Semantic recognition method for movement behavior of marine moving target

Abstract

The invention discloses a seaborne moving target motion behavior semantic recognition method, and relates to the field of data analysis services. According to the method, firstly, de-noising processing is carried out on a sea moving target trajectory Tra, 'start 'and'end' semantic trajectory points in the sea moving target trajectory Tra are identified, then'stay 'semantics are identified by utilizing a fixed time window, and then improvement is carried out on the basis of a traditional fixed window based on the motion space-time characteristics of a target. According to the method, the detour semantics of the moving target on the sea are recognized through a sliding window determined in a self-adaptive mode, further refined recognition and extraction are carried out on the turn-back semantics and the surrounding semantics through the self-adaptive sliding window on the basis of detour semantics recognition, and other track points are recognized as normal navigation semantics. According to the method, the bottleneck problems of high algorithm complexity, incomplete recognition, low recognition precision and the like of a traditional sea moving target behavior semantic recognition method can be effectively solved.

Description

Technical Field

[0001] The invention relates to the field of data analysis services, and in particular to a method for semantic recognition of motion behaviors of mobile targets at sea. Background Art

[0002] Accurate recognition of the semantics of the motion behavior of mobile targets at sea is the prerequisite for mining motion behavior patterns and predicting future trends. In order to more accurately and efficiently identify the semantics of the motion behavior of mobile targets at sea from their tracks, scholars at home and abroad have proposed a variety of technical methods and means. Recognition based on sliding windows is one of the most classic methods. It has the advantages of low complexity and strong scalability. It is also a common means of engineering application of semantic recognition of the motion behavior of mobile targets at sea. Traditional sliding window algorithms generally use fixed time windows or track point windows. Due to the inconsistent track sampling intervals, if traditional fixed-size windows are used to extract semantics, it is easy to cause semantic loss in dense areas of track points and inaccurate semantic extraction in sparse areas. Therefore, it is necessary to improve and enhance the adaptive determination of sliding windows and the corresponding recognition processes and methods based on existing research to achieve more accurate recognition of the semantics of the motion behavior of mobile targets at sea. Summary of the invention

[0003] In view of this, the present invention proposes a method for semantic recognition of motion behavior of mobile targets at sea. The method determines the recognition window by the motion attributes of the target itself, and recognizes and extracts the semantics in batches according to the behavior attributes, thereby improving the accuracy, scalability and effectiveness of semantic recognition of motion behavior of mobile targets at sea.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] A method for semantic recognition of motion behavior of a mobile target at sea, comprising the following steps:

[0006] Step 1, denoising the trajectory Tra of the marine moving target;

[0007] Step 2: for the denoised marine moving target trajectory Tra, its first trajectory point is identified as the “start” semantics, and its last trajectory point is identified as the “end” semantics;

[0008] Step 3: traverse the trajectory points whose semantics have not been recognized according to the fixed time window, and identify the trajectory points with "stay" semantics;

[0009] Step 4: For each trajectory point whose semantics has not been recognized, set its corresponding sliding time window;

[0010] Step 5: for each trajectory point whose semantics have not been identified, identify the trajectory point with the semantics of "detour" in its corresponding sliding time window;

[0011] Step 6, for each trajectory point with the semantics of "detour", the trajectory point with the semantics of "turnaround" is identified, and its "detour" semantics is re-identified as "turnaround" semantics; the trajectory point with the semantics of "surrounding" is identified, and its "detour" semantics is re-identified as "surrounding" semantics;

[0012] Step 7: Identify all trajectory points that have not yet been semantically recognized as "normal navigation" semantics, output the target trajectory with motion behavior semantic tags, and complete the motion behavior semantic recognition of the mobile target at sea.

[0013] Furthermore, the specific method of step 3 is:

[0014] Step 301: Initialize the navigation status values ​​of all trajectory points that have not yet recognized semantics, set the stop speed threshold V1, stop time threshold T1, stop distance threshold D1 and fixed time window T 固 ;

[0015] Step 302: according to the fixed time window T 固 Traverse the trajectory points whose semantics have not been recognized yet, record the navigation status value of the trajectory points whose speed is greater than V1 as 1, and record the navigation status values ​​of the remaining trajectory points as 0;

[0016] Step 303, for a trajectory segment consisting of trajectory points whose navigation status values ​​are continuously 0, if the duration of the trajectory segment is greater than T1, and the distances between any two adjacent trajectory points in the trajectory segment are less than D1, all trajectory points in the trajectory segment are identified as having "stay" semantics, otherwise no identification processing is performed.

[0017] Furthermore, the specific method of step 4 is:

[0018] Step 401: for each track point whose semantics have not been recognized, calculate the time interval between two adjacent track points. If there is a track point with the semantics of "stay" between two adjacent track points, the time interval between the two adjacent track points is recorded as 0. Count all the time intervals that are not 0, and record their median as T 中 ;

[0019] Step 402: traverse each trajectory point that has not yet been semantically recognized, and for the currently traversed trajectory point and the five trajectory points before and after it that have not yet been semantically recognized, calculate the average time interval between any two adjacent trajectory points, and record it as T 均, if there are less than 5 unrecognized trajectory points before the currently traversed trajectory point, then all the unrecognized trajectory points before the currently traversed trajectory point and the 5 unrecognized trajectory points after the currently traversed trajectory point are selected to participate in the calculation; if there are less than 5 unrecognized trajectory points after the currently traversed trajectory point, then the 5 unrecognized trajectory points before the currently traversed trajectory point and all the unrecognized trajectory points after the currently traversed trajectory point are selected to participate in the calculation;

[0020] Step 403, if T 均 <T 中 , then the sliding time window corresponding to the currently traversed trajectory point is defined as: 均 All trajectory points calculated; if T 均 ≥T 中 , then the sliding time window corresponding to the currently traversed trajectory point is defined as: the currently traversed trajectory point and the 5*T before and after 中 track points, if the current track point is less than 5*T 中 If there are no semantically recognized trajectory points, the currently traversed trajectory point, all previous trajectory points that have not yet been semantically recognized, and the next 5*T trajectory points are selected. 中 There are no trajectory points whose semantics have not been recognized yet; if the current trajectory point is less than 5*T 中 If there are no semantically recognized trajectory points, the currently traversed trajectory points and the previous 5*T 中 The first track point that has not yet been semantically recognized and all subsequent track points that have not yet been semantically recognized.

[0021] Furthermore, the specific method of step 5 is:

[0022] Step 501, initializing the steering state values ​​of all trajectory points whose semantics have not been recognized, setting the detour cumulative deviation angle threshold V2 and the detour continuous trajectory point number threshold N1;

[0023] Step 502, traverse each trajectory point that has not yet recognized semantics, and for the currently traversed trajectory point, calculate the cumulative deviation angle of the first and last trajectory points in its corresponding sliding time window. If the cumulative deviation angle is greater than V2, set the turning state value of all trajectory points in the current sliding time window to 1, otherwise set the turning state value of all trajectory points in the current sliding time window to 0;

[0024] Step 503, for a trajectory segment consisting of trajectory points whose turning state values ​​are continuously 1, if the number of trajectory points contained in the trajectory segment is greater than N1, all trajectory points in the trajectory segment are identified as "detour" semantics, otherwise no identification processing is performed.

[0025] Furthermore, in step 6, the track points with the "return" semantics are identified, and the specific method of re-identifying the "detour" semantics as the "return" semantics is as follows:

[0026] Traverse each trajectory point with "detour" semantics, and calculate the cumulative deviation angle of the first and last two trajectory points with "detour" semantics in the corresponding sliding time window, and record it as θ 总 , if 170°≤θ 总 If the angle is less than or equal to 190°, all trajectory points with “detour” semantics in the current sliding time window are re-identified as “return” semantics, otherwise no re-identification is performed.

[0027] Furthermore, in step 6, the trajectory points having the "surround" semantics are identified, and the specific method of re-identifying the "detour" semantics as the "surround" semantics is as follows:

[0028] Set the surround spatial distance deviation threshold D2, traverse each trajectory point with "bypass" semantics, and in the corresponding sliding time window, take every three adjacent trajectory points with "bypass" semantics as a group of calculation units, and calculate the center coordinates of the circle formed by the three trajectory points in each group of calculation units. In the current sliding time window, if the spatial distance difference of the center coordinates corresponding to each two adjacent calculation units is less than D2, and the cumulative deviation angle of the first and last two trajectory points with "bypass" semantics is greater than 360°, then all trajectory points with "bypass" semantics in the current sliding time window will be re-identified as "surrounding" semantics, otherwise no re-identification will be performed.

[0029] Due to the adoption of the above technical solution, the beneficial effects of the present invention compared with the prior art are:

[0030] 1. The present invention proposes a method for semantic recognition of motion behavior of mobile targets at sea, which is mainly aimed at improving the self-adaptation of sliding windows. The core idea is to achieve window adaptability by comparing the average time interval of several adjacent trajectory points of the current detection point with the median of all time intervals of the current entire trajectory. It can effectively solve the problem that using traditional fixed-size windows to extract semantics easily causes semantic loss in dense areas of trajectory points and inaccurate semantic extraction in sparse areas.

[0031] 2. The present invention proposes a method for identifying the semantics of motion behavior of mobile targets at sea, which identifies and extracts seven typical motion behavior semantics, namely, "start", "end", "stay", "normal navigation", "detour", "turnback", and "circle", and basically covers the basic behavior semantics of mobile targets at sea. The semantics are distinguished and identified according to their behavior attributes and related definitions, which not only ensures the completeness and accuracy of the behavior semantic extraction, but also reduces the amount of calculation and the problem of repeated traversal identification to a certain extent, further improving the efficiency of the semantic recognition algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The figure is a schematic diagram of the overall process of a method for semantic recognition of motion behavior of a mobile target at sea in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0034] A semantic recognition method for the motion behavior of a mobile target at sea, such as Figure 1 As shown, the following steps are included;

[0035] Step 1, denoising the trajectory Tra of the marine moving target;

[0036] Step 2: for the denoised marine moving target trajectory Tra, its first trajectory point is identified as the “start” semantics, and its last trajectory point is identified as the “end” semantics;

[0037] Step 3: traverse the trajectory points whose semantics have not been recognized according to the fixed time window, and identify the trajectory points with "stay" semantics;

[0038] Step 4: For each trajectory point whose semantics has not been recognized, set its corresponding sliding time window;

[0039] Step 5: for each trajectory point whose semantics have not been identified, identify the trajectory point with the semantics of "detour" in its corresponding sliding time window;

[0040] Step 6, for each trajectory point with the semantics of "detour", the trajectory point with the semantics of "turnaround" is identified, and its "detour" semantics is re-identified as "turnaround" semantics; the trajectory point with the semantics of "surrounding" is identified, and its "detour" semantics is re-identified as "surrounding" semantics;

[0041] Step 7: Identify all trajectory points that have not yet been semantically recognized as "normal navigation" semantics, output the target trajectory with motion behavior semantic tags, and complete the motion behavior semantic recognition of the mobile target at sea.

[0042] Furthermore, the specific method of step 3 is:

[0043] Step 301: Initialize the navigation status values ​​of all trajectory points that have not yet recognized semantics, set the stop speed threshold V1, stop time threshold T1, stop distance threshold D1 and fixed time window T 固 ;

[0044] Step 302: according to the fixed time window T 固Traverse the trajectory points whose semantics have not been recognized yet, record the navigation status value of the trajectory points whose speed is greater than C1 as 1, and record the navigation status values ​​of the remaining trajectory points as 0;

[0045] Step 303, for a trajectory segment consisting of trajectory points whose navigation status values ​​are continuously 0, if the duration of the trajectory segment is greater than T1, and the distances between any two adjacent trajectory points in the trajectory segment are less than D1, all trajectory points in the trajectory segment are identified as having "stay" semantics, otherwise no identification processing is performed.

[0046] Furthermore, the specific method of step 4 is:

[0047] Step 401: for each track point whose semantics have not been recognized, calculate the time interval between two adjacent track points. If there is a track point with the semantics of "stay" between two adjacent track points, the time interval between the two adjacent track points is recorded as 0. Count all the time intervals that are not 0, and record their median as T 中 ;

[0048] Step 402: traverse each trajectory point that has not yet been semantically recognized, and for the currently traversed trajectory point and the five trajectory points before and after it that have not yet been semantically recognized, calculate the average time interval between any two adjacent trajectory points, and record it as T 均 , if there are less than 5 unrecognized trajectory points before the currently traversed trajectory point, then all the unrecognized trajectory points before the currently traversed trajectory point and the 5 unrecognized trajectory points after the currently traversed trajectory point are selected to participate in the calculation; if there are less than 5 unrecognized trajectory points after the currently traversed trajectory point, then the 5 unrecognized trajectory points before the currently traversed trajectory point and all the unrecognized trajectory points after the currently traversed trajectory point are selected to participate in the calculation;

[0049] Step 403, if T 均 <T 中 , then the sliding time window corresponding to the currently traversed trajectory point is defined as: 均 All trajectory points calculated; if T 均 ≥T 中 , then the sliding time window corresponding to the currently traversed trajectory point is defined as: the currently traversed trajectory point and the 5*T before and after 中 track points, if the current track point is less than 5*T 中 If there are no semantically recognized trajectory points, the currently traversed trajectory point, all previous trajectory points that have not yet been semantically recognized, and the next 5*T trajectory points are selected. 中 There are no trajectory points whose semantics have not been recognized yet; if the current trajectory point is less than 5*T 中 If there are no semantically recognized trajectory points, the currently traversed trajectory points and the previous 5*T中 The first track point that has not yet been semantically recognized and all subsequent track points that have not yet been semantically recognized.

[0050] Furthermore, the specific method of step 5 is:

[0051] Step 501, initializing the steering state values ​​of all trajectory points whose semantics have not been recognized, setting the detour cumulative deviation angle threshold V2 and the detour continuous trajectory point number threshold N1;

[0052] Step 502, traverse each trajectory point that has not yet recognized semantics, and for the currently traversed trajectory point, calculate the cumulative deviation angle of the first and last trajectory points in its corresponding sliding time window. If the cumulative deviation angle is greater than V2, set the turning state value of all trajectory points in the current sliding time window to 1, otherwise set the turning state value of all trajectory points in the current sliding time window to 0;

[0053] Step 503, for a trajectory segment consisting of trajectory points whose turning state values ​​are continuously 1, if the number of trajectory points contained in the trajectory segment is greater than N1, all trajectory points in the trajectory segment are identified as "detour" semantics, otherwise no identification processing is performed.

[0054] Furthermore, in step 6, the track points with the "return" semantics are identified, and the specific method of re-identifying the "detour" semantics as the "return" semantics is as follows:

[0055] Traverse each trajectory point with "detour" semantics, and calculate the cumulative deviation angle of the first and last two trajectory points with "detour" semantics in the corresponding sliding time window, and record it as θ 总 , if 170°≤θ 总 If the angle is less than or equal to 190°, all trajectory points with “detour” semantics in the current sliding time window are re-identified as “return” semantics, otherwise no re-identification is performed.

[0056] Furthermore, in step 6, the trajectory points having the "surrounding" semantics are identified, and the specific method of re-identifying the "detour" semantics as the "surrounding" semantics is as follows:

[0057] Set the surround spatial distance deviation threshold D2, traverse each trajectory point with "bypass" semantics, and in the corresponding sliding time window, take every three adjacent trajectory points with "bypass" semantics as a group of calculation units, and calculate the center coordinates of the circle formed by the three trajectory points in each group of calculation units. In the current sliding time window, if the spatial distance difference of the center coordinates corresponding to each two adjacent calculation units is less than D2, and the cumulative deviation angle of the first and last two trajectory points with "bypass" semantics is greater than 360°, then all trajectory points with "bypass" semantics in the current sliding time window will be re-identified as "surrounding" semantics, otherwise no re-identification will be performed.

[0058] In summary, the present invention combines a fixed time window with an adaptively determined sliding window to identify and extract seven typical motion behavior semantics of mobile targets at sea, namely, "start", "end", "stay", "normal navigation", "detour", "turnaround", and "surrounding". First, the "start" and "end" semantic trajectory points are identified, and then the "stay" semantics are identified using a fixed time window. Then, based on the traditional fixed window, improvements are made based on the target's own motion spatiotemporal characteristics, and the "detour" semantics of the mobile target at sea is identified through an adaptively determined sliding window. On the basis of the "detour" semantic recognition, the "turnaround" and "surrounding" semantics are further refined and extracted through an adaptive sliding window, and the remaining trajectory points are identified as "normal navigation" semantics. The present invention can effectively overcome the bottleneck problems of high algorithm complexity, incomplete recognition, and low recognition accuracy in traditional methods for semantic recognition of mobile target behaviors at sea.

[0059] Those skilled in the art will appreciate that the embodiments described are intended to help readers understand the principles of the present invention, and should be understood that the scope of protection of the present invention is not limited to the embodiments described. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for semantic recognition of motion behavior of a mobile target at sea, characterized in that: The steps include: Step 1, denoising the trajectory Tra of the marine moving target; Step 2: for the denoised marine moving target trajectory Tra, its first trajectory point is identified as the "start" semantics, and its last trajectory point is identified as the "end" semantics; Step 3: traverse the trajectory points whose semantics have not been recognized according to the fixed time window, and identify the trajectory points with "stay" semantics; Step 4: For each trajectory point whose semantics has not been recognized, set its corresponding sliding time window; Step 5: for each trajectory point whose semantics has not been identified, identify the trajectory point with the semantics of "detour" in its corresponding sliding time window; Step 6, for each trajectory point with "detour" semantics, identify the trajectory point with "turnaround" semantics, and re-identify its "detour" semantics as "turnaround" semantics; identify the trajectory point with "surrounding" semantics, and re-identify its "detour" semantics as "surrounding" semantics; Step 7: Identify all the trajectory points that have not yet been semantically recognized as "normal navigation" semantics, output the target trajectory with motion behavior semantic tags, and complete the motion behavior semantic recognition of the mobile target at sea.

2. A method for semantic recognition of motion behavior of a mobile target at sea according to claim 1, characterized in that: The specific method of step 3 is: Step 301: Initialize the navigation status values ​​of all trajectory points that have not yet recognized semantics, set the stop speed threshold V1, stop time threshold T1, stop distance threshold D1 and fixed time window T 固 ; Step 302: according to the fixed time window T 固 Traverse the trajectory points whose semantics have not been recognized yet, record the navigation status value of the trajectory points whose speed is greater than V1 as 1, and record the navigation status values ​​of the remaining trajectory points as 0; Step 303, for a trajectory segment consisting of trajectory points whose navigation status values ​​are continuously 0, if the duration of the trajectory segment is greater than T1, and the distances between any two adjacent trajectory points in the trajectory segment are less than D1, all trajectory points in the trajectory segment are identified as having "stay" semantics, otherwise no identification processing is performed.

3. A method for semantic recognition of motion behavior of a mobile target at sea according to claim 2, characterized in that: The specific method of step 4 is: Step 401, for each trajectory point whose semantics have not been recognized, calculate the time interval between two adjacent trajectory points. If there is a trajectory point with the semantics of "stay" between two adjacent trajectory points, the time interval between the current two adjacent trajectory points is recorded as 0. Count all time intervals that are not 0, and record their median as T 中 ; Step 402: traverse each trajectory point that has not yet been semantically recognized, and for the currently traversed trajectory point and the five trajectory points before and after it that have not yet been semantically recognized, calculate the average time interval between any two adjacent trajectory points, and record it as T 均 , if there are less than 5 unrecognized trajectory points before the currently traversed trajectory point, then all the unrecognized trajectory points before the currently traversed trajectory point and the 5 unrecognized trajectory points after the currently traversed trajectory point are selected to participate in the calculation; if there are less than 5 unrecognized trajectory points after the currently traversed trajectory point, then the 5 unrecognized trajectory points before the currently traversed trajectory point and all the unrecognized trajectory points after the currently traversed trajectory point are selected to participate in the calculation; Step 403, if T 均 <T 中 , then the sliding time window corresponding to the currently traversed trajectory point is defined as: 均 All trajectory points calculated; if T 均 ≥T 中 , then the sliding time window corresponding to the currently traversed trajectory point is defined as: the currently traversed trajectory point and the 5*T before and after 中 track points, if the current track point is less than 5*T 中 If there are no semantically recognized trajectory points, the currently traversed trajectory point, all previous trajectory points that have not yet been semantically recognized, and the next 5*T trajectory points are selected. 中 There are no trajectory points whose semantics have not been recognized yet; if the current trajectory point is less than 5*T 中 If there are no semantically recognized trajectory points, the currently traversed trajectory points and the previous 5*T 中 The first track point that has not yet been semantically recognized and all subsequent track points that have not yet been semantically recognized.

4. A method for semantic recognition of motion behavior of a mobile target at sea according to claim 3, characterized in that: The specific method of step 5 is: Step 501, initializing the turning state values ​​of all trajectory points whose semantics have not been recognized, setting the detour cumulative deviation angle threshold V2 and the detour continuous trajectory point number threshold N1; Step 502, traverse each trajectory point that has not yet recognized semantics, and for the currently traversed trajectory point, calculate the cumulative deviation angle of the first and last trajectory points in its corresponding sliding time window. If the cumulative deviation angle is greater than V2, set the turning state value of all trajectory points in the current sliding time window to 1, otherwise set the turning state value of all trajectory points in the current sliding time window to 0; Step 503, for a trajectory segment consisting of trajectory points whose turning state values ​​are continuously 1, if the number of trajectory points contained in the trajectory segment is greater than N1, all trajectory points in the trajectory segment are identified as "detour" semantics, otherwise no identification processing is performed.

5. A method for semantic recognition of motion behavior of a mobile target at sea according to claim 4, characterized in that: In step 6, the track points with the "return" semantics are identified, and the specific method of re-identifying the "detour" semantics as the "return" semantics is as follows: Traverse each trajectory point with "detour" semantics, and calculate the cumulative deviation angle of the first and last two trajectory points with "detour" semantics in the corresponding sliding time window, and record it as θ 总 , if 170°≤θ 总 ≤190°, all trajectory points with "detour" semantics in the current sliding time window are re-identified as "return" semantics, otherwise no re-identification processing is performed.

6. A method for semantic recognition of motion behavior of a mobile target at sea according to claim 4, characterized in that: In step 6, the trajectory points with "surround" semantics are identified, and the specific method of re-identifying the "detour" semantics as "surround" semantics is as follows: Set the surround space distance deviation threshold D2, traverse each trajectory point with "detour" semantics, and in the corresponding sliding time window, take every three adjacent trajectory points with "detour" semantics as a group of calculation units, and calculate the center coordinates of the circle formed by the three trajectory points in each group of calculation units. In the current sliding time window, if the spatial distance difference of the center coordinates corresponding to each two adjacent calculation units is less than D2, and the cumulative deviation angle of the first and last two trajectory points with "detour" semantics is greater than 360°, then all trajectory points with "detour" semantics in the current sliding time window will be re-identified as "surrounding" semantics, otherwise no re-identification will be performed.

Images