, j is the block label, n is the total number of histogram blocks, Ht1 (j) is the j-th block in the distribution probability form histogram of thet1th positive sample image, and Ht2 (j) is the j-th block in the distribution probability form histogram of thet2th positive sample image. The intersection operation can improve the detection accuracy and reduce the interference caused by similar colors in the environment.
步骤二二、对正例样本图像的分布概率形式直方图进行并集运算，以泛化检测结果，提高模型对烟雾检出的敏感度，如公式(3)所示：Step 2: Perform a union operation on the distribution probability histogram of the positive sample image to generalize the detection results and improve the model's sensitivity to smoke detection, as shown in formula (3):
并集运算能使检测模板同时具有多个正例样本图像的特征，减少漏检的情况。The union operation can make the detection template have the features of multiple positive sample images at the same time, reducing the possibility of missed detection.
步骤二三、将正例样本图像的分布概率形式直方图与反例样本图像的分布概率形式直方图进行差运算，减少模型对环境的错误识别，提高准确率，如公式(4)所示：Step 2: Perform a difference operation on the distribution probability form histogram of the positive sample image and the distribution probability form histogram of the negative sample image to reduce the model's misrecognition of the environment and improve the accuracy, as shown in formula (4):
其中，Ht(j)是正例样本图像的分布概率形式直方图中的第j个区块，Hf(j)是反例样本图像的分布概率形式直方图中的第j个区块。Wherein, Ht (j) is the jth block in the distribution probability form histogram of the positive sample image, and Hf (j) is the jth block in the distribution probability form histogram of the counter-example sample image.
差运算保留了模板中与环境中不相同的色相与饱和度区间，删除了与环境相同的部分，确保了在检测过程中不受到环境的干扰。The difference operation retains the hue and saturation ranges in the template that are different from those in the environment, and deletes the parts that are the same as the environment, ensuring that the detection process is not disturbed by the environment.
步骤三、通过监控、航拍等途径采集施工现场视频用于识别检测，根据识别需求对视频进行抽帧采样，获得待测图像，如图1所示；利用步骤二获得的学习后的分布概率形式二维分布直方图对待测图像进行反向投影，获得目标概率分布矩阵P，然后利用高斯滤波算子对目标概率分布矩阵进行卷积，提取区域目标分布概率信息；Step 3: Collect construction site videos for identification and detection through monitoring, aerial photography, etc., and sample the video frames according to the identification requirements to obtain the image to be tested, as shown in Figure 1; use the learned distribution probability form two-dimensional distribution histogram obtained in step 2 to reversely project the image to be tested to obtain the target probability distribution matrix P, and then use the Gaussian filter operator to convolve the target probability distribution matrix to extract the regional target distribution probability information;
所述高斯滤波算子，如下式：The Gaussian filter operator is as follows:
其中，滤波方差σ2取1，滤波器半径取3。滤波器半径可根据图像分辨率及需求进行调整，滤波器半径越大，过滤噪点能力越强，准确率越高，检测范围会缩小，x'是滤波器的横轴尺寸，y'是滤波器的纵轴尺寸。Among them, the filter variance σ2 is 1, and the filter radius is 3. The filter radius can be adjusted according to the image resolution and requirements. The larger the filter radius, the stronger the ability to filter noise, the higher the accuracy, and the smaller the detection range. x' is the horizontal axis size of the filter, and y' is the vertical axis size of the filter.
步骤四、将步骤三获得的区域目标分布概率信息与预设阈值比较，将待测图像中概率大于预设阈值区域的图像掩膜置1，小于等于预设阈值区域的图像掩膜保持0值，获得初步识别结果；然后通过滑动窗口对初始识别结果中掩膜进行区域生长，获得掩膜检测结果，如图5-6所示；Step 4: Compare the regional target distribution probability information obtained in step 3 with the preset threshold, set the image mask of the area with a probability greater than the preset threshold in the image to be tested to 1, and keep the image mask of the area less than or equal to the preset threshold to 0, and obtain a preliminary recognition result; then perform regional growth on the mask in the initial recognition result through a sliding window to obtain a mask detection result, as shown in Figure 5-6;
滑动窗口取21×21的矩形窗口，对图像进行拓展后计算窗口内识别目标概率的均值与方差，通过下式(6)、(7)、(8)对包含在范围内的其余像素进行掩膜生长：The sliding window takes a 21×21 rectangular window, expands the image, and calculates the mean and variance of the probability of identifying the target in the window. The remaining pixels in the range are masked using the following equations (6), (7), and (8):
其中，P为反向投影得到的目标分布概率，M(x1,y1)通过阈值筛选的待测图像掩膜，为M的非，即未被检测到的待测图像掩膜，W为滑动窗口当前所在像素区间，Num为掩膜M内标记像素的数量；Model为最终检测的掩膜结果，beta是超参数，x1、y1滑动窗口内像素的横纵坐标，ave是均值，std是方差。Among them, P is the target distribution probability obtained by back projection, M(x1,y1) is the mask of the image to be tested filtered by the threshold, is the negation of M, that is, the mask of the image to be tested that has not been detected, W is the pixel interval where the sliding window is currently located, Num is the number of labeled pixels in the mask M; Model is the mask result of the final detection, beta is a hyperparameter, x1 and y1 are the horizontal and vertical coordinates of the pixels in the sliding window, ave is the mean, and std is the variance.
步骤五、利用步骤四获得的掩膜检测结果获得待测图像中的烟尘占比，具体为：Step 5: Use the mask detection result obtained in step 4 to obtain the smoke and dust ratio in the image to be tested, specifically:
步骤五一、对步骤四获得的掩膜检测结果进行腐蚀与膨胀，去除检测噪点，获得去除检测噪点后的掩膜检测结果Model′，具体为：Step 51: Corrode and dilate the mask detection result obtained in step 4 to remove the detection noise points, and obtain the mask detection result Model′ after removing the detection noise points, which is specifically:
腐蚀与膨胀算子采用线性、半径为4的球形算子，分别采取一次开运算与闭运算，填补图像掩膜孔洞同时清除离散点；The erosion and dilation operators use linear and spherical operators with a radius of 4, and take one opening and one closing operation respectively to fill the holes in the image mask and remove discrete points;
步骤五二、计算待测图像烟尘占比C，计算公式如下式(9)：Step 52: Calculate the smoke proportion C of the image to be tested, and the calculation formula is as follows (9):
式中，row为待测图像的行数，col为待测图像的列数，Model′是去除检测噪点后的掩膜检测结果，Sum()是求和函数。Wherein, row is the number of rows of the image to be tested, col is the number of columns of the image to be tested, Model′ is the mask detection result after removing the detection noise, and Sum() is the summation function.
通过设定烟尘预警阈值，来判断当前图像中是否存在扬尘、火灾等情况，同时得到烟尘占比情况。By setting the smoke warning threshold, we can determine whether there is dust, fire, etc. in the current image, and get the smoke ratio.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310792252.1ACN116824258B (en) | 2023-06-30 | 2023-06-30 | A construction site smoke detection method based on back projection |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310792252.1ACN116824258B (en) | 2023-06-30 | 2023-06-30 | A construction site smoke detection method based on back projection |
| Publication Number | Publication Date |
|---|---|
| CN116824258A CN116824258A (en) | 2023-09-29 |
| CN116824258Btrue CN116824258B (en) | 2024-05-14 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310792252.1AActiveCN116824258B (en) | 2023-06-30 | 2023-06-30 | A construction site smoke detection method based on back projection |
| Country | Link |
|---|---|
| CN (1) | CN116824258B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101336856A (en)* | 2008-08-08 | 2009-01-07 | 西安电子科技大学 | Information Acquisition and Transmission Method of Auxiliary Vision System |
| CN104077779A (en)* | 2014-07-04 | 2014-10-01 | 中国航天科技集团公司第五研究院第五一三研究所 | Moving object statistical method with Gaussian background model and mean value shift tracking combined |
| CN105469105A (en)* | 2015-11-13 | 2016-04-06 | 燕山大学 | Cigarette smoke detection method based on video monitoring |
| CN108765443A (en)* | 2018-05-22 | 2018-11-06 | 杭州电子科技大学 | A kind of mark enhancing processing method of adaptive color Threshold segmentation |
| JP2019046039A (en)* | 2017-08-31 | 2019-03-22 | ホーチキ株式会社 | Fire detection device and fire detection method |
| JP2019178982A (en)* | 2018-03-30 | 2019-10-17 | 日本製鉄株式会社 | Dust fall-amount estimation method |
| CN110379514A (en)* | 2019-07-12 | 2019-10-25 | 天津市德安圣保安全卫生评价监测有限公司 | A kind of welding fume occupational disease hazards methods of risk assessment |
| CN110909756A (en)* | 2018-09-18 | 2020-03-24 | 苏宁 | Convolutional neural network model training method and device for medical image recognition |
| CN111210452A (en)* | 2019-12-30 | 2020-05-29 | 西南交通大学 | A Portrait Segmentation Method Based on Graph Cut and Mean Shift |
| CN111860533A (en)* | 2019-04-30 | 2020-10-30 | 深圳数字生命研究院 | Image recognition method and device, storage medium and electronic device |
| CN113033892A (en)* | 2021-03-23 | 2021-06-25 | 河海大学 | Dynamic evaluation method for credit of main body of construction market under government supervision view angle |
| CN113496159A (en)* | 2020-03-20 | 2021-10-12 | 昆明理工大学 | Multi-scale convolution and dynamic weight cost function smoke target segmentation method |
| WO2021254205A1 (en)* | 2020-06-17 | 2021-12-23 | 苏宁易购集团股份有限公司 | Target detection method and apparatus |
| CN114970956A (en)* | 2022-04-19 | 2022-08-30 | 北京理工大学 | OLS and QOIC-based population association analysis and prediction method |
| CN115018751A (en)* | 2021-03-03 | 2022-09-06 | 上海新氦类脑智能科技有限公司 | A method and system for crack detection based on Bayesian density analysis |
| CN115731188A (en)* | 2022-11-22 | 2023-03-03 | 浙江浙大鸣泉科技有限公司 | Method for detecting smoke in ambient air based on Ringelmann blackness |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070083114A1 (en)* | 2005-08-26 | 2007-04-12 | The University Of Connecticut | Systems and methods for image resolution enhancement |
| WO2014165286A1 (en)* | 2013-03-12 | 2014-10-09 | Iowa State University Research Foundation, Inc. | Systems and methods for recognizing, classifying, recalling and analyzing information utilizing ssm sequence models |
| CN107301651A (en)* | 2016-04-13 | 2017-10-27 | 索尼公司 | Object tracking apparatus and method |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101336856A (en)* | 2008-08-08 | 2009-01-07 | 西安电子科技大学 | Information Acquisition and Transmission Method of Auxiliary Vision System |
| CN104077779A (en)* | 2014-07-04 | 2014-10-01 | 中国航天科技集团公司第五研究院第五一三研究所 | Moving object statistical method with Gaussian background model and mean value shift tracking combined |
| CN105469105A (en)* | 2015-11-13 | 2016-04-06 | 燕山大学 | Cigarette smoke detection method based on video monitoring |
| JP2019046039A (en)* | 2017-08-31 | 2019-03-22 | ホーチキ株式会社 | Fire detection device and fire detection method |
| JP2019178982A (en)* | 2018-03-30 | 2019-10-17 | 日本製鉄株式会社 | Dust fall-amount estimation method |
| CN108765443A (en)* | 2018-05-22 | 2018-11-06 | 杭州电子科技大学 | A kind of mark enhancing processing method of adaptive color Threshold segmentation |
| CN110909756A (en)* | 2018-09-18 | 2020-03-24 | 苏宁 | Convolutional neural network model training method and device for medical image recognition |
| CN111860533A (en)* | 2019-04-30 | 2020-10-30 | 深圳数字生命研究院 | Image recognition method and device, storage medium and electronic device |
| CN110379514A (en)* | 2019-07-12 | 2019-10-25 | 天津市德安圣保安全卫生评价监测有限公司 | A kind of welding fume occupational disease hazards methods of risk assessment |
| CN111210452A (en)* | 2019-12-30 | 2020-05-29 | 西南交通大学 | A Portrait Segmentation Method Based on Graph Cut and Mean Shift |
| CN113496159A (en)* | 2020-03-20 | 2021-10-12 | 昆明理工大学 | Multi-scale convolution and dynamic weight cost function smoke target segmentation method |
| WO2021254205A1 (en)* | 2020-06-17 | 2021-12-23 | 苏宁易购集团股份有限公司 | Target detection method and apparatus |
| CN115018751A (en)* | 2021-03-03 | 2022-09-06 | 上海新氦类脑智能科技有限公司 | A method and system for crack detection based on Bayesian density analysis |
| CN113033892A (en)* | 2021-03-23 | 2021-06-25 | 河海大学 | Dynamic evaluation method for credit of main body of construction market under government supervision view angle |
| CN114970956A (en)* | 2022-04-19 | 2022-08-30 | 北京理工大学 | OLS and QOIC-based population association analysis and prediction method |
| CN115731188A (en)* | 2022-11-22 | 2023-03-03 | 浙江浙大鸣泉科技有限公司 | Method for detecting smoke in ambient air based on Ringelmann blackness |
| Title |
|---|
| Fiscal sustainability，volatility and oil wealth:A stochastic analysis of fiscal spending rules;Sweder van Wijnbergen等;《Tinbergen Institute Discussion Paper》;20110430;第1-31页* |
| 基于机器视觉的按压泵缺陷检测系统研究与实现;戴凌峰;《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》;20210215(第02期);第C029-497页* |
| Publication number | Publication date |
|---|---|
| CN116824258A (en) | 2023-09-29 |
| Publication | Publication Date | Title |
|---|---|---|
| CN114897816B (en) | Mask R-CNN mineral particle recognition and particle size detection method based on improved mask | |
| CN108710865B (en) | A method for detecting abnormal behavior of drivers based on neural network | |
| CN104408707B (en) | Rapid digital imaging fuzzy identification and restored image quality assessment method | |
| Vorobel et al. | Segmentation of rust defects on painted steel surfaces by intelligent image analysis | |
| CN113393426B (en) | Steel rolling plate surface defect detection method | |
| CN112149543B (en) | Building dust recognition system and method based on computer vision | |
| CN108875600A (en) | A kind of information of vehicles detection and tracking method, apparatus and computer storage medium based on YOLO | |
| CN104990925A (en) | Defect detecting method based on gradient multiple threshold value optimization | |
| CN101699469A (en) | Method for automatically identifying action of writing on blackboard of teacher in class video recording | |
| Khalifa et al. | Malaysian Vehicle License Plate Recognition. | |
| CN106682278B (en) | Device and method for determining accuracy of supersonic flow field prediction based on image processing | |
| CN108509950B (en) | Railway contact net support number plate detection and identification method based on probability feature weighted fusion | |
| CN111738931B (en) | Shadow Removal Algorithm for Photovoltaic Array UAV Aerial Imagery | |
| CN108629762A (en) | A kind of stone age evaluation and test model reduces the image pre-processing method and system of interference characteristic | |
| Lee et al. | Adaptive local binarization method for recognition of vehicle license plates | |
| CN118735915A (en) | A paint detection method and detection system based on multi-dimensional visual analysis | |
| CN111768455A (en) | An Image-Based Method for Wood Region and Dominant Color Extraction | |
| CN105913008A (en) | Crowd exceptional event detection method based on hypothesis examination | |
| CN105138984A (en) | Sharpened image identification method based on multi-resolution overshoot effect measurement | |
| CN115731493A (en) | Rainfall micro physical characteristic parameter extraction and analysis method based on video image recognition | |
| CN114863330A (en) | A target object detection method, system and computer storage medium | |
| CN113177552B (en) | License plate recognition method based on deep learning | |
| CN116824258B (en) | A construction site smoke detection method based on back projection | |
| CN110321869A (en) | Personnel's detection and extracting method based on Multiscale Fusion network | |
| CN112927459B (en) | Sudoku fire behavior prediction method based on unmanned aerial vehicle vision and application |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |