An intelligent interactive method and system based on scenic spot video passenger flow trajectory analysis

Abstract

The present invention discloses an intelligent interaction method and system based on the analysis of the tourist flow trajectory of scenic spot videos, including S1: video data preprocessing, extracting key frames in the video stream; S2: constructing a tourist feature recognition network to identify tourist features; S3: constructing a tourist tracking network to track tourists across multiple cameras and restore the real-time location of tourists; S4: constructing a tourist heat map to visualize the movement trajectory and flow of tourists in real time; S5: combining the scenic spot's sound and light equipment to conduct personalized interaction based on the age group of tourists in the current area. The present invention solves the problems of inaccurate tracking of traditional tourist trajectory tracking methods, difficulty in judging real-time changes in flow of people, and lack of pertinence and rigidity in interaction with tourists.

Description

Technical Field

[0001] The present invention relates to the field of intelligent tourism service technology, and in particular to an intelligent interaction method and system based on scenic spot video passenger flow trajectory analysis. Background Art

[0002] Tourist flow trajectory analysis in scenic spots refers to the process of systematically analyzing the movement paths, stopover points and flow distribution of tourists in scenic spots. Traditional methods of analyzing tourists in scenic spots usually rely on questionnaires, manual counting or simple CCTV camera monitoring to collect data. Although these methods can provide some basic information, they also have obvious limitations.

[0003] Although traditional methods can know the entry and exit times of tourists, it is difficult to obtain their specific movement paths and stop points in the scenic area, and thus it is impossible to accurately track the trajectory of tourists; at the same time, the traditional scenic spot passenger flow trajectory analysis method is too rough in real-time traffic judgment and cannot accurately reflect the real-time changes and density of passenger flow; in terms of tourist feature analysis, traditional methods find it difficult to identify tourists' age groups, genders and other attributes, resulting in the inability to timely adjust the scenic spot's operating strategy and personalize the scenic spot's services.

[0004] In addition, traditional tourist interaction methods are usually static, such as fixed signboards and broadcasts, lacking personalization and dynamic interaction. These interactive methods fail to fully consider the specific needs and interests of different tourist groups, and do not use modern technology to enhance the tourist experience. Summary of the invention

[0005] In view of this, the present invention aims to provide an intelligent interaction method and system based on scenic spot video passenger flow trajectory analysis, aiming to solve the problems of inaccurate tracking of traditional tourist trajectory tracking methods, difficulty in judging real-time changes in passenger flow, and lack of pertinence and rigidity in interaction with tourists.

[0006] A method for analyzing scenic spot tourists based on video includes the following steps:

[0007] S1: Video data preprocessing, extracting key frames from the video stream;

[0008] S2: Build a tourist feature recognition network to identify tourist features;

[0009] S3: Build a tourist tracking network to track tourists across multiple cameras and restore their real-time locations;

[0010] S4: Construct a tourist heat map to visualize the movement trajectory and flow of tourists in real time;

[0011] S5: Combined with the scenic area's sound and light equipment, personalized interaction is carried out according to the age group of tourists in the current area;

[0012] Furthermore, the video data preprocessing step S1, extracting key frames in the video stream, includes:

[0013] S11: Video stream decoding, converting the original video data stream into an analyzable frame sequence;

[0014] S12: Setting a fixed time interval according to the time accuracy required for analysis;

[0015] S13: when the set time interval arrives, extract the current frame from the video stream as a key frame;

[0016] Compared with traditional video analysis methods, which consume a lot of computing resources to process continuous frames in the video stream, the present invention sets fixed time intervals for sampling according to the time accuracy requirements of the analysis, reduces the computing burden caused by too many redundant frames, and retains important information in the video while reducing the amount of data, thereby achieving the purpose of reducing data processing time and enabling the system to respond more quickly to changes in tourist behavior.

[0017] Furthermore, the step S2 constructs a tourist feature recognition network to recognize tourist features, including:

[0018] S21: Extract the feature representation of the key frame through the convolutional neural network and output the feature map;

[0019] S22: Input the feature map into the SuperPoint network, and extend the generated interest points into interest boxes. At the same time, according to the feature description vector of each interest point, the feature description vector of the interest box is calculated. The calculation method is:

[0020] p i ,d i =SuperPoint(Picture (1) )

[0021] Among them, p i is the point of interest, i is the point index, d i is the feature description vector of the point of interest, SuperPoint() is the SuperPoint network operation, Picture (1) is the feature map;

[0022] Points of interest are usually corner points in the image or points with obvious texture changes. They are generally distributed in the outline of the tourist, or the parts of the tourist with obvious features, such as clothing patterns, color boundaries, etc. For each tourist, the SuperPoint network will mark multiple points of interest on dozens or hundreds of pixels where they are located. The collection of these points of interest, together with their descriptors, constitute a comprehensive description of the tourist. This does not rely on a single feature point, but rather utilizes the information of multiple feature points on the tourist, which is conducive to subsequent tracking and feature analysis and improves the accuracy of tracking. This also means that even if part of the tourist's body is obscured or the posture changes, the system can still effectively track through other unobstructed points of interest, making the present invention more suitable for coping with complex scene changes in scenic spots;

[0023]

[0024] Among them, D j is the feature description vector of the interest box, j is the box index, and k is the side length of the interest box;

[0025] S23: Construct an age classification layer, and input each interest frame into the age classification layer to obtain the age category vector of each interest frame. The categories include: children, adults, elderly, and no category. The calculation method is:

[0026] o j =MLP(D j )

[0027] o′ j =softmax(o j )

[0028] Among them, j is the age category vector of the interest box, MLP() is the forward propagation operation, o′ j is the probability distribution of the age category of the interest box, and softmax() is the softmax operation;

[0029] Calculate o′ j Then, select o′ j The category with the highest probability is used as the age category of the interest frame;

[0030] Traditional feature recognition methods rely on simpler feature extraction techniques, such as color histogram or edge detection, which are not sufficient to accurately distinguish tourists of different age groups. Combining MLP and softmax processing, using the feature description vector of the interest frame to predict the age group can learn more complex feature representations and provide more accurate age group classification. At the same time, since the interest frame in step S22 is generated by self-supervised training, some scenic spot objects are also regarded as interest frames. At this time, age classification is performed through the age group classification layer, and the interest frame without age group information is automatically marked as "no category", which plays a role in filtering data and improving data quality.

[0031] S24: classify the age groups of tourists according to the age group category vector, and count the number of people in each age group category to obtain the flow of people at the location;

[0032] S25: Concatenate the age category vector and the feature description vector of the interest box to obtain the first tourist feature vector, which is calculated as follows:

[0033]

[0034] in, is the first tourist feature vector, Concat[,] is the concatenation operation;

[0035] By concatenating the age group category vector with the feature description vector of the interest frame, two types of information are combined: one is the feature description vector based on appearance, and the other is the classification result based on age group. This makes the feature representation of each tourist more comprehensive, providing richer information for subsequent tourist tracking across multiple cameras. At the same time, age group can also be used as a tourist feature, so that when controlling the scenic spot's audio-visual equipment, the system can provide more personalized services based on the tourist's age group and appearance characteristics.

[0036] Furthermore, the step S3 constructs a tourist tracking network to track tourists across multiple cameras and restore the real-time location of tourists, including:

[0037] S31: Using the tourist feature recognition network, the image of the camera at the next location is processed to obtain a second tourist feature vector, which is calculated as follows:

[0038]

[0039] in, is the second tourist feature vector, FeatureRec() is the tourist feature recognition network operation, Picture (2) is the feature map of the camera image at the next position;

[0040] S32: Construct a tourist tracking network, input the first tourist feature vector and the three first tourist feature vectors around the interest box into the tourist tracking relationship calculation layer, and obtain the first tourist relationship vector. The calculation method is:

[0041]

[0042] in, is the first tourist relationship vector, CNN() is the convolutional neural network operation, are the first tourist feature vectors corresponding to the three interest boxes closest to this interest box;

[0043] S33: input the second tourist feature vector and the three second tourist feature vectors around the interest box into the tourist tracking relationship calculation layer to obtain the second tourist relationship vector;

[0044]

[0045] in, is the second visitor relationship vector, are the second tourist feature vectors corresponding to the three interest boxes closest to this interest box;

[0046] S34: concatenate the first tourist feature vector with the first tourist relationship vector, concatenate the second tourist feature vector with the second tourist relationship vector, and input them into the Siamese network to calculate the vector similarity. The calculation method is:

[0047]

[0048] in, is the concatenation of the first tourist feature vector and the first tourist relationship vector, is the concatenation of the second tourist feature vector and the second tourist relationship vector, score is the similarity score output by the Siamese network, and Siamese() is the Siamese network operation;

[0049] In a scenic spot, a family or a group of people usually maintain a certain spatial relationship and interaction pattern. For example, parents follow their children closely and friends talk to each other. These people show a clustered spatial relationship in the image. By defining the features of the target interest box through the position features of the surrounding interest boxes, this spatial relationship can be better considered and the behavior patterns and characteristics of tourists can be better learned. The Siamese network is used to calculate the similarity between the feature vectors of different tourists, so as to identify and track the same group of people. In addition, in crowded scenic spots, a single tourist may be mixed with other tourists or part of the body may be blocked, which makes tracking more difficult. Considering the spatial relationship between tourists can improve the accuracy of tracking, even when the line of sight is obstructed.

[0050] S35: Associating the vector pairs with the highest similarity scores to obtain tourist trajectory information.

[0051] Furthermore, the step S4 constructs a tourist heat map to visualize the movement trajectory and flow of tourists in real time, including:

[0052] S41: Loading scenic area map;

[0053] S42: Identify the flow of people at each location output by the network according to the characteristics of the tourists, and load the real-time location data of the tourists;

[0054] S43: Calculate the density of the flow of people according to the area and flow of people at each scenic spot. The calculation method is:

[0055] ρ=S / N

[0056] Among them, ρ is the density of human traffic, S is the area of ​​the scenic spot, and N is the human traffic;

[0057] S44: Integrate heatmaps with real-time data and encode heatmaps using different colors;

[0058] S45: Overlay the generated heat map onto the scenic area map.

[0059] Furthermore, the step S5 combines the scenic spot's audio-visual equipment to conduct personalized interaction based on the age group of tourists in the current area, including:

[0060] S51: Calculate the age ratio of tourists at each scenic spot in real time according to the number of people in each age group output by the tourist feature recognition network;

[0061] S52: Establishing a scenic spot content library and matching scenic spot content based on predefined rules;

[0062] S53: Control the audio-visual equipment to play the matching scenic spot content.

[0063] Compared with the prior art, the advantages of the present invention are:

[0064] (1) In terms of the overall solution, the present invention uses video analysis methods to identify tourist features and track tourists across multiple cameras, and finally controls the scenic spot's audio-visual equipment in a targeted manner, which not only improves the tourists' interactive experience, but also enhances the scenic spot's attractiveness and operational efficiency.

[0065] (2) In terms of algorithm improvement, the present invention proposes to use the spatial features of the surrounding interest frames to represent the features of the target interest frame, and use age group information to enhance the interest frame features, thereby improving the accuracy of feature recognition and tracking, so that the system can more accurately identify and track tourists.

[0066] (3) In terms of robustness, the present invention uses SuperPoint to identify multiple features of tourists and combines them with the spatial relationship of people around them to simulate the situation where tourists appear in small groups in scenic spots, thereby improving the adaptability and robustness of the system in complex scenes. Even when part of the tourist's body is partially blocked or the posture changes, accurate tracking can continue. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A schematic flow chart of the intelligent interaction method based on scenic spot video passenger flow trajectory analysis provided by the present invention.

[0068] Figure 2 A schematic diagram of the algorithm flow of the tourist feature recognition network provided by the present invention. DETAILED DESCRIPTION

[0069] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0070] Embodiment 1: An intelligent interaction method based on scenic spot video passenger flow trajectory analysis, such as Figure 1 As shown, the following steps are included:

[0071] S1: Video data preprocessing, extracting key frames from the video stream; the video data can be obtained by directly pulling the video stream from the scenic spot camera device or video server through the network, or downloaded from the streaming media platform through an interface or other methods;

[0072] S11: Video stream decoding, converting the original video data stream into an analyzable frame sequence;

[0073] S12: Setting a fixed time interval according to the time accuracy required for analysis;

[0074] S13: when the set time interval arrives, extract the current frame from the video stream as a key frame;

[0075] Compared with traditional video analysis methods, which consume a lot of computing resources to process continuous frames in the video stream, the present invention sets fixed time intervals for sampling according to the time accuracy requirements of the analysis, reduces the computing burden caused by too many redundant frames, and retains important information in the video while reducing the amount of data, thereby achieving the purpose of reducing data processing time and enabling the system to respond more quickly to changes in tourist behavior.

[0076] For example, in a scenic spot scenario, since most tourists walk or take a tour bus and move slowly, the time accuracy requirement is not high. Therefore, it can be set to extract one frame every 2 seconds as a key frame, and extract 30 frames from the video stream every minute as key frames for analysis.

[0077] S2: Construct a tourist feature recognition network to identify tourist features, such as Figure 2 As shown;

[0078] S21: Extract the feature representation of the key frame through the convolutional neural network and output the feature map;

[0079] S22: Input the feature map into the SuperPoint network, and extend the generated interest points into interest boxes. At the same time, according to the feature description vector of each interest point, the feature description vector of the interest box is calculated. The calculation method is:

[0080] p i ,d i =SuperPoint(Picture (1) )

[0081] Among them, p i is the point of interest, i is the point index, d i is the feature description vector of the point of interest, SuperPoint() is the SuperPoint network operation, Picture (1) is the feature map;

[0082] Points of interest are usually corner points in the image or points with obvious texture changes. They are generally distributed in the outline of the tourist, or the parts of the tourist with obvious features, such as clothing patterns, color boundaries, etc. For each tourist, the SuperPoint network will mark multiple points of interest on dozens or hundreds of pixels where they are located. The collection of these points of interest, together with their descriptors, constitute a comprehensive description of the tourist. This does not rely on a single feature point, but rather utilizes the information of multiple feature points on the tourist, which is conducive to subsequent tracking and feature analysis and improves the accuracy of tracking. This also means that even if part of the tourist's body is obscured or the posture changes, the system can still effectively track through other unobstructed points of interest, making the present invention more suitable for coping with complex scene changes in scenic spots;

[0083]

[0084] Among them, D j is the feature description vector of the interest box, j is the box index, and k is the side length of the interest box;

[0085] S23: Construct an age classification layer, and input each interest frame into the age classification layer to obtain the age category vector of each interest frame. The categories include: children, adults, elderly, and no category. The calculation method is:

[0086] o j =MLP(D j )

[0087] o′ j =softmax(o j )

[0088] Among them, j is the age category vector of the interest box, MLP() is the forward propagation operation, o′ j is the probability distribution of the age category of the interest box, and softmax() is the softmax operation;

[0089] Calculate o′ j Then, select o′ j The category with the highest probability is used as the age category of the interest frame;

[0090] Traditional feature recognition methods rely on simpler feature extraction techniques, such as color histogram or edge detection, which are not sufficient to accurately distinguish tourists of different age groups. Combining MLP and softmax processing, using the feature description vector of the interest frame to predict the age group can learn more complex feature representations and provide more accurate age group classification. At the same time, since the interest frame in step S22 is generated by self-supervised training, some scenic spot objects are also regarded as interest frames. At this time, age classification is performed through the age group classification layer, and the interest frame without age group information is automatically marked as "no category", which plays a role in filtering data and improving data quality.

[0091] S24: classify the age groups of tourists according to the age group category vector, and count the number of people in each age group category to obtain the flow of people at the location;

[0092] S25: Concatenate the age category vector and the feature description vector of the interest box to obtain the first tourist feature vector, which is calculated as follows:

[0093]

[0094] in, is the first tourist feature vector, Concat[,] is the concatenation operation;

[0095] By concatenating the age group category vector with the feature description vector of the interest frame, two types of information are combined: one is the feature description vector based on appearance, and the other is the classification result based on age group. This makes the feature representation of each tourist more comprehensive, providing richer information for subsequent tourist tracking across multiple cameras. At the same time, age group can also be used as a tourist feature, so that when controlling the scenic spot's audio-visual equipment, the system can provide more personalized services based on the tourist's age group and appearance characteristics.

[0096] S3: Build a tourist tracking network to track tourists across multiple cameras and restore their real-time locations;

[0097] S31: Using the tourist feature recognition network, the image of the camera at the next location is processed to obtain a second tourist feature vector, which is calculated as follows:

[0098]

[0099] in, is the second tourist feature vector, FeatureRec() is the tourist feature recognition network operation, Picture (2) is the feature map of the camera image at the next position;

[0100] S32: Construct a tourist tracking network, input the first tourist feature vector and the three first tourist feature vectors around the interest box into the tourist tracking relationship calculation layer, and obtain the first tourist relationship vector. The calculation method is:

[0101]

[0102] in, is the first tourist relationship vector, CNN() is the convolutional neural network operation, are the first tourist feature vectors corresponding to the three interest boxes closest to this interest box;

[0103] S33: input the second tourist feature vector and the three second tourist feature vectors around the interest box into the tourist tracking relationship calculation layer to obtain the second tourist relationship vector;

[0104]

[0105] in, is the second visitor relationship vector, are the second tourist feature vectors corresponding to the three interest boxes closest to this interest box;

[0106] S34: concatenate the first tourist feature vector with the first tourist relationship vector, concatenate the second tourist feature vector with the second tourist relationship vector, and input them into the Siamese network to calculate the vector similarity. The calculation method is:

[0107]

[0108] in, is the concatenation of the first tourist feature vector and the first tourist relationship vector, is the concatenation of the second tourist feature vector and the second tourist relationship vector, score is the similarity score output by the Siamese network, and Siamese() is the Siamese network operation;

[0109] In a scenic spot, a family or a group of people usually maintain a certain spatial relationship and interaction pattern. For example, parents follow their children closely and friends talk to each other. These people show a clustered spatial relationship in the image. By defining the features of the target interest box through the position features of the surrounding interest boxes, this spatial relationship can be better considered and the behavior patterns and characteristics of tourists can be better learned. The Siamese network is used to calculate the similarity between the feature vectors of different tourists, so as to identify and track the same group of people. In addition, in crowded scenic spots, a single tourist may be mixed with other tourists or part of the body may be blocked, which makes tracking more difficult. Considering the spatial relationship between tourists can improve the accuracy of tracking, even when the line of sight is obstructed.

[0110] For example, tourist A appears in the first camera, surrounded by tourists B and C, as well as a passerby D;

[0111] After a while, three tourists A, B, and C appeared in the second camera, but B blocked half of A's body, and there was a passerby E around them;

[0112] At this point, it can be confirmed that the movement trajectory of ABC is from the scenic spot corresponding to the first camera to the scenic spot corresponding to the second camera. Since the three of them are traveling together, even if the passers-by around them change and tourist A is blocked, CNN can still track tourists A, B, and C because it extracts spatial relationship features.

[0113] S35: Associating the vector pairs with the highest similarity scores to obtain tourist trajectory information.

[0114] S4: Construct a tourist heat map to visualize the movement trajectory and flow of tourists in real time;

[0115] S41: Loading scenic area map;

[0116] S42: Identify the flow of people at each location output by the network according to the characteristics of the tourists, and load the real-time location data of the tourists;

[0117] S43: Calculate the density of the flow of people according to the area and flow of people at each scenic spot. The calculation method is:

[0118] ρ=S / N

[0119] Among them, ρ is the density of human traffic, S is the area of ​​the scenic spot, and N is the human traffic;

[0120] S44: Integrate heatmaps with real-time data and encode heatmaps using different colors; for example, red for high-density areas and blue for low-density areas;

[0121] S45: Overlay the generated heat map onto the scenic area map.

[0122] S5: Combined with the scenic area's sound and light equipment, personalized interaction is carried out according to the age group of tourists in the current area;

[0123] S51: Calculate the age ratio of tourists at each scenic spot in real time according to the number of people in each age group output by the tourist feature recognition network;

[0124] For example, when the proportion of children in the age group exceeds 30%, children-related content will be played first;

[0125] When the proportion of elderly visitors exceeds 20%, content suitable for the elderly will be given priority;

[0126] S52: Establishing a scenic spot content library and matching scenic spot content based on predefined rules;

[0127] For example, scenic spots in the forest area prepare forest animations. When it is detected that the proportion of children exceeds the threshold, the animations are played and interactive game prompts are provided, such as treasure hunt games, jigsaw puzzles, etc., using audio-visual equipment to provide clues and feedback for the games; when it is detected that the proportion of elderly people is relatively high, videos related to scenic spots such as operas, cultural promotional films, traditional handicrafts, and health knowledge lectures are played;

[0128] S53: Control the audio-visual equipment to play the matching scenic spot content.

[0129] Embodiment 2: The present invention also discloses an intelligent interaction system based on scenic spot video passenger flow trajectory analysis, which is used to implement the above-mentioned intelligent interaction method based on scenic spot video passenger flow trajectory analysis, including:

[0130] Video preprocessing module: video data preprocessing, extracting key frames from video streams;

[0131] Tourist feature recognition network construction module: construct a tourist feature recognition network to identify tourist features;

[0132] Tourist tracking network construction module: Build a tourist tracking network to track tourists across multiple cameras and restore tourists' real-time locations;

[0133] Tourist visualization module: constructs tourist heat maps to visualize tourist movement trajectories and flow in real time;

[0134] Equipment control module: Combined with the scenic area's audio-visual equipment, personalized interaction is carried out according to the age groups of tourists in the current area.

[0135] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0136] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0137] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An intelligent interactive method based on scenic spot video passenger flow trajectory analysis, characterized in that: The following steps are involved: S1: Perform video data preprocessing and extract key frames from the video stream; S2: Construct a tourist feature recognition network to identify tourist features; including: S21: Extract the feature representation of the key frame through the convolutional neural network and output the feature map; S22: Input the feature map into the SuperPoint network, and extend the generated interest points into interest boxes. At the same time, according to the feature description vector of each interest point, the feature description vector of the interest box is calculated. The calculation method is: p i ,d i =SuperPoint(Picture (1) ) Among them, p i is the point of interest, i is the point index, d i is the feature description vector of the point of interest, SuperPoint() is the SuperPoint network operation, Picture (1) is the feature map; Among them, D j is the feature description vector of the interest box, j is the box index, and k is the side length of the interest box; Through convolutional neural networks and SuperPoint networks, the age groups of tourists are classified and features are integrated, and the number of people in each age group is counted to obtain the flow of people at the location and the first tourist feature vector; S3: Build a tourist tracking network to track tourists across multiple cameras and restore their real-time locations; calculate the tourist relationship vectors in the two images and extract the spatial relationship between the target tourist and the people around the target tourist; secondly, input it into the Siamese network to calculate the similarity of the tourists in the two images and associate the vectors with the highest similarity scores to obtain the tourist trajectory information; including: S31: Using the tourist feature recognition network, the image of the camera at the next location is processed to obtain a second tourist feature vector, which is calculated as follows: in, is the second tourist feature vector, FeatureRec() is the tourist feature recognition network operation, Picture (2) is the feature map of the camera image at the next position; S32: Construct a tourist tracking network, input the first tourist feature vector and the three first tourist feature vectors around the interest box into the tourist tracking relationship calculation layer, and obtain the first tourist relationship vector. The calculation method is: in, is the first tourist relationship vector, CNN() is the convolutional neural network operation, is the first tourist feature vector, are the first tourist feature vectors corresponding to the three interest boxes closest to this interest box; S33: input the second tourist feature vector and the three second tourist feature vectors around the interest box into the tourist tracking relationship calculation layer to obtain the second tourist relationship vector; in, is the second visitor relationship vector, are the second tourist feature vectors corresponding to the three interest boxes closest to this interest box; S34: concatenate the first tourist feature vector with the first tourist relationship vector, concatenate the second tourist feature vector with the second tourist relationship vector, and input them into the Siamese network to calculate the vector similarity. The calculation method is: in, is the concatenation of the first tourist feature vector and the first tourist relationship vector, is the concatenation of the second tourist feature vector and the second tourist relationship vector, score is the similarity score output by the Siamese network, and Siamese() is the Siamese network operation; S35: Associating the vector pairs with the highest similarity scores to obtain tourist trajectory information; S4: Construct a tourist heat map to visualize the movement trajectory and flow of tourists in real time; S5: Combined with the scenic area’s audio-visual equipment, personalized interactions are conducted based on the age groups of tourists in the current area.

2. The intelligent interaction method based on scenic spot video passenger flow trajectory analysis according to claim 1 is characterized in that: The step S1 comprises the following steps: S11: Video stream decoding, converting the original video data stream into an analyzable frame sequence; S12: Setting a fixed time interval according to the time accuracy required for analysis; S13: When the set time interval arrives, extract the current frame from the video stream as a key frame.

3. The intelligent interaction method based on scenic spot video passenger flow trajectory analysis according to claim 2 is characterized in that: The step S2 comprises the following steps: S23: Construct an age classification layer, and input each interest frame into the age classification layer to obtain the age category vector of each interest frame. The categories include: children, adults, elderly, and no category. The calculation method is: O j =MLP(D j ) o′ j =softmax(o j ) Among them, j is the age category vector of the interest box, MLP() is the forward propagation operation, o′ j is the probability distribution of the age category of the interest box, and softmax() is the softmax operation; Calculate o′ j Then, select o′ j The category with the highest probability is used as the age category of the interest frame; S24: classify the age groups of tourists according to the age group category vector, and count the number of people in each age group category to obtain the flow of people at the location; S25: Concatenate the age category vector and the feature description vector of the interest box to obtain the first tourist feature vector, which is calculated as follows: Among them, Concat[,] is a concatenation operation.

4. The intelligent interaction method based on scenic spot video passenger flow trajectory analysis according to claim 1 is characterized in that: The step S4 comprises the following steps: S41: Loading scenic area map; S42: Identify the flow of people at each location output by the network according to the characteristics of the tourists, and load the real-time location data of the tourists; S43: Calculate the crowd density based on the area and crowd flow of each scenic spot. The calculation method is: p=S / N Among them, p is the density of human traffic, S is the area of ​​the scenic spot, and N is the number of human traffic; S44: Integrate heatmaps with real-time data and encode heatmaps using different colors; S45: Overlay the generated heat map onto the scenic area map.

5. The intelligent interaction method based on scenic spot video passenger flow trajectory analysis according to claim 4 is characterized in that: The step S5 comprises the following steps: S51: Calculate the age ratio of tourists at each scenic spot in real time according to the number of people in each age group output by the tourist feature recognition network; S52: Establishing a scenic spot content library and matching scenic spot content based on predefined rules; S53: Control the audio-visual equipment to play the matching scenic spot content.

6. An intelligent interactive system based on scenic spot video passenger flow trajectory analysis, characterized in that: include: Video preprocessing module: video data preprocessing, extracting key frames from video streams; Tourist feature recognition network construction module: construct a tourist feature recognition network to identify tourist features; Tourist tracking network construction module: Build a tourist tracking network to track tourists across multiple cameras and restore tourists' real-time locations; Tourist visualization module: constructs tourist heat maps to visualize tourist movement trajectories and flow in real time; Equipment control module: Combined with the scenic area's sound and light equipment, personalized interaction is carried out according to the age group of tourists in the current area; To realize the intelligent interaction method based on scenic spot video passenger flow trajectory analysis as described in any one of claims 1-5.

Images