A spatiotemporal profiling analysis method for people and places based on perception data

Abstract

The present invention relates to the field of portrait analysis technology, and discloses a method for spatiotemporal portrait analysis of people and places based on perception data, including the following steps: aggregating the perception data captured by cameras and access control sensing devices; preprocessing the captured perception data; extracting the behavioral features in the portraits and vehicle trajectory data in the perception data and classifying and summarizing them to establish a relevant database; setting the association relationship between people, vehicles, and places of different dimensions within a period according to the extracted data; aggregating the association relationship between each person and place; and constructing the spatiotemporal portrait of business-related people and places. The spatiotemporal portrait analysis method for people and places based on perception data proposed by the present invention can extract complete spatiotemporal trajectories and form continuous and complete spatiotemporal information, which can more accurately reflect the behavior patterns and activity laws of people, provide more in-depth behavior analysis capabilities, and improve the accuracy and comprehensiveness of relationship portraits.

Description

Technical Field

[0001] The present invention relates to the technical field of portrait analysis, and in particular to a method for spatiotemporal portrait analysis of people and places based on perception data. Background Art

[0002] Perception data is data perceived, measured and transmitted by perception devices or sensor devices. With the rapid development of smart city technology and the improvement of social security requirements, perception devices such as cameras and access control are spread throughout the streets and alleys of the city, and video surveillance basically covers public areas, providing hardware foundation and data support for the spatiotemporal portrait analysis of people and places. By drawing spatiotemporal portraits of people and places, relevant departments can have a clearer understanding of urban people and places, improve the city's risk event prevention and control and security dispatch capabilities, and ensure urban order and safety.

[0003] Although existing technologies can identify people in videos and draw their trajectories through algorithms, they lack effective organization and aggregation of the spatiotemporal features of these data. Specifically, existing systems usually only record the location of people at a certain point in time, but do not combine these location information with time information to form a complete spatiotemporal trajectory, resulting in an inability to fully understand the activity patterns and behavioral patterns of people. In addition, existing technologies fail to fully consider the overall characteristics of the groups of people coming and going in a place, such as the flow of people, crowd density, and the direction of crowd flow, and ignore the importance of group behavior analysis. In particular, when evaluating the connection between people, the existing technology based on simple statistics of trajectory overlap cannot accurately evaluate the relationship between people and the crowd portrait of the place;

[0004] Therefore, existing perceptual data analysis methods have obvious deficiencies in the integration of spatiotemporal features, group behavior analysis, and the refinement of multi-dimensional correlation levels, which limits the in-depth understanding and application of the common characteristics of urban people and places.

[0005] Therefore, it is necessary to design a spatiotemporal portrait analysis method for people and places based on perception data to solve the above problems. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method for spatiotemporal profiling of people and places based on perception data.

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

[0008] A method for analyzing the spatiotemporal portraits of people and places based on perception data includes the following steps:

[0009] S1: aggregates the perception data captured by cameras and access control sensing devices;

[0010] S2: preprocessing the captured sensory data;

[0011] S3: Extract the behavioral features of people in the perception data and vehicle trajectory data, classify and summarize them, and establish a relevant database;

[0012] S4: setting the associations between people, vehicles, and places in different dimensions within the period based on the extracted data;

[0013] S5: Aggregate the relationships between people and places;

[0014] S6: Build a spatiotemporal portrait of business-related people and places.

[0015] As a preferred technical solution of the present invention, in step S2, the preprocessing includes: data deduplication and redundancy processing: using hash functions to detect and remove duplicate items to reduce storage space; data missing value and outlier processing: filling missing values, removing and correcting outliers; data formatting and standardization: standardizing data from different sensors and systems into a unified data type and format; data conversion and processing: converting and processing data according to different business needs; data quality and reliability verification: after data cleaning and preprocessing, verifying and evaluating the quality and reliability of the data.

[0016] As a preferred technical solution of the present invention, the database in step S3 includes:

[0017] Place library: Based on the geographical location of the place and the longitude and latitude data of the sensing device, an association library of the place and the sensing device is generated and named as the place library. The association table of the target place obtained in the place library is the first place table, and the association tables other than the first place table in the place library are the second place table; Personnel trajectory library: Based on the trajectory information of the portrait, it is summarized into a personnel trajectory library, and the first trajectory table of the target person is obtained in the personnel trajectory library, and the trajectory tables other than the first trajectory table in the personnel trajectory library are obtained as the second trajectory table; Vehicle trajectory library: Based on the vehicle trajectory information, it is summarized into a vehicle trajectory library.

[0018] As a preferred technical solution of the present invention, the spatiotemporal portrait in step S6 includes: relationship portrait: determining the relationship portrait of the target person and the person corresponding to the second trajectory table according to the degree of association and the meeting place between the target person and the person corresponding to the second trajectory table; trajectory portrait: determining the trajectory portrait of the target person according to the number of overlaps and the overlap time between the trajectory of the target person and the places in the place library; portrait group portrait of the target place: determining the portrait group portrait of the target place according to the overlap between the target place and the person trajectory in the person trajectory library; vehicle group portrait of the target place: determining the vehicle group portrait of the target place according to the overlap between the target place and the vehicle trajectory in the vehicle trajectory library.

[0019] As a preferred technical solution of the present invention, the mechanism for obtaining the relationship portrait is as follows: 1 , t 2 ] time period, according to the number of trajectories of the target person u and the person V={v1, v2, v3…vq} in the second trajectory table that overlap within a time range less than the threshold Δt, C uV (t 1 , t 2 ), determine the degree of association A between the target person u and the person V corresponding to the second trajectory table uV (t 1 , t 2 ), where q represents the total number of people in the second trajectory table; in [t 1 , t 2 ] time period, the target person u and the person V in the second trajectory table overlap within a time range less than the threshold Δt, and the location M where the target person meets the corresponding person in the second trajectory table is determined uV (t 1 , t 2 );The target person u and the person V in the second trajectory table are in the time period [t 1 , t 2 ]The relationship portrait within is: R uV (t 1 , t 2 )=(A uV (t 1 , t 2 ), M uV (t 1 , t 2 )).

[0020] As a preferred technical solution of the present invention:

[0021] Said ;

[0022] Said ;

[0023] Said ;

[0024] Where I is the indicator function, L u (t) represents the location of the target person u at time t, L V (t) represents the position of person V at time t in the second trajectory table. u (t) = L V (t) is 1, otherwise it is 0, t∈[t 1 , t 2 ], F represents the comprehensive score of all factors affecting the degree of association.

[0025] As a preferred technical solution of the present invention, F represents the comprehensive score of all the factors affecting the degree of association:

[0026] Factors influencing the degree of association include the privacy of the meeting place, D;

[0027] The calculation mechanism of the impact score of the privacy D of the meeting place is as follows: the places in the place library P are divided into different privacy levels, P={p 1 , p 2 …p k}; Set a score for each level, S(p 1 ), S(p 2 )…S(p k ); for each meeting point m∈M uV (t 1 , t 2 ) Use spatial indexing technology to find the nearest place p in the place library P m , and obtain the score S(p m ); calculate [t 1 , t 2 ]The comprehensive impact score of the privacy D of all encounter locations within the time period:

[0028] ,

[0029] Among them, |M uV (t 1 , t 2 )| represents the number of meeting points, P j Indicates M uV (t 1 , t 2 ) in the location database corresponding to the j-th meeting point, S(P j ) represents the grade score of the place corresponding to the j-th meeting place.

[0030] As a preferred technical solution of the present invention, the factors affecting the degree of association include the encounter time T;

[0031] The impact score calculation mechanism of the encounter time T is as follows: set the relationship function between the score and time:

[0032] ,

[0033] where t X represents the Xth encounter time, and t X ∈[t 1 , t 2 ], A is the amplitude of the sine function, which determines half of the maximum fluctuation range of the score, is the phase shift, which is used to adjust the symmetry axis of the sine function to control the time corresponding to the maximum value. B is the offset, which determines the average value of the score. C is used to control the strength of the additional weighted term. 0 represents the peak position of the additional weighted term, and σ is used to control the influence range of the additional weighted term;

[0034] Calculate the comprehensive score of all encounter times T:

[0035] ,

[0036] Among them, α is a smoothing factor between 0 and 1. The closer it is to 1, the more weight is given to the most recent encounter time, and the closer it is to 0, the more uniform the weight is given to all encounter times. y is the total number of encounter times.

[0037] As a preferred technical solution of the present invention, the portrait group of people in the target place includes: crowd flow portrait: determining the crowd flow portrait according to the number of people whose trajectories overlap; population portrait: determining the population portrait according to the group information of people whose trajectories overlap.

[0038] As a preferred technical solution of the present invention, the vehicle group portrait of the target location includes: traffic flow portrait: determining the traffic flow portrait according to the number of vehicles with overlapping vehicle trajectories; vehicle portrait: determining the vehicle portrait according to the vehicle group information with overlapping vehicle trajectories.

[0039] The present invention has the following beneficial effects:

[0040] 1. Construction of complete spatiotemporal trajectory: By aggregating the data captured by various sensing devices, the complete spatiotemporal trajectory is extracted to form continuous and complete spatiotemporal information, which can more accurately reflect the behavior patterns and activity patterns of personnel and provide more in-depth behavior analysis capabilities, rather than just recording the location at a certain point in time;

[0041] 2. Establishment of group behavior model: By building a venue library, a personnel trajectory library, and a vehicle trajectory library, it is not only possible to track the behavior of a single individual, but also to analyze the group behavior characteristics in the venue, such as the flow of people, crowd density, and flow direction. This helps to identify abnormal behavior patterns, improve the level of urban safety management, and promptly discover and respond to potential safety hazards, making up for the deficiency of ignoring group behavior analysis in existing technologies;

[0042] 3. Multi-dimensional correlation evaluation: When determining the correlation between people, multi-dimensional influencing factors such as the privacy of the meeting place, the meeting time, and the weather are introduced to make the correlation evaluation more accurate and comprehensive. In addition, according to real-time data and environmental changes, the weight of each influencing factor is dynamically adjusted to improve the flexibility and accuracy of the evaluation, solving the problem that the existing technology cannot be fully evaluated based on simple statistical methods;

[0043] 4. Optimized urban management and decision-making support: By providing detailed and accurate spatiotemporal portraits and group behavior analysis results, this application provides a solid data foundation for urban management and decision-making, which helps to optimize resource allocation, improve urban operation efficiency and service levels, enable urban managers to make more scientific and reasonable decisions, and further improve the level of intelligent urban management. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of a method for analyzing the spatiotemporal portraits of people and places based on perception data proposed by the present invention;

[0045] Figure 2 A flow chart to depict space and time;

[0046] Figure 3 Flowchart derived for relationship profiling. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0048] Reference Figure 1-3 , a method for analyzing the spatiotemporal portraits of people and places based on perception data, comprising the following steps:

[0049] S1: aggregates the perception data captured by cameras and access control sensing devices;

[0050] S2: Preprocessing the captured sensory data. Specifically, the preprocessing includes:

[0051] Data deduplication and redundancy processing: Use hash functions to detect and remove duplicates to reduce storage space.

[0052] Data missing value and outlier processing: fill missing values, remove and correct outliers;

[0053] Data formatting and standardization: standardizing data from different sensors and systems into uniform data types and formats;

[0054] Data conversion and processing: convert and process data according to different business needs;

[0055] Data quality and reliability verification: After data cleaning and preprocessing, the quality and reliability of the data are verified and evaluated;

[0056] Intelligent anomaly detection: including anomaly detection based on machine learning: using unsupervised learning algorithms such as IsolationForest and Autoencoders to automatically detect outliers in data; also including rule-based anomaly detection: combining domain knowledge to define some rules to detect anomalies. For example, if a person frequently appears in multiple different locations in a short period of time, this may be an abnormal behavior;

[0057] S3: Extract the behavioral features of people in the perception data and vehicle trajectory data and classify and summarize them. The complete spatiotemporal trajectory is extracted using the trajectory comparison algorithm to form continuous and complete spatiotemporal information, and a related database is established. The database includes:

[0058] Place library: Generate a place-sensing device association library based on the place's geographical location and the longitude and latitude data of the sensing device and name it the place library. Obtain the association table of the target place in the place library as the first place table, and obtain the association table other than the first place table in the place library as the second place table;

[0059] Personnel trajectory library: The trajectory information of the portrait is summarized into a personnel trajectory library, a first trajectory table of the target person is obtained from the personnel trajectory library, and trajectory tables other than the first trajectory table in the personnel trajectory library are obtained as second trajectory tables;

[0060] Vehicle trajectory library: The vehicle trajectory information is aggregated into a vehicle trajectory library.

[0061] S4: Define the associations between people, vehicles, and places in different dimensions within the cycle based on the extracted data;

[0062] S5: Aggregate the relationships between people and places;

[0063] S6: Construct a spatiotemporal profile of business-related personnel and places, which includes:

[0064] Relationship portrait: Determine the relationship portrait between the target person and the person corresponding to the second trajectory table based on the degree of association between the target person and the person corresponding to the second trajectory table and the meeting place;

[0065] Trajectory profile: Determine the trajectory profile of the target person based on the number of overlaps and overlap times between the target person's trajectory and the locations in the location database;

[0066] The portrait group portrait of the target place: the portrait group portrait of the target place is determined according to the overlap of the target place and the personnel trajectory in the personnel trajectory library. Further, the portrait group portrait of the target place includes: the flow of people portrait: the flow of people portrait is determined according to the number of people whose personnel trajectories overlap; the population portrait: the population portrait is determined according to the information of the group of people whose personnel trajectories overlap;

[0067] Vehicle group portrait of the target location: The vehicle group portrait of the target location is determined based on the overlap of the vehicle trajectories of the target location and the vehicle trajectory library. Furthermore, the vehicle group portrait of the target location includes: traffic flow portrait: The traffic flow portrait is determined based on the number of vehicles with overlapping vehicle trajectories; vehicle portrait: The vehicle portrait is determined based on the vehicle group information with overlapping vehicle trajectories.

[0068] Furthermore, in step S6, the relationship portrait is obtained by the following mechanism:

[0069] In [t 1 , t 2 ] time period, according to the number of trajectories of the target person u and the person V={v1, v2, v3…vq} in the second trajectory table that overlap within a time range less than the threshold Δt, C uV (t 1 , t 2 ), determine the correlation degree A between the target person u and the person V corresponding to the second trajectory table uV (t 1 , t 2 ), where q represents the total number of people in the second trajectory table;

[0070] In the time period [t1, t2], the location information corresponding to the trajectory overlap between the target person u and the person V in the second trajectory table within a time range less than the threshold Δt is determined to determine the location M where the target person meets the corresponding person in the second trajectory table uV (t 1 , t 2 );

[0071] Then the relationship profile between the target person u and the person V in the second trajectory table in the time period [t1, t2] is: R uV (t 1 , t 2 )=(A uV (t 1 , t 2 ), M uV (t 1 , t 2 )).

[0072] Furthermore, the number of times the above trajectories overlap is:

[0073] ;

[0074] Where I is the indicator function, L u (t) represents the location of the target person u at time t, L V (t) represents the position of person V at time t in the second trajectory table. u (t) = L V (t) is 1, otherwise it is 0, t∈[t1 , t 2 ];

[0075] The meeting place is:

[0076] ;

[0077] The degree of association is: ;

[0078] Among them, F represents the comprehensive score of all factors affecting the degree of association. By setting this parameter, the accuracy and comprehensiveness of the degree of association can be improved.

[0079] Furthermore, the factors affecting the degree of association include the privacy D of the meeting place, and the calculation mechanism of the impact score of the privacy D of the meeting place is as follows:

[0080] Different privacy levels are assigned to the places in the place library P, P = {p 1 , p 2 …p k}, set a score for each level, S(p 1 ), S(p 2 )…S(p k ), all scores are ∈[H 1 , H 2 ] Specifically, according to existing research and standards, the privacy of places is divided into several levels, namely, very public: such as parks, squares, etc.; relatively public: such as cafes, shopping malls, etc.; moderately private: such as offices, libraries, etc.; relatively private: such as living rooms of private homes, hotel lobbies, etc.; very private: such as bedrooms of private homes, hotel rooms, etc., and scores are set for these levels in turn, for example, 1, 3, 5, 7, 10;

[0081] For each meeting point, use spatial indexing technology to find the nearest place p in the place library P. m , and obtain the score S(p m ) and used as the score for that encounter location;

[0082] Calculate [t 1 , t 2 ]The comprehensive impact score of the privacy D of all encounter locations within the time period:

[0083] ;

[0084] Among them, |M uV (t 1 , t 2 )| represents the number of meeting points, P j Indicates M uV (t 1, t 2 ) in the location database corresponding to the j-th meeting point, S(P j ) represents the grade score of the place corresponding to the j-th meeting place.

[0085] Furthermore, the influencing factors of the association degree include the encounter time T, and the calculation mechanism of the impact score of the encounter time T is as follows:

[0086] Set the relationship function between score and time:

[0087] ,

[0088] where t X represents the Xth encounter time, and t X ∈[t 1 , t 2 ], A is the amplitude of the sine function, which determines half of the maximum fluctuation range of the score, is the phase shift, which is used to adjust the symmetry axis of the sine function to control the time corresponding to the maximum value. B is the offset, which determines the average value of the score. C is used to control the strength of the additional weighted term. 0 represents the peak position of the additional weighted term, and σ is used to control the influence range of the additional weighted term. For example, setting A=4.5, B=5.5, =3π / 2, C=2, t 0 =0, σ=3, it means that the score is highest in the midnight (00:00 to 06:00) time period, and the score is lowest during the day, especially from noon to dusk. It should be noted that when integrating multiple influencing factors to calculate the degree of association, the value of the parameter in the formula needs to be set so that S T (t X )∈[H 1 , H 2 ], which can correspond to the range of score values ​​for the privacy of the meeting place;

[0089] Calculate the scores of all encounter times T using the relationship function between score and time;

[0090] Calculate the comprehensive score of all encounter times T:

[0091] ,

[0092] Among them, S T (t X ) is the score of the Xth encounter time, α is a smoothing factor between 0 and 1, the closer it is to 1, the more weight is given to the most recent encounter time, and the closer it is to 0, the more uniform the weight is given to all encounter times, and y is the number of encounter times.

[0093] For example, there are three encounter time points and their corresponding scores: t A = 0 (midnight), score S(t A )=10.00; t B =6 (6 am), score S(t B )=3.55; t C =12 (12 noon), score S(t C )=1.00; set the smoothing factor α=0.5;

[0094] Calculate the weight for each time period:

[0095] For t 3 (Most recent time period): weight is α(1-α) 3-3 =0.5×(1-0.5) 0 =0.5;

[0096] For t 2 (Next most recent time period): weight is α(1-α) 3-2 =0.5×(1-0.5) 1 =0.25;

[0097] For t 1 (Further time period): weight is α(1-α) 3-1 =0.5×(1-0.5) 2 =0.125;

[0098] The comprehensive score of the factors affecting the encounter time is: T (M uV (t 1 , t 2 ))= 0.5×1.00+ 0.25×3.55+0.125×10.00=2.64.

[0099] This method is used to calculate the comprehensive score of the encounter time influencing factor T. As time goes by, the weight of earlier data points will gradually decrease, thereby reducing the impact of old data on current values. This helps the model better adapt to new trends and patterns, reflect the latest changes more quickly, smooth the data to a certain extent, and reduce data fluctuations caused by single outliers or noise. Embodiment 1

[0100] Factors affecting the degree of association include weather W;

[0101] The weather is divided into several levels, and a basic score is set for each level. For example, the basic score is between 1-10, sunny: 1; cloudy: 2; overcast: 3; light rain: 4; moderate rain: 5; heavy rain: 6; heavy rain: 7; snow: 8; blizzard: 9; extreme weather (such as typhoons, tornadoes, etc.): 10;

[0102] Set the relationship function between the encounter weather and the score, and calculate the final score of the encounter weather:

[0103] S W (t Z ) = ε·W(t Z )·S T (t Z ),

[0104] Where W(t Z ) represents the basic score corresponding to the Zth encounter weather, S T (t Z ) is the score of the Zth encounter time, ε is the adjustment factor. It is worth noting that when the score is calculated in combination with other influencing factors, the basic score ∈[H 1 , H 2 ], set the value of ε so that S W The value range of (t) can fall into [H 1 , H 2 ], so that the values ​​can be normalized and then unified with the score range of other influencing factors. Using this formula, combined with the two dimensions of time and weather, it can provide richer information and help better understand the reasons behind the data. For example, heavy rain in the middle of the night may have a greater impact on people than heavy rain in the daytime, because visibility is low at night, transportation is inconvenient, and people may be less willing to go out. By combining the time factor, the degree of association between the target person u and the person V in the second trajectory table can be more accurately evaluated.

[0105] By calling the weather API to obtain the weather conditions at the time of encounter, the scores of all encounter weather conditions are obtained according to the above formula;

[0106] Calculate the comprehensive score of weather influencing factor W:

[0107] ,

[0108] Among them, S W (t Z ) represents the final score of the Zth encounter weather. Embodiment 2

[0109] The influencing factors include the degree of change of user’s expression E;

[0110] Capture the user's facial expressions in real time through facial recognition sensors such as cameras. Use tools such as OpenCV or Dlib to detect key points on the face, such as eyes, mouth, eyebrows, etc. The positions and shapes of these key points can be used to infer the user's expressions.

[0111] Classify facial expressions: There are two classification methods: rule-based methods: define some rules to judge facial expressions based on the position and shape changes of key points. For example, if the corners of the mouth are raised and the eyes are curved, it can be judged as a smile; machine learning methods: train a classifier such as SVM, random forest, neural network, etc., use the labeled expression data set for training, and then classify new facial images; pre-trained models: use pre-trained expression recognition models, such as Fer2013, AffectNet, etc. These models have been trained on large-scale data sets and can be directly applied;

[0112] Calculate the expression change score: record the key point position at each time point, and calculate the change in the key point position between adjacent time points. For example, calculate the Euclidean distance or cosine similarity, and calculate the change in the key point position between adjacent time points. The greater the change, the higher the score.

[0113] The degree of expression change is graded and a basic score is set, such as no change: 0; slight change: 2; moderate change: 4; significant change: 7; extreme change: 10, thereby obtaining a basic score for the user's expression change during each encounter.

[0114] Set up a formula to calculate the final score for each expression change during an encounter:

[0115] ,

[0116] in, is the starting time of the first encounter, i.e., the time when the expression data is first collected, is the time after the second encounter, i.e. the final expression data collection time, and , ∈[t1,t2], λ is the adjustment factor, and the basic score range is [H 1 , H 2 ], the purpose is to normalize the final score so that the output result is between [H 1 , H 2 ], and then corresponds to the score range of other influencing factors, E u is the expression change score of the target person, E V is the expression change score of the person in the second trajectory table, E maxis the maximum value of the expression change score. This formula can fully reflect the degree of expression change of two people every time they meet, and when the degree of expression change of both people is high, the score will be higher.

[0117] Calculate the combined score of expression changes during all encounters:

[0118]

[0119] ;

[0120] in For the The weight factor of the expression change score of each encounter is γ, which is a power exponent used to adjust the nonlinearity of the score. β is a constant that controls the strength of the weight factor. Smax is the maximum value of the final score of expression changes in all encounters. By setting the weight factor, encounters in certain time periods such as meetings and activities can be given higher weights, thereby occupying a larger proportion in the final score. The weight factor can be dynamically adjusted according to actual data, making the model more flexible and adaptable. By introducing the power exponent γ, it helps to emphasize or weaken the influence of certain scores. For example, if γ>1, high scores will be further amplified; if γ<1, low scores will be relatively amplified.

[0121] Based on the above analysis, the comprehensive score of all influencing factors is calculated. Embodiment 3

[0122] The calculation formula for the comprehensive score of the factors affecting the degree of association is:

[0123] ;

[0124] Among them, i represents different influencing factors, n represents the number of influencing factors, and f i (t 1 , t 2 ) indicates that in [t 1 , t 2 ]The comprehensive impact score of the i-th influencing factor in the time period. Using this calculation method can reduce the impact of extreme values ​​on the results, make the contribution of each factor more balanced, and provide a more stable estimate. Embodiment 4

[0125] The calculation formula for the score of the correlation factor is:

[0126] ;

[0127] Among them, i represents different influencing factors, n represents the number of influencing factors, and w i represents the weight of the i-th influencing factor, fi (t 1 , t 2 ) means that in [t 1 , t 2 ] is the comprehensive impact score of the i-th influencing factor in the time period, where w i It can be determined through statistical methods such as principal component analysis (PCA), factor analysis, and regression analysis, or it can be adjusted using machine learning methods such as feature importance and optimization algorithms, which can highlight the importance of certain influencing factors.

[0128] Furthermore, the mechanism for obtaining the target person's trajectory profile is as follows:

[0129] Determine coincidence events: For each place P in the place library P O , check the location L of the target person u u (t) Whether in the time period [t 1 , t 2 ] falls within the geographical scope of the venue. If so, a coincidence event B is recorded. u，O (ts, te), where ts is the time of first entry and te is the time of last exit;

[0130] Calculate the number of overlaps: For each location P O , count the number of all coincidence events to calculate the number of coincidences N u (P O , t 1 , t 2 ):

[0131] N u (P O , t 1 , t 2 ) = ∑ B u,O 1,

[0132] Where ∑B u,O For all locations P O The sum of the coincidence events.

[0133] Calculate the overlap time: For each location P O , calculate the total coincidence time D by accumulating the time difference of all coincidence events u (P O , t 1 , t 2 );

[0134] ;

[0135] Constructing trajectory profiles:

[0136] G u(P O , t 1 , t 2 )={(P O ,(N u (P O , t 1 , t 2 ), D u (P O , t 1 , t 2 ))),P O ∈P}.

[0137] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for analyzing the spatiotemporal portraits of people and places based on perception data, characterized in that: The following steps are involved: S1: aggregates the perception data captured by cameras and access control sensing devices; S2: preprocessing the captured sensory data; S3: Extract the behavioral features of people in the perception data and vehicle trajectory data, classify and summarize them, and establish a relevant database; S4: setting the associations between people, vehicles, and places in different dimensions within the period based on the extracted data; S5: Aggregate the relationships between people and places; S6: Build a spatiotemporal portrait of business-related people and places; The database in step S3 includes: Place library: Generate a place-sensing device association library based on the place's geographical location and the longitude and latitude data of the sensing device and name it the place library. Obtain an association table of the target place in the place library as the first place table, and obtain an association table other than the first place table in the place library as the second place table; Personnel trajectory library: according to the trajectory information of the portrait, a personnel trajectory library is aggregated, a first trajectory table of the target person is obtained from the personnel trajectory library, and a trajectory table other than the first trajectory table in the personnel trajectory library is obtained as a second trajectory table; Vehicle trajectory library: The vehicle trajectory information is aggregated into a vehicle trajectory library; The spatiotemporal portrait in step S6 includes: Relationship portrait: Determine the relationship portrait between the target person and the person corresponding to the second trajectory table according to the degree of association and the meeting place between the target person and the person corresponding to the second trajectory table; Trajectory profile: Determine the trajectory profile of the target person according to the number of overlaps and overlap times between the trajectory of the target person and the places in the place library; Portrait group portrait of the target location: Determine the portrait group portrait of the target location based on the overlap between the target location and the trajectory of the people in the personnel trajectory database; Vehicle group portrait at the target location: Determine the vehicle group portrait at the target location based on the overlap between the target location and the vehicle trajectory in the vehicle trajectory library; The mechanism for obtaining the relationship portrait is as follows: In the time period [t1, t2], the number of times C that the trajectory of the target person u overlaps with the person V={v1, v2, v3…vq} in the second trajectory table within a time range less than the threshold Δt is calculated. uV (t1, t2), determine the association degree A between the target person u and the person V corresponding to the second trajectory table uV (t1, t2), where q represents the total number of people in the second trajectory table; In the time period [t1, t2], the location information corresponding to the overlap of the trajectory of the target person u and the person V in the second trajectory table within a time range less than the threshold Δt is determined to determine the location M where the target person meets the corresponding person in the second trajectory table uV (t1, t2); The relationship profile between the target person u and the person V in the second trajectory table in the time period [t1, t2] is: R uV (t1, t2) = (A uV (t1, t2), M uV (t1, t2)).

2. According to claim 1, a method for analyzing the spatiotemporal portrait of people and places based on perception data is characterized in that: In step S2, the preprocessing includes: Data deduplication and redundancy processing: Use hash functions to detect and remove duplicate items to reduce storage space; Data missing value and outlier processing: fill missing values, remove and correct outliers; Data formatting and standardization: standardizing data from different sensors and systems into uniform data types and formats; Data conversion and processing: convert and process data according to different business needs; Data quality and reliability verification: After data cleaning and preprocessing, the quality and reliability of the data are verified and evaluated.

3. The method for analyzing spatiotemporal portraits of people and places based on perception data according to claim 1, characterized in that: Said ; Said ; Said ; Where I is the indicator function, L u (t) represents the location of the target person u at time t, L V (t) represents the position of person V at time t in the second trajectory table. u (t) = L V (t) is 1, otherwise it is 0, t∈[t1, t2], and F represents the comprehensive score of all the factors affecting the degree of association.

4. The method for analyzing the spatiotemporal portrait of people and places based on perception data according to claim 3 is characterized in that: The F represents the comprehensive score of all the factors affecting the degree of association: Factors influencing the degree of association include the privacy of the meeting place, D; The calculation mechanism of the impact score of the privacy D of the meeting place is as follows: Different privacy levels are assigned to the places in the place library P, P = {p1, p2…p k }; Set a score for each level, S(p1), S(p2)…S(p k ); For every meeting point m∈M uV (t1, t2) Use spatial indexing technology to find the nearest place p in the place library P m , and obtain the score S(p m ); Calculate the comprehensive impact score of the privacy D of all meeting locations in the time period [t1, t2]: ， Among them, |M uV (t1, t2)| represents the number of meeting points, P j Indicates M uV The place corresponding to the j-th meeting point in (t1, t2) in the place library, S(P j ) represents the grade score of the place corresponding to the j-th meeting place.

5. The method for analyzing spatiotemporal portraits of people and places based on perception data according to claim 3 is characterized by: The factors affecting the degree of association include the encounter time T; The impact score calculation mechanism of the encounter time T is as follows: Set the relationship function between score and time: ， where t X represents the Xth encounter time, and t X ∈[t1, t2], A is the amplitude of the sine function, which determines half of the maximum fluctuation range of the score, is the phase shift, which is used to adjust the symmetry axis of the sine function to control the time corresponding to the maximum value, B is the offset, which determines the average value of the score, C is used to control the strength of the additional weighted term, t0 represents the peak position of the additional weighted term, and σ is used to control the influence range of the additional weighted term; Calculate the comprehensive score of all encounter times T: ， Among them, α is a smoothing factor between 0 and 1. The closer it is to 1, the more weight is given to the most recent encounter time, and the closer it is to 0, the more uniform the weight is given to all encounter times. y is the total number of encounter times.

6. According to the method for analyzing the spatiotemporal portraits of people and places based on perception data according to claim 1, the portrait group portrait of the target place includes: Crowd flow profile: Determine the crowd flow profile based on the number of people whose trajectories overlap; Demographic portrait: Determine the demographic portrait based on the information of groups of people whose trajectories overlap.

7. The method for analyzing the spatiotemporal portrait of people and places based on perception data according to claim 1, characterized in that: The vehicle group portrait of the target location includes: Traffic flow profile: Determine the traffic flow profile based on the number of vehicles with overlapping vehicle trajectories; Vehicle portrait: Determine the vehicle portrait based on the information of the vehicle group with overlapping vehicle trajectories.

Images