Multivariate object intention recognition method and system based on space-time co-occurrence mining

Abstract

The invention provides a multivariate object intention recognition method based on space-time co-occurrence mining, and relates to the field of data information mining. According to the method, an intermediate result is constructed into a multi-level index through an object track area index construction link, and various data such as tracks, areas and co-occurrence events can be efficiently mined and recognized; in a space-time co-occurrence object extraction link, specific models of aircraft and ship co-occurrence objects are utilized to replace individuals which are difficult to identify in practical application, and an individual co-occurrence problem is converted into co-occurrence of object types; after sufficient type co-occurrence data is accumulated through a multivariate object intention recognition link, subsequent behavior prediction can be carried out according to the combination of similar historical data of the type and the type of a task executed after manual confirmation, and the more the accumulated data is, the more the behavior preference of each co-occurrence combination can be represented. The intention and behavior recognition effect is more and more accurate, and the user is assisted to carry out more comprehensive recognition and judgment.

Description

Technical Field

[0001] The present invention relates to the field of data information mining, and in particular to a multi-object intention recognition method based on spatiotemporal co-occurrence mining. Background Art

[0002] Spatiotemporal co-occurrence mining is a technology used to analyze objects or events that co-occur within a specific time and space range. By integrating data in the time and space dimensions, it can more comprehensively reveal the correlation between objects or events, thus playing an important role in many fields, such as social network analysis, mobile data analysis, and urban planning.

[0003] With the advancement of technology and the expansion of application scenarios, a variety of algorithms have emerged to process spatiotemporal data. For example, graph-based methods, clustering-based methods, and machine learning methods (such as deep learning) are widely used to identify co-occurrence patterns. The continuous development of advanced sensors, remote sensing technology, and automated data collection systems has greatly promoted the collection and accumulation of spatiotemporal data, making spatiotemporal data increasingly rich and comprehensive.

[0004] The advancement of these technologies not only improves the speed and accuracy of data collection, but also expands the scope of data coverage, providing a more solid data foundation for research and application in the field of space and time. Space and time data characterize the activity characteristics of objects in the objective world, and space and time co-occurrence reflects the activity patterns and behavioral intentions of objects. Using big data and artificial intelligence technology, how to timely and accurately mine the behavior prediction and intention analysis of space and time co-occurrence objects is of great significance.

[0005] At present, most of the research results related to spatiotemporal co-occurrence patterns are obtained by expanding the time dimension based on spatial co-occurrence patterns, and the methods used are mainly based on the Apriori algorithm. For example, Celik defined a hybrid spatiotemporal co-occurrence pattern measurement method and proposed a hybrid spatiotemporal co-occurrence pattern algorithm, analyzed the correctness and completeness of the algorithm, and developed a spatiotemporal partial co-occurrence pattern mining algorithm for discovering fixed patterns and common task actions in animal migration.

[0006] James W, Muraki T, Donna M and others studied the problem of mining frequent co-occurrence subsequences between different mobile objects from spatiotemporal data and proposed a two-stage mining algorithm. The algorithm first uses a hash function to convert the original trajectory into a subsequence with similar features, and then uses the Apriori algorithm to mine frequent segments. Liu B and Matwin S combined the measurement problems of the time dimension and the space dimension and proposed a spatiotemporal co-occurrence pattern discovery algorithm COSTCOP.

[0007] The algorithm comprehensively considers the time and space dimensions in the process of pattern mining. However, the massive location data of aircraft and ships has put forward new requirements for the mining of spatiotemporal co-occurrence patterns. At present, the amount of information has reached the TB level. The existing research results are mostly based on a small number of data samples, which is difficult to cope with such a huge amount of data. In addition, the serial calculation method of traditional mining algorithms and their insensitivity to spatial features limit the speed and efficiency of co-occurrence pattern mining.

[0008] Therefore, it is necessary to provide a multi-object intention recognition method based on spatiotemporal co-occurrence mining to solve the technical problems that the serial calculation method of traditional mining algorithms and their insensitivity to spatial features limit the speed and efficiency of co-occurrence pattern mining. Summary of the invention

[0009] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an easy-to-operate method for multi-object intention recognition based on spatiotemporal co-occurrence mining, which solves the technical problems that the serial calculation method of traditional mining algorithms and their insensitivity to spatial features limit the speed and efficiency of co-occurrence pattern mining, and assists users in making more comprehensive identification and verification.

[0010] To achieve the above purpose, the present application proposes a multi-object intention recognition method based on spatiotemporal co-occurrence mining, including an object track area index construction link, a spatiotemporal co-occurrence object extraction link and a multi-object intention recognition link; wherein,

[0011] Object track area index construction link: by acquiring the trajectory data of the object in the track area, construct the corresponding object track area index data; wherein, the track area is provided with a number of sensitive areas, and the object track area index data is used to describe the object's entry and exit of the sensitive area;

[0012] Spatiotemporal co-occurrence object extraction phase: the sensitive area originally delineated based on the distance threshold is expanded, and the result set is determined through the expanded sensitive area. The object track area index data is recalled through the result set, and the object track area index data is screened to obtain the spatiotemporal co-occurrence object set;

[0013] Multi-object intention recognition link: obtain each multi-model combination through the spatiotemporal co-occurrence object set, and determine the task or intention corresponding to the multi-model combination through prior knowledge; among them, if the current multi-model combination is an unknown combination, the corresponding support and confidence are output and the unknown combination is confirmed manually.

[0014] As a further solution, the object track area index construction step is performed through the following steps:

[0015] Step 1: pre-divide several sensitive areas in the track area;

[0016] Step 2: Obtain a result set of sensitive areas passed by the object track through a spatiotemporal database; wherein the spatiotemporal database records the trajectory data of each object;

[0017] Step 3: Obtain the entry and exit time of the object track entering and exiting each sensitive area through the result set;

[0018] Step 4: Output the required object track area index data.

[0019] As a further solution, in step 3, the entry and exit times are also rounded; the entry time for entering the sensitive area is rounded down, and the entry time for leaving the sensitive area is rounded up; if the same object continuously enters and exits the same area, the earliest entry time and the latest exit time are selected as the entry and exit times.

[0020] As a further solution, in step 4, the object track area index data includes object number, area number, object model, area time step and track date; wherein the object number is uniquely corresponding, the area time step is divided based on the time step of 0-23, and is determined by the entry and exit time.

[0021] As a further solution, the region set is constructed by the following steps:

[0022] Set spatiotemporal co-occurrence conditions, including distance threshold and time threshold;

[0023] Select the current sensitive area, and expand the current sensitive area outward according to the distance threshold to obtain the expanded sensitive area;

[0024] Determine an intersection sensitive area; wherein the intersection sensitive area is a sensitive area that intersects with the expansion sensitive area;

[0025] The construction of the region set of the current sensitive region is completed by dilating the sensitive region and taking the union of the intersecting sensitive regions as the region set.

[0026] As a further solution, the following steps are used to select objects that co-occur in time and space:

[0027] Get the area set and recall the object track area index data corresponding to the area set;

[0028] Add the track area index data of the recalled object into the same recall candidate set according to the same track date;

[0029] Select the recall candidate set, and select the time period by the preset time threshold, filter out the object track area index data in the same time period, and add it to the same spatiotemporal co-occurrence object set;

[0030] Acquire each multivariate model combination in the spatiotemporal co-occurrence object set; wherein the multivariate model combination is a combination of different object models in the selected object track area index data;

[0031] Determine the tasks or intentions that the multi-model combination corresponds to through prior knowledge;

[0032] If the acquisition is successful, the multivariate model combination is a known combination, and the corresponding task or intention to be performed is output;

[0033] If it cannot be determined, the multivariate model combination is an unknown combination, and the corresponding support and confidence are output.

[0034] As a further solution, when selecting a time period by a preset time threshold, if there are multiple object track area index data with the same object number in the same time period, they are regarded as one object track area index data with the same object number; wherein, the time period selection formula is as follows:

[0035] Time=[Tt,T+t]

[0036] Where T is the current time point, t is the time threshold, and Time is the selected time period.

[0037] As a further solution, the confidence level is the instance participation rate that conforms to the spatiotemporal co-occurrence pattern.

[0038] As a further solution, the support degree is the accumulated count value of the selected object track area index data; wherein the accumulated count ignores the object number and takes the object model as the statistical basis.

[0039] On the other hand, the present invention provides a multi-object intention recognition system based on spatiotemporal co-occurrence mining, which uses a multi-object intention recognition method based on spatiotemporal co-occurrence mining as described in any of the above items to perform object track area index construction, spatiotemporal co-occurrence object extraction and multi-object intention recognition.

[0040] Compared with related technologies, the multi-object intention recognition method based on spatiotemporal co-occurrence mining provided by the present invention has the following advantages:

[0041] The present invention constructs the intermediate results into a multi-level index through the object track area index construction link, which can efficiently mine and identify various data such as trajectories, areas, co-occurrence events, etc.; in the spatiotemporal co-occurrence object extraction link, the specific models of aircraft and ship co-occurrence objects are used to replace individuals that are difficult to identify in practical applications, and the problem of individual co-occurrence is converted into the co-occurrence of object types; after accumulating enough type co-occurrence data through the multi-object intention recognition link, subsequent behavior prediction can be performed based on the combination of similar historical data of its type and the type of task performed after manual confirmation, and the more accumulated data, the more it can characterize the behavioral preferences of each co-occurrence combination, and the intention and behavior recognition effects will become more and more accurate, assisting users in more comprehensive identification and verification. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 A schematic flow chart of a multi-object intention recognition method based on spatiotemporal co-occurrence mining provided by the present invention;

[0045] Figure 2 A schematic diagram of the track data provided by the present invention;

[0046] Figure 3 A schematic diagram of the division of sensitive areas provided by the present invention;

[0047] Figure 4 A schematic diagram of a combination of multiple models of sensitive areas provided by the present invention;

[0048] Figure 5 This is a schematic diagram of the statistical results provided by the present invention.

[0049] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0051] See also Figure 1 The embodiment of the present application provides a multi-object intention recognition method based on spatiotemporal co-occurrence mining, including an object track area index construction link, a spatiotemporal co-occurrence object extraction link and a multi-object intention recognition link; wherein,

[0052] Object track area index construction link: by acquiring the trajectory data of the object in the track area, construct the corresponding object track area index data; wherein, the track area is provided with a number of sensitive areas, and the object track area index data is used to describe the object's entry and exit of the sensitive area;

[0053] Spatiotemporal co-occurrence object extraction phase: the sensitive area originally delineated based on the distance threshold is expanded, and the result set is determined through the expanded sensitive area. The object track area index data is recalled through the result set, and the object track area index data is screened to obtain the spatiotemporal co-occurrence object set;

[0054] Multi-object intention recognition link: obtain each multi-model combination through the spatiotemporal co-occurrence object set, and determine the task or intention corresponding to the multi-model combination through prior knowledge; among them, if the current multi-model combination is an unknown combination, the corresponding support and confidence are output and the unknown combination is confirmed manually.

[0055] It should be noted that: this embodiment first obtains the trajectory data of the object in the track area to determine which preset areas the object has passed through, and obtains the entry and exit time of each area, so as to perform time step division based on 0-23, and then finds out the specific models of aircraft and ships that co-occur within the given time and space range by determining the time threshold and space threshold.

[0056] If there is prior knowledge as a basis for judging the combination of its type, the specific intention of its multi-object can be given; if the combination has no prior knowledge support, the candidate set that meets the support and confidence thresholds is manually judged, and the final result is added to the prior knowledge set for accumulation.

[0057] like Figure 2As shown in the figure, for the spatiotemporal co-occurrence mining of aircraft and ships, track data is the basis. Since the volume of historically accumulated track data is very large, it is necessary to efficiently extract historical co-occurrence objects from it as a basis for subsequent object recognition and intention judgment. In order to quickly and accurately mine high-value spatiotemporal co-occurrence objects in massive track data, a method for efficiently generating track time step area index information is proposed. The specific steps are as follows:

[0058] Step 1: pre-divide several sensitive areas in the track area;

[0059] Step 2: Obtain a result set of sensitive areas passed by the object track through a spatiotemporal database; wherein the spatiotemporal database records the trajectory data of each object;

[0060] Step 3: Obtain the entry and exit time of the object track entering and exiting each sensitive area through the result set;

[0061] Step 4: Output the required object track area index data.

[0062] Among them, several sensitive areas are pre-divided according to actual needs and experience, and the result set of the preset areas passed by the track is obtained through the space-time database; the time for the track to enter and exit each area needs to be rounded down for the entry time and up for the exit time; in addition, if there is continuous entry and exit in the same area, the earliest entry and latest exit time are selected.

[0063] The output required object track area index data includes object number, area number, object model, area time step and track date, wherein the object number is uniquely corresponding, the area time step is divided based on the time step of 0-23 and is determined by the entry and exit time. The object track area index is shown in Table 1:

[0064] Table 1 Object track area index table

[0065]

[0066] The method for producing track time step region index information provided in this embodiment has a very high efficiency, and can quickly process a large amount of data to be processed and generate track region indexes, thereby supporting the subsequent recognition of spatiotemporal co-occurring objects.

[0067] For the recognition of spatiotemporal co-occurrence objects, it is necessary to establish the co-occurrence conditions for sea and air objects, namely the distance threshold and the time threshold respectively; traverse the preset sensitive areas and expand the current area according to the distance threshold, and determine whether the expanded area contains other preset sensitive areas to obtain the area set; based on the area set and date, the spatial co-occurrence data of the day can be recalled in the object track area index data, and then the spatiotemporal co-occurrence objects can be obtained by filtering based on the time threshold.

[0068] As the co-occurring object data accumulates, the support and confidence will continue to change. Finally, the co-occurring data pairs will be included in the database after manual confirmation. Then, the subsequent behavior prediction will be carried out based on the similar historical data of the objects. The more data is accumulated, the more likely it is to find the behavior preferences of each object, and the intention and behavior recognition effect will become more and more accurate.

[0069] The regional set is constructed through the following steps:

[0070] Set spatiotemporal co-occurrence conditions, including distance threshold and time threshold;

[0071] Select the current sensitive area, and expand the current sensitive area outward according to the distance threshold to obtain the expanded sensitive area;

[0072] Determine an intersection sensitive area; wherein the intersection sensitive area is a sensitive area that intersects with the expansion sensitive area;

[0073] The construction of the region set of the current sensitive region is completed by dilating the sensitive region and taking the union of the intersecting sensitive regions as the region set.

[0074] It should be noted that each business unit defines some sensitive areas according to different areas of concern. These areas may overlap or be far apart. When constructing the area set, the original demarcated area will be expanded, and then the union of the areas contained in it will be taken as the area set to recall the object track area index data.

[0075] like Figure 3 As shown in the figure, A, B, C, D, E, and F represent the originally delineated sensitive areas. Assuming that B is the current area to be identified, its original area intersects with C. After B is expanded according to the distance threshold, the area shown by the dotted line in the figure intersects with A. Therefore, A, B, and C will be used as the designated area set when recalling the local area index data in the object track area.

[0076] The reason for expanding the spatiotemporal co-occurrence area is that in many cases, the co-occurrence between multiple objects is not limited to a narrow fixed space; multiple objects with a common object performing different actions in different areas can also be regarded as a kind of co-occurrence, and different combinations of different objects may also be for the purpose of achieving different specific tasks.

[0077] For example, Figure 4 As shown in (a), there is an X-type ship in area A and an S-type aircraft in area B. When X and S co-occur and the task performed is T1; Figure 4 In (b), there is an X-type ship in area A and a P-type aircraft in area B. When X and P co-occur and the task performed is T2; Figure 4In (c), there is an X-type ship in area F, an S-type aircraft in area D, and a P-type aircraft in area E, and the mission they are performing is T3.

[0078] After obtaining the area set in the previous step, recall the candidate set in the object track area index table based on the date, then select the time period according to the preset time threshold, and select the spatiotemporal co-occurrence objects through the following steps:

[0079] Get the area set and recall the object track area index data corresponding to the area set;

[0080] Add the track area index data of the recalled object into the same recall candidate set according to the same track date;

[0081] Select the recall candidate set, and select the time period by the preset time threshold, filter out the object track area index data in the same time period, and add it to the same spatiotemporal co-occurrence object set;

[0082] Acquire each multivariate model combination in the spatiotemporal co-occurrence object set; wherein the multivariate model combination is a combination of different object models in the selected object track area index data;

[0083] Determine the tasks or intentions that the multi-model combination corresponds to through prior knowledge;

[0084] If the acquisition is successful, the multivariate model combination is a known combination, and the corresponding task or intention to be performed is output;

[0085] If it cannot be determined, the multivariate model combination is an unknown combination, and the corresponding support and confidence are output.

[0086] When selecting a time period by using a preset time threshold, if there are multiple object track area index data with the same object number in the same time period, they are regarded as one object track area index data with the same object number. The time period selection formula is as follows:

[0087] Time=[Tt,T+t]

[0088] Where T is the current time point, t is the time threshold, and Time is the selected time period.

[0089] After merging the redundant data, the cumulative count is performed according to the object model, and the following is obtained: Figure 4 The result set shown in the figure is as follows. The final result set ignores the individual numbers and takes the model as the statistical basis, and the result is as follows Figure 5 The statistical results are shown.

[0090] The prior knowledge is the tasks or intentions performed by the multivariate model combination that has been manually confirmed in the past, as shown in Table 2 below:

[0091] Table 2 Schematic diagram of prior knowledge

[0092]

[0093] At this point, based on prior knowledge, we can know that the aircraft of type 0301 and the ship of type 0203 in the above result set are performing fishing operations.

[0094] If the result is not in the prior knowledge, the support and confidence levels are used to filter out the combinations that need to be sent for manual confirmation.

[0095] Support refers to the number of instances of co-occurrence of objects in the above result set. When the number of co-occurrences exceeds the set threshold, it is determined as a spatiotemporal co-occurrence object that requires manual confirmation. Confidence refers to the participation rate of instances that conform to the spatiotemporal co-occurrence pattern. The confidence calculation formula is:

[0096]

[0097] In the formula, f i ∈E, E is the recall candidate set, f i is the current spatiotemporal co-occurring object set, i is the number of the time step in the candidate set region, and k is the number of objects contained in the current instance.

[0098] The following is an example of the above formula in combination with Table 3:

[0099] Table 3 Confidence calculation diagram

[0100]

[0101] The total number of k-ary instances in fi is the number of k-ary instances in fi: 6+10+3+1=20;

[0102] The number of unknown k-ary instances in fi, that is, the k-ary instances in fi that are not in the prior knowledge; assuming that there are 3 unknown combinations; then confidence = 1-(3 / 20) = 0.85

[0103] Here, support and confidence jointly define whether manual intervention is needed. Assuming that the confidence is 0 when the unknown number = the total number, it does not mean that manual intervention is not needed. At this time, the unknown number should have met the support threshold setting, so manual intervention is also required. Only when the confidence and support are not met, manual confirmation is not required.

[0104] Among them, various combinations of aircraft and ships constitute candidate spatiotemporal co-occurrence sets. If they still satisfy the confidence level > θ, they are spatiotemporal co-occurrence objects, where θ is the confidence threshold. Finally, the intention and task of the co-occurrence set are manually judged and included in the prior knowledge database.

[0105] The above are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A multi-object intention recognition method based on spatiotemporal co-occurrence mining, characterized in that: It includes object track area index construction, spatiotemporal co-occurrence object extraction and multi-object intention recognition. Object track area index construction link: by acquiring the trajectory data of the object in the track area, construct the corresponding object track area index data; wherein, the track area is provided with a number of sensitive areas, and the object track area index data is used to describe the object's entry and exit of the sensitive area; Spatiotemporal co-occurrence object extraction phase: the sensitive area originally delineated based on the distance threshold is expanded, and the result set is determined through the expanded sensitive area. The object track area index data is recalled through the result set, and the object track area index data is screened to obtain the spatiotemporal co-occurrence object set; Multi-object intention recognition link: obtain each multi-model combination through the spatiotemporal co-occurrence object set, and determine the task or intention corresponding to the multi-model combination through prior knowledge; among them, if the current multi-model combination is an unknown combination, the corresponding support and confidence are output and the unknown combination is confirmed manually.

2. The multi-object intention recognition method based on spatiotemporal co-occurrence mining according to claim 1 is characterized in that: The object track area index construction phase is performed through the following steps: Step 1: pre-divide several sensitive areas in the track area; Step 2: Obtain a result set of sensitive areas passed by the object track through a spatiotemporal database; wherein the spatiotemporal database records the trajectory data of each object; Step 3: Obtain the entry and exit time of the object track entering and exiting each sensitive area through the result set; Step 4: Output the required object track area index data.

3. The multi-object intention recognition method based on spatiotemporal co-occurrence mining according to claim 2 is characterized in that: In step 3, the entry and exit times are also rounded; the entry time for entering the sensitive area is rounded down, and the entry time for leaving the sensitive area is rounded up; if the same object continuously enters and exits the same area, the earliest entry time and the latest exit time are selected as the entry and exit times.

4. The multi-object intention recognition method based on spatiotemporal co-occurrence mining according to claim 2 is characterized in that: In step 4, the object track area index data includes object number, area number, object model, area time step and track date; wherein the object number is uniquely corresponding, the area time step is divided based on the time step of 0-23, and is determined by the entry and exit time.

5. The multi-object intention recognition method based on spatiotemporal co-occurrence mining according to claim 1 is characterized in that: The regional set is constructed through the following steps: Set spatiotemporal co-occurrence conditions, including distance threshold and time threshold; Select the current sensitive area, and expand the current sensitive area outward according to the distance threshold to obtain the expanded sensitive area; Determine an intersection sensitive area; wherein the intersection sensitive area is a sensitive area that intersects with the expansion sensitive area; The construction of the region set of the current sensitive region is completed by dilating the sensitive region and taking the union of the intersecting sensitive regions as the region set.

6. The multi-object intention recognition method based on spatiotemporal co-occurrence mining according to claim 4 is characterized in that: The spatiotemporal co-occurrence objects are selected through the following steps: Get the area set and recall the object track area index data corresponding to the area set; Add the track area index data of the recalled object into the same recall candidate set according to the same track date; Select the recall candidate set, and select the time period by the preset time threshold, filter out the object track area index data in the same time period, and add it to the same spatiotemporal co-occurrence object set; Acquire each multivariate model combination in the spatiotemporal co-occurrence object set; wherein the multivariate model combination is a combination of different object models in the selected object track area index data; Determine the tasks or intentions that the multi-model combination corresponds to through prior knowledge; If the acquisition is successful, the multivariate model combination is a known combination, and the corresponding task or intention to be performed is output; If it cannot be determined, the multivariate model combination is an unknown combination, and the corresponding support and confidence are output.

7. The multi-object intention recognition method based on spatiotemporal co-occurrence mining according to claim 6 is characterized in that: When selecting a time period by using a preset time threshold, if there are multiple object track area index data with the same object number in the same time period, they are regarded as one object track area index data with the same number. The time period selection formula is as follows: Time=[Tt,T+t] Where T is the current time point, t is the time threshold, and Time is the selected time period.

8. The multi-object intention recognition method based on spatiotemporal co-occurrence mining according to claim 6 is characterized in that: The confidence level is the participation rate of instances that conform to the spatiotemporal co-occurrence pattern.

9. The multi-object intention recognition method based on spatiotemporal co-occurrence mining according to claim 6 is characterized in that: The support degree is the accumulated count value of the selected object track area index data; wherein the accumulated count ignores the object number and takes the object model as the statistical basis.

10. A multi-object intention recognition system based on spatiotemporal co-occurrence mining, characterized in that: A multi-object intention recognition method based on spatiotemporal co-occurrence mining as described in any one of claims 1 to 9 is used to perform object track area index construction, spatiotemporal co-occurrence object extraction and multi-object intention recognition.

Images