Motion group classification and behavior pattern recognition method based on trajectory semantic similarity

Abstract

The invention discloses a motion group classification and behavior pattern recognition method based on track semantic similarity. The sports group classification method comprises the following steps: step 1, acquiring riding sports data to obtain riding track data; step 2, performing road network matching to obtain a road type of each riding track point; segmenting each riding track according to the road type to obtain a plurality of riding track segments; behavior attributes and motion attributes are determined, attribute calculation is carried out, and semantic assignment is carried out on the riding track segments to obtain semantic track segments; step 3, calculating semantic trajectory similarity to obtain a similarity matrix; and step 4, a similarity network is constructed according to the similarity matrix, community detection is performed by using the similarity network to obtain a plurality of communities, and each community is a sports group category. According to the method, exercise group classification and complex riding exercise behavior pattern recognition can be effectively carried out, and a scientific basis is provided for planning and construction of cities such as greenways and personalized services and management for different exercise groups.

Description

Technical Field

[0001] The invention belongs to the field of data mining and sports behavior pattern data analysis, and relates to a sports group classification and behavior pattern recognition method based on trajectory semantic similarity. Background Art

[0002] With the improvement of health awareness and the increasing emphasis on sustainable tourism, cycling tourism has become an emerging trend in the modern tourism industry. As a form of travel that is healthy, environmentally friendly, flexible and can explore local scenery, cycling tourism has not only developed significantly, but has also gradually evolved into a key development driver for promoting sustainable growth in local environmental, social and economic issues. However, due to the complexity of cycling tourism behavior and the diverse experiences of different individuals due to their different cycling purposes, methods and needs, there are many challenges and difficulties in fully revealing cycling tourism behavior. Therefore, in-depth exploration of cycling tourism behavior is of great significance for promoting destination planning, providing personalized services and management, and promoting the development of sustainable tourism.

[0003] Behavioral complexity requires the support of big data, but the research on bicycle tourism has not been fully developed in terms of data. Data acquisition in the field of bicycle tourism mainly adopts traditional questionnaire surveys to reveal tourists' preferences through pre-prepared survey content, rather than based on big data. Questionnaire surveys and travel diaries rely on respondents' subjective feedback and memories of travel activities, requiring them to record carefully and diligently, which is a relatively coarse-grained measurement method. In addition, traditional data collection methods rely on the cooperation of respondents, and this high degree of cooperation usually leads to the problem of limited sample size. With the development and widespread application of big data technology, new data sources represented by GPS trajectory data are increasingly used in tourism behavior research. GPS data objectively and accurately records tourists' time, geographic location, distance, speed and other information, and realizes fine-grained tracking of travel activity paths. GPS trajectory data generated by users not only has a large sample size, long sample period and high resolution, but is also applicable to a variety of complex environments (such as cities, natural environments, etc.), and can accurately describe information such as cyclists' routes, speeds and distances. It provides data support for revealing the complexity of cyclists' behavior from a fine-grained perspective.

[0004] Time, space and activity are three important areas of tourism behavior research. The spatiotemporal behavior pattern can reveal "when", "where" and "how" tourists move, but it cannot deeply reveal the purpose and preference of tourists' behavior. In terms of space, the study of spatiotemporal behavior pattern simplifies tourist behavior into a topological structure and only focuses on the trajectory of travelers. Although the topological structure and spatiotemporal pattern can explain the structure of tourism activities, they greatly ignore the content of the activities themselves and the quality of tourists' experience. The mode of cycling tourism is not limited to the spatiotemporal pattern, but the style and experience of tourists are more critical.

[0005] As individual behaviors become increasingly complex and diverse, semantic trajectory similarity calculation becomes the key to trajectory similarity mining and application. Trajectory similarity can identify travelers who exhibit similar movement patterns, and then classify movement behaviors of individuals with similar interests and preferences according to homogeneous travel groups. Existing studies on tourist behavior focus on exploring the spatiotemporal characteristics of tourists, such as considering the spatial and temporal dimensions of tourist travel flows and conducting spatiotemporal analysis of the trajectory similarity of individual travelers. From the perspective of time and space, a heuristic method combining dynamic time warping (DTW) and the earth mover's distance (EMD) is proposed to measure the trajectory similarity of tourists. However, the above methods only use temporal and spatial information to explore the flow of tourist movements, and do not conduct in-depth exploration of the complexity of tourist behavior. Semantic trajectories can contain multi-faceted information, bringing rich semantic knowledge to the interpretation of the complexity of tourist behavior, such as contextual information and background geographic information. Semantic trajectory similarity also has many representative algorithms, such as multidimensional similarity measure (MSM), multi-aspect semantic trajectory similarity (MUITAS), maximum minimum trajectory distance and sum of minimum trajectory distance (i.e., MMTD and SUMTD). Therefore, semantic trajectories are an effective way to deeply reveal the behavior of cycling tourists. Summary of the invention

[0006] In view of the defects or shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a method for sports group classification and behavior pattern recognition based on trajectory semantic similarity, so as to solve the problem that the existing recognition method has a traditional data collection method, only considers the topological structure and spatiotemporal pattern, and fails to consider the rider's preferences and style.

[0007] In order to achieve the above tasks, the present invention adopts the following technical solutions:

[0008] On the one hand, the present invention provides a method for classifying a group of athletes based on trajectory semantic similarity, which specifically includes the following steps:

[0009] Step 1, obtaining cycling motion data, and performing data preprocessing to obtain cycling track data; the cycling motion data includes cycling GPS track data and road network data, and the cycling GPS track data is composed of relevant data of all cycling track points;

[0010] Step 2, using the cycling GPS trajectory data to perform road network matching, and obtaining the road type of each cycling trajectory point on each cycling trajectory; using the road type to segment each cycling trajectory into multiple cycling trajectory segments; then determining the information of the behavior attribute and the motion attribute, performing attribute calculation and using the attribute calculation result to perform semantic assignment on each segmented cycling trajectory segment, and obtaining the semantic trajectory segment of each cycling trajectory segment, and the semantic trajectory segments of all the cycling trajectory segments constitute the semantic trajectory corresponding to the cycling trajectory;

[0011] Step 3, using the semantic trajectory corresponding to each riding trajectory obtained in step 2, calculate the semantic trajectory similarity between each riding trajectory, and obtain a similarity matrix composed of the semantic trajectory similarities of all riding trajectories;

[0012] Step 4: Based on the similarity matrix obtained in step 3, a similarity network is constructed using the KNN algorithm. The nodes in the similarity network represent semantic trajectories, and the edges represent the similarity between two semantic trajectories. The similarity network is used to perform community detection to obtain multiple communities, each of which is a movement group.

[0013] Furthermore, in step 1, the data preprocessing includes: (1) eliminating trajectory data outside the geographical scope of the study area; (2) deleting trajectory points with missing attribute information from the data, and identifying and deleting duplicate trajectory data through all field attribute information; (3) eliminating trajectory data above the highest altitude; and (4) removing abnormal data based on riding speed.

[0014] Furthermore, in step 2, the road network matching is performed based on the nearest neighbor search method of KDTree, including the following process:

[0015] Step 1), road position representation: abstract the roads in the road network into a number of inflection points, and the spatial position of each road can be represented by the coordinates of these inflection points;

[0016] Step 2), KDTree construction: construct KDTree for the coordinates of all turning points of all roads in the road network to obtain a KDTree;

[0017] Step 3), neighbor search: perform neighbor search on the cycling trajectory points in step 1 and the KDTree obtained in step 2), obtain the road breakpoint closest to each cycling trajectory point, and use the road type of the road breakpoint as the road type of the corresponding cycling trajectory point.

[0018] Furthermore, in step 2, the behavior attributes include riding time, road altitude, and road type, and the motion attributes include average speed and riding distance; the attribute calculation is divided into behavior attribute calculation and motion attribute calculation.

[0019] Furthermore, in step 3, the semantic trajectory similarity between the riding trajectories is obtained by the following formula:

[0020]

[0021] in:

[0022] sim_ST(ep i ,ep k ) represents any two cycling trajectories ep i ,ep k The similarity of the semantic trajectories;

[0023] and represents the i-th and k-th semantic tracks, m and n refer to the number of semantic track segments in these two semantic tracks respectively; Represents the semantic track ST i Semantic track fragments in and semantic track ST k Semantic track fragments in similarity;

[0024] Represents the semantic track ST i Each semantic track segment Instead of being in the semantic track k The most similar semantic trajectory fragments The sum of similarities;

[0025] Represents the semantic track ST k Each semantic track segment Instead of being in the semantic track i The most similar semantic trajectory fragments The sum of similarities;

[0026] From two semantic tracks ST i 、ST k The similarity of two semantic trajectory segments is obtained by the following formula:

[0027]

[0028] Among them, w P For each attribute's weight, p=5; matching function match p(A p ) are the corresponding attributes A of any two semantic trajectory segments P The similarity between the behavior attributes of the two semantic trajectory segments is calculated by the semantic matching function, and the similarity between the motion attributes of the two semantic trajectory segments is calculated by the distance matching function.

[0029] Furthermore, the weight of each attribute is set to w 1 =0.063, w 2 =0.098, w 3 =0.372, w 4 =0.103, w 5 =0.364.

[0030] Furthermore, the semantic matching function is obtained by the following formula:

[0031]

[0032] in, Is to get attribute A P The function of the category to which it belongs, The equal sign in the slash indicates that any two semantic track segments have the same attribute A. P Belong to the same category.

[0033] Furthermore, the distance matching function is obtained by the following formula:

[0034]

[0035] in, and Attribute A P The distance function and distance threshold are as follows: the distance function adopts d(X,Y)=|XY|, where X and Y represent the motion attribute values ​​corresponding to any two semantic trajectory segments.

[0036] Furthermore, in step 4, in the construction of the similarity network, K=10 in the KNN algorithm.

[0037] On the other hand, the present invention provides a method for identifying motion behavior patterns based on trajectory semantic similarity, the method comprising: merging groups with the same cycling tourism pattern among multiple cycling sports groups to obtain pattern classification results of the cycling sports groups, the cycling sports groups being obtained by the above-mentioned motion group classification method based on trajectory semantic similarity of the present invention.

[0038] Compared with the prior art, the present invention has the following technical effects:

[0039] (1) By introducing multi-dimensional semantic information such as riding time, road altitude, riding road type, average speed and riding distance, the present invention realizes the multi-dimensional feature integration and fragmentation of the original riding GPS trajectory, thereby converting it into a more expressive riding semantic trajectory. This improvement not only enhances the interpretability of trajectory representation, but also overcomes the limitations of traditional trajectory representation methods, laying the foundation for in-depth data analysis.

[0040] (2) The present invention calculates the semantic similarity of multiple attributes (such as road type, speed, etc.) of semantic trajectory segments and comprehensively evaluates the similarity between semantic trajectories from multiple dimensions. It can comprehensively and accurately reflect the actual similarity between riding trajectories. This method is superior to the traditional GPS-based trajectory similarity calculation method, which mainly considers the similarity of spatial position and time, but does not fully consider other complex situations of the riding environment.

[0041] (3) The present invention integrates road types into coarse-grained categories (e.g., unifying highways and highway connecting lines into highways), avoiding excessive subdivision of road types that results in too many semantic trajectory segment categories for a semantic trajectory, preventing frequent changes in road types matched by riding trajectory points, and reducing the complexity of semantic trajectory similarity calculation. This simplified classification method can better aggregate trajectories with similar semantic features, thereby improving the accuracy and efficiency of riding trajectory group classification.

[0042] (4) The present invention adopts a community detection algorithm to identify groups with intrinsic associations by analyzing the complex network relationships between semantic tracks. Unlike traditional clustering algorithms that only rely on spatial similarity, community detection algorithms can reveal the underlying structure in semantic track data and consider the topological relationships and interactions between semantic tracks, thereby providing more accurate group classification. This method effectively captures complex group characteristics and relationships and improves the accuracy of group classification.

[0043] In summary, the method of the present invention can effectively classify sports groups and identify complex cycling sports behavior patterns, thereby promoting destination planning, providing personalized services and management, and promoting sustainable tourism development. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is the research framework of the method of the present invention;

[0045] Figure 2 It is the process of semantic trajectory construction;

[0046] Figure 3 It is the relationship between K value and Silhouette coefficient of KNN;

[0047] Figure 4 is the community detection algorithm result in the embodiment of the present invention;

[0048] Figure 5 It is the cycling tourism mode and cycling tourism activity behavior;

[0049] Figure 6 The distribution of leisure and sightseeing riding modes and typical riding routes in the embodiment of the present invention;

[0050] Figure 7 The distribution of sports challenge riding modes and typical riding routes in the embodiments of the present invention;

[0051] Figure 8 It is the distribution of fitness exercise riding modes and typical riding routes in the embodiment of the present invention.

[0052] The present invention is further explained in detail below with reference to the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0053] Example 1

[0054] See also Figure 1 , Figure 2 This embodiment provides a method for classifying a sports group based on trajectory semantic similarity, which specifically includes the following steps:

[0055] Step 1: Obtain cycling motion data, perform data preprocessing, and obtain cycling trajectory data.

[0056] Specifically, the cycling sports data includes cycling GPS track data and road network data (including road types). The cycling GPS track data is track data uploaded by cycling enthusiasts in Xi'an area crawled from the 2bulu outdoor travel website; the road network data is urban road network data obtained from OpenStreetMap (OSM for short).

[0057] The cycling GPS trajectory data consists of the relevant data of all cycling trajectory points. For example, the cycling GPS trajectory data shown in Table 1 records in detail the participant's user ID, trajectory longitude, latitude, date and time, speed, altitude and other multi-dimensional motion information.

[0058] Table 1. Example of cycling GPS trajectory data

[0059] User ID longitude Dimensions time Speed ​​(m / s) Altitude(m) 1024***67 108.*** 34.*** 2019-11-8 14:11:43 0.4853 411.2028 1024***67 108.*** 34.*** 2019-11-8 14:11:47 1.6229 411.7378 1024***67 108.*** 34.*** 2019-11-8 14:11:41 3.1387 412.2087 1024***67 108.*** 34.*** 2019-11-8 14:11:55 3.1514 412.4787 1024***67 108.*** 34.*** 2019-11-8 14:11:59 3.0886 412.7186

[0060] Specifically, in this embodiment, due to factors such as GPS signal loss, equipment error, and weather environment, the open GPS trajectory data has missing information and inaccurate information. In order to eliminate the impact of data errors on the research results, this embodiment adopts the following data preprocessing: (1) Trajectory data outside the geographical scope of the study area are eliminated, and the administrative area of ​​Xi'an is selected as the study area of ​​this embodiment. (2) Trajectory points with missing attribute information are deleted from the data, and duplicate trajectory data are identified and deleted through all field attribute information. (3) With reference to the highest altitude of Xi'an, trajectory data beyond 4000m are eliminated. (4) Abnormal data is removed according to riding speed. The riding data collected in this embodiment are distributed in urban areas and mountainous areas. Considering that the instantaneous speed during mountain riding downhill can reach very high, this embodiment sets the speed threshold to a relatively large value of 30m / s. Through the data cleaning and screening steps, 3532 valid riding trajectories were finally obtained, which contained a total of 6079556 trajectory points.

[0061] Step 2, using the cycling GPS trajectory data obtained in step 1 to perform road network matching, and obtain the road type of each cycling trajectory point on each cycling trajectory; using the road type to segment each cycling trajectory into multiple cycling trajectory segments; then determining the information of behavior attributes and motion attributes, performing attribute calculation and using the attribute calculation results to perform semantic assignment on each segmented cycling trajectory segment, and obtain the semantic trajectory segment of each cycling trajectory segment, and the semantic trajectory segments of all the cycling trajectory segments constitute the semantic trajectory corresponding to the cycling trajectory.

[0062] In this embodiment, a neighbor search method based on KDTree (K-Dimensional Tree) is used to match the road network, including the following process: Step 1), road position representation: roads in the road network are abstracted into a number of turning points (also called inflection points or intersections), and the spatial position of each road can be represented by the coordinates of these turning points. Step 2), KDTree construction: KDTree is constructed for the coordinates of all turning points of all roads in the road network to obtain a KDTree. Step 3), neighbor search: A neighbor search is performed on the riding trajectory points in step 1 and the KDTree obtained in step 2) to obtain the road turning point closest to each riding trajectory point, and the road type of the road turning point is used as the road type of the corresponding riding trajectory point; this method can quickly find the closest road turning point between the riding trajectory point and the road network, thereby achieving road network matching.

[0063] Specifically, segmenting the cycling track means: splitting a given cycling track into a series of non-intersecting sub-tracks with similar road types. Since cycling tourism activities are constrained by the road network, different road types directly reflect the cycling preferences of travelers. Therefore, in the present invention, based on the road type information of the track points, the track points with the same road type in the cycling track are merged to achieve segmentation of the cycling track and obtain multiple cycling track segments.

[0064] Specifically, attributes are the cycling behavior characteristics and movement characteristics of the cycling community in five dimensions: cycling time, road altitude, cycling road type, average speed, and cycling distance. Behavioral attributes include cycling time, road altitude, and road type, and movement attributes include average speed and cycling distance. Attribute calculation is divided into behavioral attribute calculation and movement attribute calculation.

[0065] Among them, the behavior attribute calculation is after the trajectory is segmented by road type information. The riding time of each riding trajectory segment is represented by the time of the first GPS point, that is, the time when the user arrives at the road segment of this road type; the road altitude is the difference between the maximum and minimum altitudes of the GPS points in each riding trajectory segment.

[0066] In the calculation of motion attributes, the average speed refers to the average speed of all GPS track points in each cycling track segment, and the cycling distance refers to the sum of the Euclidean distances between all GPS track points in each cycling track segment.

[0067] Step 3: Using the semantic trajectory corresponding to each riding trajectory obtained in step 2, the semantic trajectory similarity between two riding trajectories is calculated to obtain a similarity matrix composed of the semantic trajectory similarities of all riding trajectories.

[0068] Preferably, the semantic trajectory similarity is obtained by formula 1:

[0069]

[0070] in:

[0071] sim_ST(ep i ,ep k ) represents any two cycling trajectories ep i ,ep k The similarity of the semantic trajectories;

[0072] and represents the i-th and k-th semantic tracks, m and n refer to the number of semantic track segments in these two semantic tracks respectively; Represents the semantic track ST i Semantic track fragments in and semantic track ST k Semantic track fragments in similarity;

[0073] Represents the semantic track ST i Each semantic track segment Instead of being in the semantic track k The most similar semantic trajectory fragments The sum of similarities;

[0074] Represents the semantic track ST k Each semantic track segment Instead of being in the semantic track i The most similar semantic trajectory fragments The sum of similarities;

[0075] Preferably, from two semantic tracks ST i 、ST k The similarity of two semantic trajectory segments is obtained by formula 2:

[0076]

[0077] Among them, w P For each attribute's weight, p=5; matching function match p (A p ) are the corresponding attributes A of any two semantic trajectory segments P Each semantic segment contains P attributes, including 3 behavioral attributes and 2 motion attributes. The similarity between the behavioral attributes of any two semantic trajectory segments is calculated by the semantic matching function, while the similarity between the motion attributes of any two semantic trajectory segments is calculated by the distance matching function.

[0078] The weight of each attribute w P The determination is based on information entropy, which is a commonly used statistical tool to measure the uncertainty or information content of a variable. It can effectively evaluate the dispersion and importance of each attribute in the overall data set. Specifically, in this embodiment, the weight of each attribute is set to w 1 =0.063, w 2 =0.098, w 3 =0.372, w 4 =0.103, w 5 =0.364.

[0079] Preferably, the semantic matching function is obtained by Formula 3:

[0080]

[0081] in, Is to get attribute A P The function of the category to which it belongs, The equal sign in the slash indicates that any two semantic track segments have the same attribute A. P Belong to the same category. In this embodiment, the riding road types in the behavior attributes are classified into highways, urban roads, bicycle lanes, residential roads, sightseeing roads and suburban roads (Appendix 2); the riding time is classified into morning (6:00-12:00), afternoon (12:00-18:00) and evening (18:00-6:00); the road altitude is classified into gentle (road altitude difference ≤ 100) and challenging (road altitude difference > 100).

[0082] Table 2 Classification of road types

[0083]

[0084]

[0085]

[0086] Preferably, the distance matching function is obtained by formula 4:

[0087]

[0088] in, and Attribute A P In the present invention, the distance function is: d(X,Y)=|XY|, where X and Y represent the motion attribute values ​​corresponding to any two semantic trajectory segments; in this embodiment, according to the application background of cycling tourism activities and the distribution of the cycling data set of the present invention, the distance threshold for the average speed is set to 1.4m / s, and the distance threshold for the cycling distance is set to 3038m.

[0089] Step 4: Based on the similarity matrix obtained in step 3, a similarity network is constructed using the KNN algorithm. The nodes in the similarity network represent semantic trajectories, and the edges represent the similarity between two semantic trajectories. The similarity network is used to perform community detection to obtain multiple communities, each of which is a sports group. These groups show similar cycling tourism activities, such as Figure 5 shown.

[0090] Specifically, in the actual construction of the similarity network, the setting of the K value in the KNN algorithm is closely related to the result of community classification. The present invention selects the Silhouette coefficient as the evaluation index of the clustering effect to determine the appropriate K value for setting KNN. Figure 3 It can be seen that when K=7, 10 and 15, the Silhouette coefficient value is relatively high, and the corresponding number of communities is 2, 9 and 7. Through comparative analysis, the clustering effect is most significant when K=10, so the present invention selects 9 communities under this parameter as the classification result of Xi'an's cycling behavior pattern.

[0091] Specifically, the community detection algorithm (Infomap algorithm) detects multiple communities by analyzing the similarity network.

[0092] In this example, the cycling preferences of different communities are sorted out from three aspects: cycling behavior characteristics (i.e., behavior attributes), cycling movement characteristics (i.e., movement attributes), and other characteristics, and each community is sorted into cycling tourism activity behaviors with differentiated characteristics (see Figure 5 In this example, 9 communities with significant cycling behavior characteristics were successfully detected. Each community collected a large number of cycling tracks, fully demonstrating different cycling tourism activities (see Figure 4 Among these communities, Comm1 is the largest community with 664 cycling tracks (19% of the total), followed by Comm2 with 607 cycling tracks (17%), while the smallest community is Comm9, which contains 143 cycling tracks, accounting for 4% of the total sample.

[0093] Embodiment 2:

[0094] This embodiment provides a method for identifying a motion behavior pattern based on trajectory semantic similarity, the method comprising the following steps: merging groups with the same cycling travel pattern among multiple cycling sports groups to obtain a pattern classification result of the cycling sports groups.

[0095] Specifically, the cycling sports group is obtained by the sports group classification method based on trajectory semantic similarity of the present invention.

[0096] In order to discover and summarize the internal connections between behaviors with the same functional characteristics and simplify the complexity caused by too many subdivisions, in this embodiment, activities with similar functions are combined into one pattern (see Figure 5 ). This approach helps to identify and meet the needs of different tourists and promote the diversification and personalization of cycling tourism activities. Figure 5 In the riding time Represents morning, afternoon and evening respectively; and They represent gentle riding and challenging riding respectively; the background color of the cell under average speed indicates the speed rating: light green - low speed, light yellow - medium speed, light red - high speed; They represent distance ratings: short ride, medium distance ride, and long ride.

[0097] Finally, this embodiment divides Xi'an cycling behavior into three cycling tourism modes:

[0098] (1) Leisure and sightseeing riding mode.

[0099] This mode covers two types of cycling tourism activities: sightseeing and exploration cycling (comm6) and night leisure cycling (comm7). It is a cycling tourism mode that pursues leisure and entertainment and enjoys the city's daytime and nighttime scenery. This type of cycling tourists focuses on the leisure function of cycling rather than pure mobility efficiency, which is mainly manifested in slow cycling speed, short cycling distance, and larger groups of people riding in the afternoon and evening. Cycling routes are distributed in both urban and suburban areas, and mainly tend to surround or pass through roads and areas with sightseeing value (sightseeing roads and bicycle roads), such as the city wall, Bahe Greenway, Weihe Greenway, Qujiang Lake, Daming Palace, Kunlun Lake and other riverside, greenways and parks ( Figure 6 ). It is worth noting that with West Street and East Street inside the city wall, West Youyi Road and East Youyi Road south of the city wall as the main sections, specific trajectories outline the route shapes of "2021" and "2024" in the urban space, further demonstrating the creativity and fun of leisure sightseeing cycling mode in cycling tourism.

[0100] (2)Sports challenge riding mode.

[0101] This mode combines the two cycling tourism activities of short-distance challenge cycling (comm2) and extreme speed challenge cycling (comm8). It is a cycling mode that emphasizes short-distance mountain challenges, pursues high-speed experience and skill challenges. This mode has flexible cycling time and simple and direct cycling routes. It is distributed in cities, suburbs and mountainous areas, covering multiple terrain areas. This cycling group shows a strong preference for mountainous terrain. It not only requires high cycling skills in terms of cycling speed, but also requires them to be able to flexibly cope with changing terrain and road conditions. Figure 7 As shown in the figure, Fengyu, Lishan, Dayu, Taipingyu, Daiyu and Wangyu are hot spots for mountain biking, such as Fengyu→Dabagou→Jiwozi, Lishan National Forest Park→Longwangmiaogou→Hongqingshan National Forest Park, Dayu→Xianyukou→Xigou, Taipingyu→Hongyuhe→Qinling Alpine Camp→Taiping National Forest Park, etc. are popular mountain biking routes for cyclists. At the same time, there are also cyclists who prefer to start from the city, cross the suburbs, and challenge the extreme routes in the mountainous area, which reflects the needs of this cycling group in pursuit of speed, skill challenges and cycling endurance.

[0102] (3) Fitness training cycling mode.

[0103] This mode is mainly composed of three cycling tourism activities: long-distance exercise cycling (comm1), urban exercise cycling (comm3, 4, 5) and commuting exercise cycling (comm9). It is a cycling mode that focuses on long-distance fitness cycling, daily cycling in urban areas, and morning commuting fitness (see Figure 8 ). This type of cycling group focuses on the convenience of cycling, and mainly chooses to carry out cycling activities in urban areas close to residential areas, using convenient daily methods such as commuting. This type of cycling mode is usually manifested as cycling around the city, cycling in the suburbs of the city, and the main cycling routes cover the city, suburbs and Mount Li area, mainly choosing direct and efficient urban roads and living roads. In addition, the study found that although participants generally showed a preference for bicycle lanes, their cycling starting points and destinations were often significantly away from bicycle lanes, which caused a certain degree of difficulty in use in practice.

Claims

1. A method for classifying sports groups based on trajectory semantic similarity, characterized in that: The specific steps include: Step 1, obtaining cycling motion data, and performing data preprocessing to obtain cycling track data; the cycling motion data includes cycling GPS track data and road network data, and the cycling GPS track data is composed of relevant data of all cycling track points; Step 2, using the cycling GPS trajectory data to perform road network matching, and obtaining the road type of each cycling trajectory point on each cycling trajectory; using the road type to segment each cycling trajectory into multiple cycling trajectory segments; then determining the information of the behavior attribute and the motion attribute, performing attribute calculation and using the attribute calculation result to perform semantic assignment on each segmented cycling trajectory segment, and obtaining the semantic trajectory segment of each cycling trajectory segment, and the semantic trajectory segments of all the cycling trajectory segments constitute the semantic trajectory corresponding to the cycling trajectory; Step 3, using the semantic trajectory corresponding to each riding trajectory obtained in step 2, calculate the semantic trajectory similarity between each riding trajectory, and obtain a similarity matrix composed of the semantic trajectory similarities of all riding trajectories; Step 4: Based on the similarity matrix obtained in step 3, a similarity network is constructed using the KNN algorithm. The nodes in the similarity network represent semantic trajectories, and the edges represent the similarity between two semantic trajectories. The similarity network is used to perform community detection to obtain multiple communities, each of which is a movement group.

2. The method for classifying sports groups based on trajectory semantic similarity according to claim 1, characterized in that: In step 1, the data preprocessing includes: (1) eliminating trajectory data outside the geographical scope of the study area; (2) deleting trajectory points with missing attribute information from the data, and identifying and deleting duplicate trajectory data through all field attribute information; (3) eliminating trajectory data above the highest altitude; and (4) removing abnormal data based on riding speed.

3. The method for classifying sports groups based on trajectory semantic similarity according to claim 1, characterized in that: In step 2, the road network matching is performed based on the KDTree neighbor search method, which includes the following process: Step 1), road position representation: abstract the roads in the road network into a number of inflection points, and the spatial position of each road can be represented by the coordinates of these inflection points; Step 2), KDTree construction: construct KDTree for the coordinates of all turning points of all roads in the road network to obtain a KDTree; Step 3), neighbor search: perform neighbor search on the cycling trajectory points in step 1 and the KDTree obtained in step 2), obtain the road breakpoint closest to each cycling trajectory point, and use the road type of the road breakpoint as the road type of the corresponding cycling trajectory point.

4. The method for classifying sports groups based on trajectory semantic similarity according to claim 1, characterized in that: In step 2, the behavior attributes include riding time, road altitude, and road type, and the motion attributes include average speed and riding distance; the attribute calculation is divided into behavior attribute calculation and motion attribute calculation.

5. The method for classifying sports groups based on trajectory semantic similarity according to claim 1, characterized in that: In step 3, the semantic trajectory similarity between the riding trajectories is obtained by the following formula: in: sim_ST(ep i ,ep k ) represents any two cycling trajectories ep i ep k The similarity of the semantic trajectories; and represents the i-th and k-th semantic tracks, m and n refer to the number of semantic track segments in these two semantic tracks respectively; Represents the semantic track ST i Semantic track fragments in and semantic track ST k Semantic track fragments in similarity; Represents the semantic track ST i Each semantic track segment Instead of being in the semantic track ST k The most similar semantic trajectory fragments The sum of similarities; Represents the semantic track ST k Each semantic track segment Instead of being in the semantic track ST i The most similar semantic trajectory fragments The sum of similarities; From two semantic tracks ST i 、ST k The similarity of two semantic trajectory segments is obtained by the following formula: Among them, w P For each attribute's weight, p=5; matching function match p (A p ) are the corresponding attributes A of any two semantic trajectory segments P The similarity between the behavior attributes of the two semantic trajectory segments is calculated by the semantic matching function, and the similarity between the motion attributes of the two semantic trajectory segments is calculated by the distance matching function.

6. The method for classifying sports groups based on trajectory semantic similarity according to claim 5, characterized in that: The weight of each attribute is set to w1 = 0.063, w2 = 0.098, w3 = 0.372, w4 = 0.103, and w5 = 0.

364.

7. The method for classifying sports groups based on trajectory semantic similarity according to claim 5, characterized in that: The semantic matching function is obtained by the following formula: in, Is to get attribute A P The function of the category to which it belongs, The equal sign in the slash indicates that any two semantic track segments have the same attribute A. P Belong to the same category.

8. The method for classifying sports groups based on trajectory semantic similarity according to claim 5, characterized in that: The distance matching function is obtained by the following formula: in, and Attribute A P The distance function and distance threshold are as follows: the distance function adopts d(X,Y)=|XY|, where X and Y represent the motion attribute values ​​corresponding to any two semantic trajectory segments.

9. The method for classifying sports groups based on trajectory semantic similarity according to claim 5, characterized in that: In step 4, in the similarity network construction, K=10 in the KNN algorithm.

10. A motion behavior pattern recognition method based on trajectory semantic similarity, characterized in that: The method comprises: merging groups with the same cycling travel mode among multiple cycling sports groups to obtain a pattern classification result of the cycling sports groups, wherein the cycling sports groups are obtained by the sports group classification method based on trajectory semantic similarity described in any one of claims 1 to 8.

Images