Motion extraction method without trajectory spatiotemporal statistics and wildfire motion extraction method

Abstract

The present invention discloses a method for motion extraction of trackless spatiotemporal statistical data, including obtaining original trackless spatiotemporal statistical data; establishing a density map of the starting frame and a density map of the ending frame of the data; establishing a constraint map for constraining the motion extraction process; selecting an optimal transmission method to extract motion information and visualize it, obtaining a density map and a vector field form, and completing the motion extraction of trackless spatiotemporal statistical data. The present invention also provides a wildfire motion extraction method including the motion extraction method of trackless spatiotemporal statistical data. Based on an innovative algorithm, the present invention can extract motion information that is physically feasible, easy to understand, and takes into account environmental factors and user domain knowledge; the method of the present invention can extract motion information of trackless spatiotemporal statistical data, not only taking environmental factors into account, but also having high reliability and good practicality.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a motion extraction method without trajectory spatiotemporal statistical data and a wildfire motion extraction method. Background Art

[0002] Since spatiotemporal statistics record the location and time information of objects or events, they are generally considered to be an important data source for visualizing and understanding object motion. Generally speaking, spatiotemporal statistics can be visualized as a series of density maps; for complex motion, users generally need a series of temporally dense data frames to better understand the motion process. However, due to data requirements and storage space limitations, the stored data frames are usually temporally sparse. Therefore, there is a great need for a technology that can extract and visualize object motion from spatiotemporal statistics for further visual analysis.

[0003] Extracting motion between two static frames without trajectory information and origin-destination mapping is a very difficult task; the reason is that these frames do not contain explicit motion information, such as trajectory or origin-destination information. For example, air pollution data collected by distributed static sensors does not contain explicit information about the motion of pollutants. Recently, two interpolation-based methods have been proposed to extract motion from temporally sparse frames. The first method is based on a diffusion model: it assumes that the object motion process is a diffusion process and uses it as a constraint to interpolate between two density maps. The extracted motion is implicitly defined by the interpolated frames; the second method is based on a data-driven model: it transforms the motion extraction problem into an image generation problem, calculates the possible dynamic changes of the image, and then interpolates smoothly deformed interpolated frames to generate perceptually smooth motion patterns.

[0004] However, both existing methods have problems: first, the movement may not be physically feasible, for example, the object movement process may not be a diffusion process; second, the existing methods do not consider environmental factors that may affect the movement of objects. For example, the diffusion of air pollution may be affected by terrain factors, and urban traffic must be constrained by road design. In addition, for more complex scenarios, it is critical to support user-guided interactive processes so that users can incorporate their own understanding of environmental factors into the results. However, existing methods do not support user participation. Summary of the invention

[0005] One of the purposes of the present invention is to provide a motion extraction method for trackless spatiotemporal statistical data which takes environmental factors into consideration and has high reliability and good practicability.

[0006] A second object of the present invention is to provide a wildfire motion extraction method that includes the motion extraction method without track-free spatiotemporal statistical data.

[0007] The motion extraction method without trajectory spatiotemporal statistical data provided by the present invention comprises the following steps:

[0008] S1. Obtain original trajectory-free spatiotemporal statistics;

[0009] S2. According to the data obtained in step S1, a density map of the starting frame and a density map of the ending frame of the data are established;

[0010] S3. Establish a constraint graph for constraining the motion extraction process based on user domain knowledge and environmental context information;

[0011] S4. Based on the constraint graph obtained in step S3, select the optimal transmission method to extract motion information;

[0012] S5. Visualize the motion information obtained in step S4 to obtain a density map and a vector field, thereby completing the motion extraction without trajectory spatiotemporal statistical data.

[0013] The step S2 is specifically to use a kernel density estimation algorithm for each data point based on the data obtained in step S1 and based on the same kernel radius and Gaussian kernel function, so as to obtain a starting frame density map and an ending frame density map of the data.

[0014] The step S3 is specifically to calculate the shortest path distance between the starting frame and the ending frame point pairs and generate a distance matrix based on the user's domain knowledge and environmental context information, using the Euclidean distance measurement method, the density-weighted shortest path measurement method, or the custom-weighted shortest path measurement method, and use the obtained distance matrix as a constraint graph for constraining the motion extraction process.

[0015] The step S3 specifically includes the following steps:

[0016] A. Select the corresponding measurement method according to the following rules:

[0017] If there is no need to consider the impact of external factors (such as environmental factors) on movement, the selected measurement method is the Euclidean distance measurement method; for example, for a data set with a very short time interval, the movement can be approximated by a straight line, so using the Euclidean distance as a measurement of the transmission cost is intuitive and easy to explain; another scenario where this measurement is applicable is that the movement is mainly based on diffusion and is less restricted by the environment;

[0018] If the effect of external factors (such as environmental factors) on movement cannot be determined, the metric of choice is the shortest path metric based on density weights; for example, the movement of fish schools is affected by ocean currents, and the spread of air pollution is driven by wind fields; but analysts may not be able to determine how these factors affect movement; for this case, it can be simply assumed that the distribution reflects hidden factors, or that mass tends to be transmitted through locations that are already high in density; under this assumption, the metric is set to the shortest path length, with density-based weights;

[0019] If the impact of external factors (such as environmental factors) on movement can be determined, the metric of choice is the shortest path metric based on custom weights; with domain knowledge, analysts can freely customize the weight map; for example, knowing that a mountain can block the movement of air pollutants, analysts can customize the weight map to extract the movement of pollutants, which will be closer to reality;

[0020] B. Generate the corresponding distance matrix based on the measurement method selected in step A:

[0021] C. Use the distance matrix obtained in step B as the final constraint graph.

[0022] The step B specifically comprises the following steps:

[0023] If the selected measurement method is the Euclidean distance measurement method, the Euclidean distance between the starting frame and the ending frame point pair is directly calculated to generate a distance matrix;

[0024] If the selected metric method is the density-weighted shortest path metric method, the cost value of each pixel grid is first calculated based on the density formula to generate a cost map; then the obtained cost map is used as the input of the density-weighted shortest path metric method to calculate the shortest distance between the starting frame and the ending frame point pair, thereby generating a distance matrix;

[0025] If the selected metric method is the shortest path metric method based on custom weights, first load the environmental context information of the data; then set the cost value of each map network on the local map where the data is located to generate a cost map; finally, use the obtained cost map as the input of the shortest path metric method based on custom weights to calculate the shortest distance between the starting frame and the ending frame point pairs, thereby generating a distance matrix.

[0026] The density-based formula is used to calculate the cost value of each pixel grid, specifically using the following formula to calculate the cost value loss(x) of each pixel network:

[0027]

[0028] Where x is the pixel network currently being calculated; ρ(x) is the density value of the x pixel grid; ε is the set regularization term, which is used to ensure that the denominator in the formula is not 0.

[0029] The step S4 specifically takes the starting frame density map, the ending frame density map and the constraint map as input, and uses the iterative Bregman projection algorithm to calculate the mapping relationship between the starting frame and the ending frame point pairs; then based on the obtained mapping relationship, a number of interpolation frames located between the starting frame and the ending frame are generated, thereby completing the extraction of motion information.

[0030] The iterative Bregman projection algorithm specifically uses the following formula as the calculation formula of the iterative Bregman projection algorithm:

[0031]

[0032] Where W p (u,v) is the p-Wasserstein distance; inf is the infimum; π is a transmission scheme in the set of transmission schemes; Π(u,v) is the set of all transmission schemes; Ω×Ω is the product space of the probability measure Ω; c(x,y) is the transmission cost of quality from position x to position y; p is the order; π(x,y) is the transmission quality from position x to position y; γ is the influence coefficient; H(π) is the entropy function.

[0033] The generating of the plurality of interpolation frames between the start frame and the end frame is specifically generating the plurality of interpolation frames between the start frame and the end frame based on the following calculation formula:

[0034] If the quality of the start frame and the end frame is the same, the following formula is used to calculate the number of interpolation frames between the start frame and the end frame:

[0035]

[0036] Where μ t (z) is the mass value at position z at time t; z is the position of the mass at time t when it moves along the geodesic path from position x to position y; geo (x,y) (t) is the geodesic position calculation function; π(x,y) is the transmission quality from position x to position y;

[0037] If the quality of the start frame and the end frame is inconsistent, the following formula is used to calculate several interpolation frames between the start frame and the end frame:

[0038]

[0039] Where r(n) is the function for calculating the rate of mass change, and M t is the total quality of the termination frame data, Ms is the total quality of the starting frame data, n is the number of interpolation frames; t(n+1) is the time coefficient.

[0040] The step S5 is specifically to visualize the motion information based on the mass distribution information of each frame, the mass value and the motion speed, using the kernel density estimation algorithm and the arrow construction algorithm to obtain the density map and the vector field form, thereby completing the motion extraction of trackless spatiotemporal statistical data.

[0041] The density map is specifically an adaptive kernel density estimation method based on environmental constraints, which is calculated using the following formula to constrain the density area to a physically accessible area:

[0042]

[0043] Where d(x,y) is the density value at position (x,y), u is the attenuation factor, and the value of u is the distance between the constraint pixel closest to (x,y) and (x,y).

[0044] The vector field form is specifically calculated using the following formula to obtain the vector field form:

[0045]

[0046] Where V t (z) is the vector field at position z; z is the position of the mass at time t when it moves along the geodesic path from position x to position y; geo (x,y) (t) is the geodesic position calculation function; π(x,y) is the transmission quality from position x to position y; ε is the set time interval.

[0047] The present invention also provides a wildfire motion extraction method including the motion extraction method without trajectory spatiotemporal statistical data, which specifically includes the following steps:

[0048] S1. Obtain the distribution data of current wildfires in the target area as the original trackless spatiotemporal statistics;

[0049] S2. Using the above-mentioned motion extraction method of non-track spatiotemporal statistical data, complete the motion extraction of non-track spatiotemporal statistical data;

[0050] S3. The motion extraction result obtained in step S2 is used as the final wildfire motion extraction result.

[0051] The motion extraction method for trackless spatiotemporal statistical data and the wildfire motion extraction method provided by the present invention are based on innovative algorithms and can extract motion information that is physically feasible, easy to understand and takes into account environmental factors and user domain knowledge; the method of the present invention can extract motion information of trackless spatiotemporal statistical data, which not only takes into account environmental factors, but also has high reliability and good practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the method flow of the motion extraction method of the present invention.

[0053] Figure 2 It is a schematic diagram of the result of using the Euclidean distance metric of the motion extraction method of the present invention.

[0054] Figure 3 It is a schematic diagram of the result of using the density-based shortest path metric of the motion extraction method of the present invention.

[0055] Figure 4 It is a schematic diagram of the result of using the shortest path metric based on custom weights in the motion extraction method of the present invention.

[0056] Figure 5 It is a schematic diagram for comparing the adaptive kernel density estimation method of the motion extraction method of the present invention with the existing method.

[0057] Figure 6 It is a schematic diagram comparing the extraction results of the motion extraction method of the present invention and the existing method.

[0058] Figure 7 The figure is a schematic diagram of the method flow of the wildfire movement extraction method of the present invention. DETAILED DESCRIPTION

[0059] like Figure 1 The figure shows a flow chart of the motion extraction method of the present invention: the motion extraction method without trajectory spatiotemporal statistical data provided by the present invention comprises the following steps:

[0060] S1. Obtain original trajectory-free spatiotemporal statistics;

[0061] S2. According to the data obtained in step S1, a density map of the starting frame and a density map of the ending frame of the data are established; specifically, according to the data obtained in step S1, based on the same kernel radius and Gaussian kernel function, a kernel density estimation algorithm is used for each data point to obtain a density map of the starting frame and a density map of the ending frame of the data;

[0062] S3. Establish a constraint graph for constraining the motion extraction process according to the user domain knowledge and environmental context information; specifically, according to the user domain knowledge and environmental context information, use the Euclidean distance measurement method, the density-weighted shortest path measurement method, or the custom-weighted shortest path measurement method to calculate the shortest path distance between the start frame and the end frame point pair and generate a distance matrix, and use the obtained distance matrix as a constraint graph for constraining the motion extraction process;

[0063] The specific implementation includes the following steps:

[0064] A. Select the corresponding measurement method according to the following rules:

[0065] If there is no need to consider the impact of external factors (such as environmental factors) on movement, the selected measurement method is the Euclidean distance measurement method; for example, for a data set with a very short time interval, the movement can be approximated by a straight line, so using the Euclidean distance as a measurement of the transmission cost is intuitive and easy to explain; another scenario where this measurement is applicable is that the movement is mainly based on diffusion and is less restricted by the environment;

[0066] If the effect of external factors (such as environmental factors) on movement cannot be determined, the metric of choice is the shortest path metric based on density weights; for example, the movement of fish schools is affected by ocean currents, and the spread of air pollution is driven by wind fields; but analysts may not be able to determine how these factors affect movement; for this case, it can be simply assumed that the distribution reflects hidden factors, or that mass tends to be transmitted through locations that are already high in density; under this assumption, the metric is set to the shortest path length, with density-based weights;

[0067] If the impact of external factors (such as environmental factors) on movement can be determined, the metric of choice is the shortest path metric based on custom weights; with domain knowledge, analysts can freely customize the weight map; for example, knowing that a mountain can block the movement of air pollutants, analysts can customize the weight map to extract the movement of pollutants, which will be closer to reality;

[0068] B. Generate the corresponding distance matrix according to the measurement method selected in step A: specifically, the following steps are included:

[0069] If the selected measurement method is the Euclidean distance measurement method, the Euclidean distance between the starting frame and the ending frame point pair is directly calculated to generate a distance matrix;

[0070] Figure 2 This is a schematic diagram of the results using the Euclidean distance metric: This figure shows the movement of air pollution in Central China and East China from 15:00 on November 2, 2019 to 23:00 on November 3, 2019; Figure 2 (a) and Figure 2 (b) is a pollutant distribution density diagram representing the start and end time of the air quality index distribution at the beginning; Figure 2 (c) is the wind field during the same period; Figure 2 (d) is the average vector field of the extracted air pollution motion; Figure 2 (e1)~ Figure 2 (e4) is the interpolated density map; Figure 2 (f1)~ Figure 2 (f4) is the vector field corresponding to the interpolated frame;

[0071] If the selected metric method is the density-weighted shortest path metric method, the cost value of each pixel grid is first calculated based on the density formula to generate a cost map; then the obtained cost map is used as the input of the density-weighted shortest path metric method to calculate the shortest distance between the starting frame and the ending frame point pair, thereby generating a distance matrix;

[0072] Among them, the cost value of each pixel grid is calculated based on the density formula. Specifically, the cost value loss(x) of each pixel network is calculated using the following formula:

[0073]

[0074] Where x is the pixel network currently being calculated; ρ(x) is the density value of the x pixel grid; ε is the set regularization term, which is used to ensure that the denominator in the formula is not 0;

[0075] Figure 3 A diagram showing the results of using the density-weighted shortest path metric: This diagram shows the spread of fire in the southern part of Kakadu National Park in Australia; Figure 3 (a) is the density map starting on October 3, 2019; Figure 3 (b)~ Figure 3 (g) is the interpolated frame; Figure 3 (h) is the density map ending on October 5, 2019;

[0076] If the selected metric method is the shortest path metric method based on custom weights, first load the environmental context information of the data; then set the cost value of each map network on the local map where the data is located to generate a cost map; finally, use the obtained cost map as the input of the shortest path metric method based on custom weights to calculate the shortest distance between the starting frame and the ending frame point pair, thereby generating a distance matrix;

[0077] Figure 4 The result graph using the shortest path metric based on custom weights: This graph shows the flow of people around Meixi Lake in Changsha, China; Figure 4(a) is the density map starting at 17:00 on April 4, 2020; Figure 4 (b) The map ends at 21:00 on April 4, 2020; Figure 4 (c) is the first round weight graph when the cost of setting lakes and rivers is higher except for the two bridges; Figure 4 (d) is also the weight graph of the second round when setting up a bridge at high cost; Figure 4 (e1)~ Figure 4 (e4) is the interpolated frame of the first round; Figure 4 (f1)~ Figure 4 (f4) is the interpolated frame of the second round;

[0078] C. Use the distance matrix obtained in step B as the final constraint graph;

[0079] S4. Based on the constraint graph obtained in step S3, an optimal transmission method is selected to extract motion information; specifically, the starting frame density map, the ending frame density map and the constraint graph are used as inputs, and an iterative Bregman projection algorithm is used to calculate the mapping relationship between the starting frame and the ending frame point pairs; then, based on the obtained mapping relationship, a number of interpolation frames located between the starting frame and the ending frame are generated, thereby completing the extraction of motion information;

[0080] In specific implementation, the following formula is used as the calculation formula of the iterative Bregman projection algorithm:

[0081]

[0082] Where W p (u,v) is the p-Wasserstein distance; inf is the infimum; π is a transmission scheme in the set of transmission schemes; Π(u,v) is the set of all transmission schemes; Ω×Ω is the product space of the probability measure Ω; c(x,y) is the transmission cost of quality from position x to position y; p is the order; π(x,y) is the transmission quality from position x to position y; γ is the influence coefficient; H(π) is the entropy function;

[0083] Generate a plurality of interpolation frames between the start frame and the end frame, specifically generate a plurality of interpolation frames between the start frame and the end frame based on the following calculation formula:

[0084] If the quality of the start frame and the end frame is the same, the following formula is used to calculate the number of interpolation frames between the start frame and the end frame:

[0085]

[0086] Where μ t (z) is the mass value at position z at time t; z is the position of the mass at time t when it moves along the geodesic path from position x to position y; geo(x,y) (t) is the geodesic position calculation function; π(x,y) is the transmission quality from position x to position y;

[0087] If the quality of the start frame and the end frame is inconsistent, the following formula is used to calculate several interpolation frames between the start frame and the end frame:

[0088]

[0089] Where r(n) is the function for calculating the rate of mass change, and M t is the total quality of the termination frame data, M s is the total quality of the starting frame data, n is the number of interpolation frames; t(n+1) is the time coefficient;

[0090] S5. Visualize the motion information obtained in step S4 to obtain a density map and a vector field form, thereby completing the motion extraction of trackless spatiotemporal statistical data; specifically, based on the mass distribution information of each frame, the mass value and the motion speed, a kernel density estimation algorithm and an arrow construction algorithm are used to visualize the motion information to obtain a density map and a vector field form, thereby completing the motion extraction of trackless spatiotemporal statistical data;

[0091] Specifically, an adaptive kernel bandwidth based on constraint graph is proposed; for each pixel e, the distance d(e) to its nearest constraint pixel is calculated; in specific implementation, the upper limit of d(e) is set to 35 pixels; the kernel bandwidth at e is proportional to d(e);

[0092] The density map is specifically an adaptive kernel density estimation method based on environmental constraints, which uses the following formula for calculation, thereby constraining the density area to a physically reachable area:

[0093]

[0094] Where d(x,y) is the density value at position (x,y), u is the attenuation factor, and the value of u is the distance between the constraint pixel closest to (x,y) and (x,y);

[0095] Among them, the adaptive kernel density estimation method can constrain the density area to a physically reachable area to avoid visual errors; Figure 5 Schematic diagram for comparing the adaptive kernel density estimation method with existing methods: Figure 5 The dots in (a) indicate people’s locations; Figure 5 (b) is the estimation result obtained by the kernel density estimation method using the same kernel bandwidth; Figure 5 (c) is the estimation result obtained by using the kernel density estimation method with mask; Figure 5(d) is the estimation result obtained by the adaptive kernel density estimation method of the present invention; Figure 5 (a)~ Figure 5 (d) It can be seen that the adaptive kernel density estimation method proposed in the present invention can obtain better prediction results;

[0096] The vector field form is specifically calculated using the following formula to obtain the vector field form:

[0097]

[0098] Where V t (z) is the vector field at position z; z is the position of the mass at time t when it moves along the geodesic path from position x to position y; geo (x,y) (t) is the geodesic position calculation function; π(x,y) is the transmission quality from position x to position y; ε is the set time interval.

[0099] The effect of the method of the present invention is further described below in conjunction with an embodiment: Taking the movement of water droplets as an example: Figure 6 is a schematic diagram comparing the extraction results of the motion extraction method of the present invention and the existing method, Figure 6 (a) shows the actual process of water drop falling; Figure 6 (b) is the result obtained by the motion extraction method based on the diffusion model; Figure 6 (c) is the result obtained by the motion extraction method based on the data-driven model; Figure 6 (d) is the result obtained by the motion extraction method of the present invention. Figure 6 It can be seen that the effect of the method of the present invention is closest to the actual situation, and therefore has the best effect.

[0100] In addition, combined with Figure 4 , the present invention is further described:

[0101] Firstly, based on the distribution data of the crowd at two different times, the density graphs of the start frame and the end frame of the data are established;

[0102] Then, based on user knowledge, humans cannot walk on water, so the water area is set to a high cost through user interaction ( Figure 4 (c)) to construct a constraint graph to constrain crowd movement;

[0103] Next, based on the constraint graph, the optimal transmission method is used to extract movement information such as the trajectory of the crowd and the origin-destination mapping relationship;

[0104] Then, the extracted motion information is visualized as a density map; it can be found that a large group of people moved from the lower left area to the upper left area through Bridge No. 1. Another group of people on the right side of the lake walked through Bridge No. 2 and gathered in the upper left area (corresponding to Figure 4 (e1)~ Figure 4 (e4));

[0105] Finally, we learned that Bridge 1 is for vehicles only and pedestrians cannot use this bridge, so we edited the constraint graph again to set the Bridge 1 area to a high cost (corresponding to Figure 4 (d)), and re-execute the optimal transmission method to obtain motion information; the new result shows that the crowd walks around the right side of the lake and reaches the destination through Bridge 2 (corresponding to Figure 4 (f1)~ Figure 4 (f4)).

[0106] like Figure 7 The method flow chart of the wildfire motion extraction method of the present invention is shown as follows: The wildfire motion extraction method provided by the present invention, which includes the motion extraction method without trajectory spatiotemporal statistical data, specifically includes the following steps:

[0107] S1. Obtain the distribution data of current wildfires in the target area as the original trackless spatiotemporal statistics;

[0108] S2. Using the above-mentioned motion extraction method of non-track spatiotemporal statistical data, complete the motion extraction of non-track spatiotemporal statistical data;

[0109] S3. The motion extraction result obtained in step S2 is used as the final wildfire motion extraction result.

[0110] In specific implementation, the following steps can be used to extract wildfire movement:

[0111] S1. Obtain wildfire data in the southern part of Kakadu National Park, Australia, including the distribution of wildfires at 14:00 on October 3, 2019 and 10:00 on October 4, 2019; and use this as the original trajectory-free spatiotemporal statistical data;

[0112] S2. Using the above-mentioned motion extraction method of trackless spatiotemporal statistical data, complete the motion extraction of trackless spatiotemporal statistical data:

[0113] According to the data obtained in step S1, a starting frame density map and an ending frame density map of the wildfire data are established;

[0114] According to the user's domain knowledge, possible factors that affect the spread of wildfires include terrain, vegetation distribution, wind field, etc. However, the analyst cannot determine the impact of external factors on movement, so the analyst assumes that the distribution of fire indicates the propagation path and uses the density-weighted shortest path metric method to establish a constraint graph to constrain the movement extraction process;

[0115] Based on the obtained constraint graph, the optimal transmission method is selected to extract the motion information; in addition, due to the inconsistent quality during the fire spread process, we also used the quality mixing method mentioned above;

[0116] The obtained motion information is visualized and obtained Figure 3 The density map shown here completes the motion extraction of wildfire data.

[0117] like Figure 3 As shown in the figure, the wildfire first moves to the lower right to the green area indicating lush vegetation; then, the wildfire is divided into two parts. The spread of the left part is restricted by the river and spreads only along the river bank; the bottom part continues to move downward and when it encounters another river, it is restricted and changes direction to spread along the river bank; based on the above movement process analysis, we can conclude that the spread of fire is affected by terrain and vegetation distribution.

Claims

1. A motion extraction method without trajectory spatiotemporal statistical data, characterized in that The steps include: S1. Obtain original trajectory-free spatiotemporal statistics; S2. According to the data obtained in step S1, a density map of the starting frame and a density map of the ending frame of the data are established; S3. Establish a constraint graph for constraining the motion extraction process based on user domain knowledge and environmental context information; Specifically, according to the user's domain knowledge and environmental context information, the Euclidean distance measurement method, the density-weighted shortest path measurement method or the custom-weighted shortest path measurement method is used to calculate the shortest path distance between the start frame and the end frame point pair and generate a distance matrix, and the obtained distance matrix is ​​used as a constraint graph to constrain the motion extraction process; The specific implementation includes the following steps: A. Select the corresponding measurement method according to the following rules: If the influence of external factors on the movement does not need to be considered, the selected measurement method is the Euclidean distance measurement method; If the influence of external factors on the movement cannot be determined, the metric of choice is the density-weighted shortest path metric; If the influence of external factors on the movement can be determined, the selected metric is the shortest path metric based on custom weights; B. Generate the corresponding distance matrix according to the measurement method selected in step A: specifically, the following steps are included: If the selected measurement method is the Euclidean distance measurement method, the Euclidean distance between the starting frame and the ending frame point pair is directly calculated to generate a distance matrix; If the selected metric method is the density-weighted shortest path metric method, the cost value of each pixel grid is first calculated based on the density formula to generate a cost map; then the obtained cost map is used as the input of the density-weighted shortest path metric method to calculate the shortest distance between the starting frame and the ending frame point pair, thereby generating a distance matrix; Among them, the cost value of each pixel grid is calculated based on the density formula. Specifically, the cost value loss(x) of each pixel network is calculated using the following formula: Where x is the pixel network currently being calculated; ρ(x) is the density value of the x pixel grid; ε is the set regularization term, which is used to ensure that the denominator in the formula is not 0; If the selected metric method is the shortest path metric method based on custom weights, first load the environmental context information of the data; then set the cost value of each map network on the local map where the data is located to generate a cost map; finally, use the obtained cost map as the input of the shortest path metric method based on custom weights to calculate the shortest distance between the starting frame and the ending frame point pair, thereby generating a distance matrix; C. Use the distance matrix obtained in step B as the final constraint graph; S4. Based on the constraint graph obtained in step S3, an optimal transmission method is selected to extract motion information; specifically, the starting frame density map, the ending frame density map and the constraint graph are used as inputs, and an iterative Bregman projection algorithm is used to calculate the mapping relationship between the starting frame and the ending frame point pairs; then, based on the obtained mapping relationship, a number of interpolation frames located between the starting frame and the ending frame are generated, thereby completing the extraction of motion information; S5. Visualize the motion information obtained in step S4 to obtain a density map and a vector field, thereby completing the motion extraction without trajectory spatiotemporal statistical data.

2. The motion extraction method without trajectory spatiotemporal statistical data according to claim 1 is characterized in that The step S2 is specifically to use a kernel density estimation algorithm for each data point based on the data obtained in step S1 and based on the same kernel radius and Gaussian kernel function, so as to obtain a starting frame density map and an ending frame density map of the data.

3. The motion extraction method without trajectory spatiotemporal statistical data according to claim 2 is characterized in that The iterative Bregman projection algorithm specifically uses the following formula as the calculation formula of the iterative Bregman projection algorithm: Where W p (u,v) is the p-Wasserstein distance; inf is the infimum; π is a transmission scheme in the set of transmission schemes; Π(u,v) is the set of all transmission schemes; Ω×Ω is the product space of the probability measure Ω; c(x,y) is the transmission cost of quality from position x to position y; p is the order; π(x,y) is the transmission quality from position x to position y; γ is the influence coefficient; H(π) is the entropy function.

4. The motion extraction method without trajectory spatiotemporal statistical data according to claim 3 is characterized in that The generating of the plurality of interpolation frames between the start frame and the end frame is specifically generating the plurality of interpolation frames between the start frame and the end frame based on the following calculation formula: If the quality of the start frame and the end frame is the same, the following formula is used to calculate the number of interpolation frames between the start frame and the end frame: Where μ t (z) is the mass value at position z at time t; z is the position of the mass at time t when it moves along the geodesic path from position x to position y; geo (x,y) (t) is the geodesic position calculation function; π(x,y) is the transmission quality from position x to position y; If the quality of the start frame and the end frame is inconsistent, the following formula is used to calculate several interpolation frames between the start frame and the end frame: Where r(n) is the function for calculating the rate of mass change, and M t is the total quality of the termination frame data, M s is the total quality of the starting frame data, n is the number of interpolation frames; t(n+1) is the time coefficient.

5. The motion extraction method without trajectory spatiotemporal statistical data according to claim 4 is characterized in that The step S5 is specifically to visualize the motion information based on the mass distribution information of each frame, the mass value and the motion speed, using the kernel density estimation algorithm and the arrow construction algorithm to obtain the density map and the vector field form, thereby completing the motion extraction of trackless spatiotemporal statistical data.

6. The motion extraction method without trajectory spatiotemporal statistical data according to claim 5 is characterized in that The density map is specifically an adaptive kernel density estimation method based on environmental constraints, which is calculated using the following formula to constrain the density area to a physically accessible area: Where d(x,y) is the density value at position (x,y), u is the attenuation factor, and the value of u is the distance between the constraint pixel closest to (x,y) and (x,y); The vector field form is specifically calculated using the following formula to obtain the vector field form: Where V t (z) is the vector field at position z; z is the position of the mass at time t when it moves along the geodesic path from position x to position y; geo (x,y) (t) is the geodesic position calculation function; π(x,y) is the transmission quality from position x to position y; ε is the set time interval.

7. A method for extracting wildfire motion comprising the motion extraction method without trajectory spatiotemporal statistical data as claimed in any one of claims 1 to 6, characterized in that The specific steps include: S1. Obtain the distribution data of current wildfires in the target area as the original trackless spatiotemporal statistics; S2. Using the above-mentioned motion extraction method of non-track spatiotemporal statistical data, complete the motion extraction of non-track spatiotemporal statistical data; S3. The motion extraction result obtained in step S2 is used as the final wildfire motion extraction result.

Images