Multi-node Parallel Fusion Method for Spatiotemporal Trajectories

Abstract

A multi-node parallel fusion method for spatio-temporal trajectories disclosed by the present invention can reduce the processing time and improve the fusion efficiency. The present invention is implemented through the following technical solutions: First, use high-resolution data to search for pixels similar to the central pixel spectrum in the moving window, filter the samples and assign nodes to the tracks, assign combined weights to each similar pixel, and determine the conversion coefficient through regression analysis; the track assignment node calculates the associated area of the track according to the track association assignment rule and outputs the track to the track association node; Second, the track association node performs track association calculation according to the track association rule and outputs the track association result; Then, the track association node performs track association calculation according to the track association rule, generates the track association result, and outputs the association result to the track fusion node according to the fusion assignment rule; Finally, the track fusion node fuses the tracks according to the fusion rule to generate the final fusion result.

Description

Technical Field

[0001] The present invention relates to the field of spatio-temporal data processing, and particularly to a method for fast distributed fusion of spatio-temporal trajectories in a mobile target real-time monitoring system, and more particularly to a multi-node parallel fusion method. Background Art

[0002] With the rapid development of mobile Internet, location sensing technology, and earth observation technology, aerospace sensing data represented by mobile objects, space / spatio-temporal, and remote sensing has grown explosively. The increasingly large amount of data easily leads to the bottleneck of algorithm performance. The use process of aerospace big data is complex, with high usage thresholds and low application efficiency. Aerospace data mainly comes from space-based and airborne platforms, such as GNSS (Global Navigation Satellite System) data based on space-based platforms. Aerospace big data covers various location-related data of land, sea, air, and space, including Spatial (space, i.e., geospatial) and Space (sky, i.e., cosmic space). Aerospace data has characteristics such as diverse types, highly unstructured, large monomers, and multi-dimensions, posing great challenges to integrated data management and efficient query retrieval. In terms of storage management, traditional centralized storage highly depends on the performance of a single machine, greatly limiting the scalability of storage capacity and unable to support low-latency access and high-concurrency access to massive unstructured data. In terms of processing and analysis, traditional serial analysis algorithms can no longer meet the real-time processing requirements of massive spatio-temporal data and cannot fully utilize the advantages of current new hardware architectures and parallel models / frameworks. In specific research and applications, traditional data processing and analysis methods can no longer meet the performance requirements of efficient access and real-time processing of spatio-temporal big data. Therefore, the integration of spatio-temporal big data with high-performance computing / cloud computing is an inevitable development trend.

[0003] The number of spatio-temporal trajectories is synthesized from position coordinates and corresponding time stamps. Therefore, the time feature is crucial for analyzing the running patterns of moving objects. Spatio-temporal trajectory data records a series of information about the continuous moving object appearing at the corresponding position at a certain moment, which is the manifestation of the spatio-temporal attributes in its motion state. Applied to the actual situation, there may be a large difference in the time span of two trajectory segments. There are different orders of magnitude between the spatial distance and the time distance. The shape of the sub-trajectory segment is irregular, and the set contains a large amount of noise. Since the result quality of dividing the trajectory segment according to the change angle of adjacent sample points is low. This leads to a high QMeasure value. Traditional spatio-temporal data processing takes the geographic information system (GIS) or remote sensing image processing platform software as the core, emphasizing the platform professionalism. However, due to the professional strengthening, a semi-closed system with a high degree of professionalism is formed, which will also inversely weaken the fusion processing ability with other multi-modal type data. Traditional aerospace data applications require a large amount of slice preprocessing, resulting in inflexible data applications. To make the data more flexible, the industry has introduced an algorithm of pre-static caching + dynamic slicing, but this algorithm is very complex. The increase in devices with global positioning system (GPS) functions generates a large amount of spatio-temporal trajectory data, bringing a heavy burden to the storage, transmission, and processing of the data. Therefore, spatio-temporal trajectories are time-consuming to process when there are many targets, and the running speed is slow. The traditional single-node processing method is limited by the processor performance and cannot process the fusion calculation of a large number of targets in a timely manner. The traditional multi-node parallel processing method mainly distributes the calculation for parallel processing to improve the computing power and reduce the processing time. However, in a mobile target real-time monitoring system, the time-consuming part is the comparison with all targets, and all targets are synchronously updated in real time. At the same time, it is necessary to ensure that all calculations are carried out synchronously, so more synchronization time is required. Due to the destruction of the correlation of the trajectories before and after the segmentation point, the distributed calculation of big data and other processing platforms aimed at enhancing the computing power is difficult to be used to improve the fusion processing speed. Therefore, there is an urgent need to study a multi-node parallel fusion method that uses multiple nodes to process simultaneously, reduces the processing time, and improves the fusion efficiency. Summary of the Invention

[0004] To solve the problem of rapid fusion of such a large number of mobile targets, the present invention provides a spatio-temporal trajectory multi-node parallel fusion method that can reduce the processing time and improve the fusion efficiency, so as to solve the problem of slow fusion speed of a large number of targets.

[0005] To achieve the above object, the present invention proposes a spatio-temporal trajectory multi-node parallel fusion method, including the following steps:

[0006] Search for pixels with spectral similarity to the central pixel in the moving window using high-resolution data, filter the samples and assign nodes to the tracks, assign combined weights to each similar pixel, and determine the conversion coefficients through regression analysis; based on the data transformation model, screen and process the information of the transformed data, construct the spatio-temporal trajectory dataset, the set of feature points, and the set of sub-trajectories, determine the weights according to the weight function, and establish a spatio-temporal adaptive fusion model; the track assignment node calculates the association area of the track according to the track association assignment rule and outputs the track to the track association node; the track association node performs track association calculation according to the track association rule of fuzzy double thresholds, and according to the current track target number perform track association calculation, obtain the track association result of each frame according to the correlation and output it to the track association node; the track association node performs track association calculation according to the track association rule, and according to the current track target number , the target number of the successfully associated , the track association membership degree generate the track association result, and output the track association result to the track fusion result according to the fusion assignment rule; the track fusion result performs track fusion calculation at the track fusion node 1, the track fusion node 2... the track fusion node perform track fusion calculation to generate the final fusion result.

[0007] The present invention has the following beneficial effects compared with the prior art:

[0008] The present invention uses high-resolution data to search for pixels with spectral similarity to the central pixel in the moving window, filters the samples and assigns nodes to the tracks, assigns combined weights to each similar pixel, and determines the conversion coefficients through regression analysis; based on the data transformation model, screens and processes the information of the transformed data, constructs the spatio-temporal trajectory dataset, the set of feature points, and the set of sub-trajectories, determines the weights according to the weight function, and establishes a spatio-temporal adaptive fusion model; this parallel calculation of track association by region and parallel calculation of track fusion by number can reduce the fusion processing time and improve the fusion efficiency, and can solve the problem of low real-time fusion efficiency of a large number of targets.

[0009] The present invention adopts a track assignment node, calculates the association area of the track according to the track association assignment rule, and outputs the track to the track association node; the track association node performs track association calculation according to the track association rule of fuzzy double thresholds at the track association node 1, the track association node 2... the track association node perform track association calculation, obtain the track association result of each frame according to the correlation and output it to the track association node; the track association node performs track association calculation according to the track association rule, and according to the current track target number , the target number of the successfully associated , associated membership degree Generate track association results and output the track association results to the point track fusion node according to the fusion distribution rule; the point track fusion node performs point track fusion calculation at point track fusion node 1, point track fusion node 2... point track fusion node Perform point track fusion calculation to generate the final fusion result. Each technical node only needs to be configured with an ordinary commercial computer, with low cost and high fusion efficiency, and can improve the fusion efficiency in the absence of big data or cloud computing environment. Brief Description of the Drawings

[0010] To understand the present invention more clearly, the present invention will be described through specific implementation examples and with reference to the accompanying drawings, where:

[0011] Figure 1 is a schematic diagram of the multi-node parallel fusion process of the spatio-temporal trajectory of the present invention.

[0012] Figure 2 is a schematic diagram of the point track association calculation process.

[0013] Figure 3 is a schematic diagram of the track association calculation process.

[0014] Figure 4 is a schematic diagram of the point track fusion calculation process. Detailed Description of the Invention

[0015] Refer to Figure 1 . According to the present invention, use high-resolution data to search for pixels similar to the central pixel spectrum in the moving window, filter the samples and the point track assignment node, assign combined weight values to each similar pixel, and determine the conversion coefficient through regression analysis; based on the data transformation model, screen and process the information of the transformed data, construct a spatio-temporal trajectory data set, a set of feature points, and a set of sub-trajectories, determine the weight according to the weight function, and establish a spatio-temporal adaptive fusion model; the point track assignment node calculates the association area of the point track according to the point track association assignment rule and outputs the point track to the point track association node; the point track association node performs point track association calculation according to the point track association rule of the fuzzy double threshold at point track association node 1, point track association node 2... point track association node Perform point track association calculation, obtain the point track association result of each frame according to the correlation and output it to the track association node; the track association node performs track association calculation according to the track association rule, according to the current point track target number , the target number of successful association , point track association membership degree Generate track association results and output the track association results to the point track fusion result according to the fusion distribution rule; the point track fusion result performs point track fusion calculation at point track fusion node 1, point track fusion node 2... point track fusion node Perform track fusion calculation to generate the final fusion result.

[0016] The track assignment node calculates the associated area of the track according to the track association assignment rule :

[0017]

[0018] Among them, is the number of regions divided in the longitude direction, is the number of regions divided in the latitude direction, is the total number of regions, is the minimum longitude of the track to be fused, is the maximum longitude of the track to be fused, is the minimum latitude of the track to be fused, is the maximum latitude of the track to be fused; the symbol represents rounding down, such as .

[0019] The track is the spatio-temporal information of the target, including the original target number , position time , WGS-84 coordinate longitude , latitude , altitude , positioning error , three-dimensional geocentric coordinates , , Among them, is in seconds, is in degrees, is in degrees, is in meters, is in meters, , , is in meters. The track association calculation process can be referred to Figure 2 .

[0020] Refer to Figure 2 . The track association calculation process of the track association node is: The track association node sets the input track association threshold , time threshold , track : , , , , , , , , , save the trace to the original linked list according to the trace target number, and the spatio-temporal adaptive fusion model traverses all the original linked lists to find the time threshold in the original linked list the shortest distance within , the spatio-temporal adaptive fusion model uses the following calculation formula to calculate the trace association membership degree : , judge whether the association is successful according to the association membership degree When is greater than , the association is successful, and the association result is output; otherwise, traverse other original targets for trace association. Among them, 、 are input empirical constants, is the target number, is the position time, is the longitude, is the latitude, is the altitude, is the trace 's positioning error, , , is the trace 's three-dimensional geocentric coordinates.

[0021] Find the shortest distance within the time threshold in the original linked list of the trace association node within In: Find all traces in the original linked list whose position time difference from the trace is to form a set of all traces , is the trace in the set , and then, according to the three-dimensional geocentric coordinates , , …, , calculate the Euclidean distance between the trace and each trace in the set to obtain as the minimum value in : , where represents the minimum value in the set, represents the Euclidean distance between the trace and the th trace in the set

[0022] ,

[0023] Among them, , , is the three - dimensional geocentric coordinate of the trace, , , is the set in the th trace of the three - dimensional geocentric coordinate,

[0024] The track association node conducts track association calculation according to the track association rules. According to the current trace target number , the associated target number , and the association membership degree generate the track association result, and output the track association result to the trace fusion node according to the fusion distribution rules. The track association calculation process can be referred to Figure 3 .

[0025] Refer to Figure 3 . The track association calculation process of the track association node is as follows: input the trace association result, save the trace association result to the association linked list according to the trace association target number, and save the association membership degrees of the same target number and to the same association linked list in sequence. According to the set of all trace association results within the time threshold , calculate the track association membership degree : , generate the fusion number and allocate the fusion node, where, is the trace association result, , , is the total number of traces.

[0026] In generating the fusion number and allocating the fusion node, the track association node: judges whether the association is successful according to the track association membership degree . If is greater than the track association threshold, the association is successful; otherwise, the association fails. If the association is successful, the fusion number is the associated target number , and the fusion node remains unchanged. If the association fails, a new fusion target is generated. The target number of the new fusion target is recorded as , and the allocated fusion node is , where, is the current number of fusion targets, is the number of fusion nodes, represents taking the remainder, that is, represents divided by remainder.

[0027] The track fusion node performs track fusion calculations according to the track fusion rules at track fusion node 1, track fusion node 2... track fusion node to generate the final fusion result. The process of track fusion calculation can be referred to Figure 4 .

[0028] Refer to Figure 4 . The track fusion calculation process of the track fusion node is as follows: input track ( , , , , , , , , )and the fusion target number of this track. According to the fusion target number , save the track to the fusion linked list, find all tracks within the time threshold and calculate the fusion result according to the covariance convex combination. Among them, is the target number of the track, is the position time, is the longitude, is the latitude, is the altitude, is the positioning error, , , is the three-dimensional geocentric coordinate of the target.

[0029] The track fusion node finds all tracks within the time threshold : In the fusion linked list with the same fusion target number as the track , find the set of all tracks with a position time difference from the track , and find the tracks that meet the requirements: , is the track in the set ; identify the common information of the tracks participating in the fusion, set the fused target as the fusion point, and set , , as the three-dimensional geocentric coordinate of the fusion point, form the sub-track segment set of this track using the generated set of feature points, cluster the track data, and perform standard preprocessing on the time series. Generate the sub-track segment set of all tracks according to the set of feature points of all tracks, extract the track feature points, add the track feature points to the set of feature points, and calculate the fusion result using the covariance convex combination calculation formula shown below: The fusion result includes the positioning error of the fusion point and the three-dimensional geocentric coordinates of the fusion point , , :[[]]END]]

[0030]

[0031] Predict the associated matching trajectory, and the isolated points that are not matched are noise point traces; reorganize the trajectory sequence in the order of the trajectories, establish a circular association gate with a size larger than the preset threshold. If the estimated errors of the trajectories to be fused are uncorrelated, then fuse with other estimates, and expand the gate to recapture the lost target, remove duplicates and output the trajectory sequence after matching and merging. Among them, , , is the set in the th point trace of the three-dimensional geocentric coordinates, is the set in the th point trace of the square of the positioning error.

Claims

1. A multi-node parallel fusion method for spatio-temporal trajectories, characterized in that The method includes the following steps: searching for pixels with spectral similarity to the central pixel in the moving window using high-resolution data, filtering the samples and assigning nodes to the tracks, assigning combined weights to each similar pixel, and determining the conversion coefficients through regression analysis; based on the data transformation model, screening and processing the information of the transformed data, constructing a spatio-temporal trajectory dataset, a set of feature points, and a set of sub-trajectories, determining the weights according to the weight function, and establishing a spatio-temporal adaptive fusion model; the track assignment node calculates the association area of the tracks according to the track association assignment rules and outputs the tracks to the track association node; the track association node performs track association calculations according to the fuzzy double-threshold track association rules at track association node 1, track association node 2... track association node performs track association calculations, and obtains the track association results for each frame according to the correlation and outputs them to the track association node; The track association node performs track association calculations according to the track association rules and based on the current plot target number , the target number of the successfully associated target , the plot association membership degree Generate the track association result, and output the track association result to the plot fusion result according to the fusion distribution rules; the plot fusion result performs plot fusion calculations at the plot fusion node 1, plot fusion node 2... plot fusion node to generate the final fusion result.

2. The multi-node parallel fusion method for spatio-temporal trajectories according to claim 1, wherein: The track assignment node calculates the correlation area of the track according to the track association assignment rule : Wherein, is the number of regions divided in the longitude direction, is the number of regions divided in the latitude direction, is the total number of regions, is the minimum longitude of the tracks to be fused, is the maximum longitude of the tracks to be fused, is the minimum latitude of the tracks to be fused, is the maximum latitude of the tracks to be fused; the symbol represents rounding down.

3. The multi-node parallel fusion method for spatio-temporal trajectories according to claim 1, wherein: Tracks are the spatio-temporal information of targets, including the original target numbers , position time , WGS-84 coordinate longitude , latitude , altitude , positioning error , three-dimensional geocentric coordinates , , , where is in seconds, is in degrees, is in degrees, is in meters, is in meters, , , is in meters.

4. The multi-node parallel fusion method for spatio-temporal trajectories according to claim 1, wherein: The point track association calculation process of the point track association node is as follows: The point track association node sets the input point track association threshold and the time threshold and the point track : , , , , , , , , , According to the point track target number, the point track is saved to the original linked list. The spatio-temporal adaptive fusion model traverses all the original linked lists and searches for the shortest distance within the time threshold in the original linked list. The spatio-temporal adaptive fusion model uses the following calculation formula to calculate the point track association membership degree : : , According to the association membership degree judge whether the association is successful. When is greater than , the association is successful and the association result is output. Otherwise, it turns to traverse other original targets for point track association. Among them, and are the input empirical constants, is the target number, is the position time, is the longitude, is the latitude, is the altitude, is the point track 's positioning error, , , is the point track 's three-dimensional geocentric coordinate.

5. The multi-node parallel fusion method for spatio-temporal trajectories according to claim 1, characterized in that: The time threshold for searching in the original linked list of the track association node The shortest distance within Among them: Find all tracks in the original linked list that have a Position time difference with the track To form a set of all tracks , Denoted as the set For the tracks in the set, then, according to the three-dimensional geocentric coordinates , , …, , calculate the Euclidean distance between the track And each track in the set To obtain As The minimum value in : , where Represents the minimum value in the set, Represents the track And the set The th track In the Euclidean distance: , ; where , , Is the three-dimensional geocentric coordinate of the track , , , Is the set The th track In the three-dimensional geocentric coordinate, Is the total number of tracks.

6. The multi-node parallel fusion method for spatio-temporal trajectories according to claim 1, characterized in that: The track association calculation process of the track association node is as follows: Input the point track association result, save the point track association result to the association linked list according to the point track association target number, and the same target number and associated membership degree are sequentially saved to the same association linked list. According to the time threshold the set of all point track association results within , calculate the track association membership degree : , generate a fusion number and allocate a fusion node, where is the point track association result, , is the total number of point tracks.

7. The multi-node parallel fusion method for spatio-temporal trajectories according to claim 1, wherein: During the track association node generating the fusion number and allocating the fusion node: According to the track association membership degree judge whether the association is successful. If it is greater than the track association threshold, the association is successful; otherwise, the association fails. If the association is successful, the fusion number is the target number of the associated target , and the fusion node remains unchanged; If the association fails, a new fusion target is generated, and the target number of the new fusion target is recorded as , and the allocated fusion node is , where is the current number of fusion targets, is the number of fusion nodes, represents taking the remainder, that is represents divided by of the remainder.

8. The multi-node parallel fusion method for spatio-temporal trajectories according to claim 1, characterized in that: Track Fusion Node Search Time Threshold All tracks within: In the fusion linked list with the same fusion target number as track , search for the set of all tracks with a position time difference from track , and find the tracks that meet the requirements: , , is the track in the set ; identify the common information of the tracks participating in the fusion, set the fused target as the fusion point, and set , , as the three-dimensional geocentric coordinates of the fusion point, form a set of sub-track segments of this track using the generated set of feature points, cluster the track data, and perform standardized preprocessing on the time series. Generate a set of sub-track segments for all tracks based on the set of feature points of all tracks, extract the track feature points, add the track feature points to the set of feature points, and calculate the fusion result using the covariance convex combination calculation formula shown below: The fusion result includes the positioning error of the fusion point and the three-dimensional geocentric coordinates of the fusion point, , : Predict the associated matching trajectory, and the isolated points that are not matched are noise point traces; among them, , , is the set the th point trace in the three-dimensional geocentric coordinates, is the set the th point trace in the squared positioning error.

9. The multi-node parallel fusion method for spatio-temporal trajectories according to claim 8, wherein: Re - form the trajectory sequence according to the order of the trajectories, establish an annular association gate with a size larger than a preset threshold. If the estimated errors of the trajectories to be fused are uncorrelated, then fuse them with other estimations, and expand the gate to re - capture the lost targets, remove duplicates and output the trajectory sequence after matching and merging.

Images