Travel shooting short video generation system and method

Abstract

The invention discloses a travel shooting short video generation system and method, and relates to the technical field of short video generation, and the method comprises the following steps: deploying a high-speed camera with an inertial measurement unit in a dynamic scene of a scenic spot, deploying an environment light sensation adaptive camera in a static scene, and connecting all cameras with an edge computing gateway; the camera starts a corresponding shooting mode according to the scene type, the dynamic scene tracks the motion track through a multi-camera space-time synchronization algorithm, and the static scene adjusts shooting parameters according to light and the flow of people; the cloud AI engine extracts behavior features, emotional features and basic features of the user, screens an adaptive video template from a template library through a collaborative filtering algorithm in combination with scene features, and generates an exclusive short video; the system associates and binds the generated exclusive short video with the user feature data, and establishes a one-to-one or many-to-one short video affiliation relationship for the user to preview and download online; and inconvenience caused by the fact that a camera device is manually held by hand for shooting is avoided.

Description

Technical Field

[0001] This invention relates to the field of short video generation technology, specifically to a travel photography short video generation system and method. Background Technology

[0002] Tourist tourism has become a popular leisure activity, with various scenic spots serving as must-visit destinations. Furthermore, with the rise of short-video social media, capturing moments of enjoyment during travel has become a core need for tourists. Currently, tourists primarily obtain video content through selfies, assistance from companions, or hiring travel photographers. However, manual filming presents several inconveniences in high-speed or dangerous environments such as slides, roller coasters, and rafting. Firstly, tourists find it difficult to film independently, and handheld devices pose safety hazards. Secondly, shaky footage is common among ordinary tourists, affecting video quality. Thirdly, the videos require manual editing, which is time-consuming and contradicts the desire for a relaxed and enjoyable experience. Summary of the Invention

[0003] This invention provides a travel photography short video generation system and method to address the shortcomings of manual shooting in the prior art and improve the convenience of taking photos and videos at scenic spots.

[0004] This invention provides the following technical solution:

[0005] In a first aspect, the present invention provides a method for generating short travel videos, comprising the following steps:

[0006] S1. Deploy high-speed cameras with inertial measurement units in dynamic scenes of the scenic area, and deploy ambient light adaptive cameras in static scenes. All cameras are connected to the edge computing gateway.

[0007] S2. The camera activates the corresponding shooting mode according to the scene type. In dynamic scenes, the camera tracks the motion trajectory through a multi-camera spatiotemporal synchronization algorithm. In static scenes, the shooting parameters are adjusted according to the light and the flow of people.

[0008] S3, the cloud-based AI engine extracts the user's behavioral, emotional, and basic characteristics, and combines them with scene characteristics to select suitable video templates from the template library through a collaborative filtering algorithm to generate exclusive short videos;

[0009] S4. The system will associate and bind the generated exclusive short videos with user feature data to establish a one-to-one or many-to-one short video ownership relationship, so that users can preview and download them online.

[0010] As a further improvement of the present invention, in S2, the workflow of the dynamic scene shooting mode includes the following steps:

[0011] After the high-speed camera is started, a multi-camera communication link is established through the edge computing gateway. The multi-camera timestamp calibration is completed by using a dual synchronization mechanism of Network Time Protocol (NTP) and Precision Time Protocol (PTP).

[0012] Based on GPS positioning data and the scenic area's electronic map, a unified spatial coordinate system was established to determine the spatial coordinates of each camera position and complete the spatial reference calibration.

[0013] The high-speed camera collects the device's own motion posture data in real time, with the collection frequency consistent with the camera's frame rate, and binds the motion data and video frame data to the edge computing gateway;

[0014] The system detects video frames captured by each camera position and extracts the user's human body contour and key feature points to achieve cross-view locking of the same user from multiple camera positions.

[0015] The edge computing gateway fuses video frame data from multiple cameras with motion data from high-speed cameras, and uses a Kalman filter algorithm to eliminate noise in the high-speed camera data to obtain a smooth trajectory of device motion.

[0016] Based on the pixel coordinates of key user feature points and the spatial coordinates of the camera position, the three-dimensional position of the user in a unified spatial coordinate system is calculated by triangulation. Combined with the position data of continuous frames, the particle filter algorithm is used to predict the user's motion trajectory, realizing real-time tracking of the user's motion trajectory by multiple cameras.

[0017] As a further improvement of the present invention, the step of extracting the user's human body contour and key feature points to achieve cross-view locking of the same user from multiple camera angles includes the following steps:

[0018] Each camera position uses the YOLOv8 model to detect human targets in real-time video frames and outputs the bounding box coordinates of valid human targets;

[0019] Extract and compare the global human body contour features, key human body feature points, and clothing features of adjacent frames to confirm that the target in the current frame and the previous frame is the same user.

[0020] Based on the target features of the high-point camera position, the cosine similarity of the contour feature vectors of other camera positions and the weighted cosine similarity of the local feature vectors of the benchmark target are calculated. When the similarity meets the threshold, the global user ID is bound to achieve cross-view locking of the same user by multiple camera positions.

[0021] As a further improvement of the present invention, S3 specifically includes the following steps:

[0022] The cloud-based AI engine receives encrypted video clips uploaded by the edge computing gateway and performs structured extraction of user features;

[0023] Based on scene feature label matching templates in the same scene library, several candidate templates were initially selected.

[0024] Based on historical template selection data of users with similar scenarios and characteristics, the remaining templates are sorted, and the optimal template is selected.

[0025] Edit videos using the optimal template, matching core user behavior segments with template slow-motion or special effects nodes, and ordinary segments with regular editing nodes, and automatically synthesize short videos.

[0026] As a further improvement of the present invention, the step of structured extraction of user features is as follows:

[0027] Based on the human pose estimation model, key user actions are identified, the duration of each action is counted, and core behavioral segments are marked.

[0028] By analyzing key facial expressions of users through a facial expression recognition model, emotional labels such as pleasure and calmness are output.

[0029] As a further improvement of the present invention, in step S4, the user's download step is as follows:

[0030] Users can initiate an access request through at least one of the following channels: QR code signs set up in the scenic area, the official mini-program of the scenic area, or the self-service terminal in the visitor center;

[0031] After receiving the request, the system pops up a verification interface. The user enters the order number or ID card number, and the system compares the data with the scenic spot's ticketing system. If the match is successful, the verification is successful.

[0032] After verification, users enter the short video preview interface. After selecting the target resolution, the system immediately generates a short video of the corresponding resolution for users to download to their local devices or share to social media platforms.

[0033] As a further improvement of the present invention, step S4 also involves a secondary optimization step for the short video:

[0034] After users obtain short videos, they can initiate optimization requests through the official mini-program to change background music, adjust transition styles, add personalized text, change filters, and trim clips.

[0035] The cloud-based AI engine re-invokes template library resources or adjusts generation parameters based on user optimization needs to generate optimized short videos;

[0036] Users can preview the optimized short videos through the original identity verification channel and can initiate optimization requests again or download and share them directly.

[0037] Secondly, the present invention provides a travel photography short video generation system, comprising:

[0038] The scene camera deployment module deploys high-speed cameras with inertial measurement units in dynamic scenes of the scenic area, and ambient light-sensing adaptive cameras in static scenes. All cameras are connected to the edge computing gateway.

[0039] The scene-based shooting module allows the camera to activate the corresponding shooting mode based on the scene type. In dynamic scenes, the camera tracks the motion trajectory through a multi-camera spatiotemporal synchronization algorithm, while in static scenes, the shooting parameters are adjusted according to the light and the flow of people.

[0040] The dedicated video generation module uses a cloud-based AI engine to extract users' behavioral, emotional, and basic characteristics, and combines these with scene features to select suitable video templates from the template library through a collaborative filtering algorithm, generating a personalized short video.

[0041] The user preview and download module associates the generated exclusive short videos with user feature data, establishing a one-to-one or many-to-one short video ownership relationship, so that users can preview and download them online.

[0042] The present invention has the following beneficial effects:

[0043] By installing high-speed cameras with inertial measurement units in dynamic scenes within scenic areas, videos are captured and stored in the cloud. These videos are then processed by a cloud-based AI engine to generate personalized short videos, which users can download via mobile devices. This avoids the inconvenience of manually holding camera equipment. Furthermore, videos captured by high-speed cameras are shaky, ensuring high image quality. In addition, the cloud-based AI engine can automatically generate corresponding short videos, reducing the need for manual editing. Attached Figure Description

[0044] Figure 1 This is a diagram showing the overall architecture of the travel photography short video generation system in this invention.

[0045] Figure 2 This is a flowchart of the video template matching and optimized sorting process in this invention.

[0046] Figure 3 This is a schematic diagram illustrating the spatiotemporal synchronous fusion of dynamic scenes from multiple cameras in a travel photography short video generation system.

[0047] Figure 4 This is a flowchart of the travel photography short video generation method in this invention. Detailed Implementation

[0048] The technical solutions of the embodiments of this specification will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of this specification and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of this specification.

[0049] The terms "first," "second," "third," etc., in the description, claims, and accompanying drawings are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0050] In the following description, terms such as “inner,” “outer,” “upper,” “lower,” “left,” and “right” are used only to facilitate the description of the embodiments and to simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this specification.

[0051] All data involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0052] Before introducing the technical solutions described in this manual, the application scenarios and related technologies of the technical solutions will be introduced.

[0053] Scenic spot tourism has become a mainstream leisure activity, with major attractions being must-visit destinations for tourists. With the widespread use of short video social media, recording moments during travel has become a core need for tourists. However, tourists mainly rely on themselves or their companions to shoot videos, which presents many inconveniences in high-speed or dangerous scenarios such as slides, roller coasters, and rafting. To address this, this invention provides a method for generating short travel videos. By installing high-speed cameras with inertial measurement units in dynamic scenes within scenic areas, the videos are captured and downloaded via mobile devices, avoiding the inconvenience of manually holding the camera. Furthermore, the videos captured by the high-speed cameras are shaky, ensuring high image quality. In addition, a cloud-based AI engine can automatically generate the corresponding short videos, reducing the need for manual editing.

[0054] Please see Figure 1-4 As shown, a method for generating travel-themed short videos includes the following steps:

[0055] S1. Deploy high-speed cameras with inertial measurement units in dynamic scenes of the scenic area, and deploy ambient light adaptive cameras in static scenes. All cameras are connected to the edge computing gateway.

[0056] High-speed industrial cameras equipped with 9-axis inertial measurement units are deployed in dynamic scenic scenes such as roller coasters, glass walkways, rafting rivers, and zip lines. The cameras have a frame rate of ≥120fps, a resolution of ≥4K, and a shutter speed of ≤1 / 1000s, and support real-time tracking of dynamic targets.

[0057] In static scenes within scenic areas, such as landmark viewing platforms, folk experience points, historical sites, and viewing corridors, ambient light adaptive cameras are deployed. These cameras integrate a CMOS light sensor and a people flow statistics module, with a light intensity detection range of 0-20000 lux and support for automatic exposure adjustment.

[0058] All cameras establish communication connections with the edge computing gateway via a 5G industrial module or a gigabit wired network. The edge computing gateway is equipped with an 8-core processor, 16GB of memory, and 1TB of local cache, and is pre-loaded with a lightweight basic feature recognition model.

[0059] After the camera is turned on, it automatically completes self-calibration, including lens distortion correction, high-speed camera posture calibration, and timestamp synchronization. It identifies basic information tags such as user age, gender, and clothing style in real time through the captured images. Age is divided into teenagers: 6-17 years old, young adults: 18-35 years old, middle-aged: 36-55 years old, and elderly: 56 years old and above. Clothing styles include casual, sports, and ethnic styles.

[0060] S2. The camera activates the corresponding shooting mode according to the scene type. In dynamic scenes, the motion trajectory is tracked through a multi-camera spatiotemporal synchronization algorithm, while in static scenes, the shooting parameters are adjusted according to the light and the flow of people.

[0061] The workflow of the dynamic scene shooting mode includes the following steps:

[0062] After all high-speed cameras deployed in dynamic scenes are started, a multi-camera communication link is established through the edge computing gateway. The network time protocol NTP and the precision time protocol PTP are used for dual synchronization mechanism to complete the multi-camera timestamp calibration, so that the synchronization accuracy is ≤3ms.

[0063] For example, 10 minutes before the roller coaster starts, the system automatically wakes up all high-speed cameras in dynamic scenes. The cameras establish a dedicated UDP communication link through the edge computing gateway and adopt a dual synchronization mechanism of Network Time Protocol (NTP) and Precision Time Protocol (PTP) to calibrate the timestamps of each camera position. The final synchronization accuracy reaches 2.8ms, which meets the timing consistency requirements of multi-camera frame data.

[0064] Based on GPS positioning data and the scenic area's electronic map, a unified spatial coordinate system was established to determine the spatial coordinates (X, Y, Z axes) of each camera position, and spatial benchmark calibration was completed.

[0065] For example, based on the latitude, longitude, and altitude data collected by the built-in GPS modules of each camera, and combined with the scenic area's electronic map, a three-dimensional spatial coordinate system is established: the X-axis represents the east-west direction of the scenic area (the origin is the scenic area gate), the Y-axis represents the north-south direction of the scenic area, and the Z-axis represents the altitude (the origin is sea level). The spatial coordinates of 14 track cameras and 1 reference camera are calculated and stored through the gateway. For example, the coordinates of track camera position 5 are (320.5m, 186.3m, 720.8m), and the coordinates of the high-point reference camera position are (350.2m, 210.7m, 800.0m).

[0066] The 9-axis inertial measurement unit of the high-speed camera collects the device's own motion attitude data in real time, including three-axis acceleration, three-axis angular velocity, and three-axis magnetic field strength. The acquisition frequency is consistent with the camera's frame rate, and the motion data and video frame data are bound and transmitted to the edge computing gateway.

[0067] For example, after the roller coaster starts, each high-speed camera captures video at a frame rate of 120fps, while a 9-axis inertial measurement unit collects the device's own motion attitude data in real time (including three-axis acceleration: 0-5g, three-axis angular velocity: 0-2000° / s, and three-axis magnetic field strength: 0-80μT), with the acquisition frequency consistent with the frame rate (120Hz). Each frame of video data is bound to the motion attitude data with the corresponding timestamp and transmitted to the edge computing gateway in the form of encrypted data packets.

[0068] S2.4. Detect video frames captured by each camera position, and achieve cross-viewpoint locking of the same user from multiple camera positions by extracting the user's human body contour and key feature points; including the following steps:

[0069] Step 1: Each camera position uses the YOLOv8 model to perform human detection on real-time video frames, sets a confidence threshold of 0.85, filters out invalid targets such as staff members next to the track and tourists in the distance, and outputs the bounding box coordinates of tourists inside the roller coaster car (e.g., the bounding box coordinates of a tourist are x1=320px, y1=180px, x2=450px, y2=360px).

[0070] Step 2: Extract the 256-dimensional global contour feature vector, coordinates of 21 key feature points (head, shoulders, elbows, knees, etc.), and RGB values ​​of the main color tone of clothing (e.g., R=255, G=120, B=80) of the same tourist in two adjacent frames (1 / 120s interval). Verify that the feature point displacement deviation is ≤15 pixels through optical flow tracking algorithm. Combined with the clothing feature similarity ≥0.78, confirm that the current frame and the previous frame are the same tourist.

[0071] Step 3: Using the tourist characteristics of the benchmark camera position at the highest point as a baseline, calculate the cosine similarity (threshold ≥ 0.75) of the contour feature vectors of tourists at the other 14 track-mounted cameras and the benchmark tourist, as well as the weighted cosine similarity of the local feature vectors (weights of key feature points such as the head and shoulders ≥ 0.6, weights of feature points at the extremities ≤ 0.3, threshold ≥ 0.80). A tourist at track-mounted camera position 3 has a contour similarity of 0.82 and a local similarity of 0.86 with the benchmark camera position, both meeting the thresholds. The system then assigns a globally unique user ID (e.g., UID-20240520-0036) to this tourist, enabling cross-view locking of this tourist from all 15 camera positions.

[0072] The formula for calculating the cosine similarity of the contour feature vectors is: ;

[0073] A = [A1, A2, ... A n [B1, B2, ..., B] represents the contour feature vector of the first camera's video frame. n ] represents the contour feature vector of the corresponding video frame from the second camera position, and n represents the dimension of the contour feature vector (n≥256).

[0074] A i Let B be the i-th component of vector A. i Let i be the i-th component of vector B;

[0075] X takes the value [0,1]. When X≥0.75, the two contour features are considered to match.

[0076] The formula for calculating the weighted cosine similarity of local feature vectors is as follows: ;

[0077] C = [C1, C2, ... C n [D1, D2, ...D] represents the contour feature vector of the first camera's video frame. n ] represents the contour feature vector of the corresponding video frame from the second camera position, and m is the dimension of the contour feature vector (m≥128).

[0078] C j Let D be the j-th component of vector C. j Let j be the j-th component of vector D;

[0079] w j The weight of the j-th component (the weight of the component corresponding to key feature points such as the head and shoulders is ≥0.6, and the weight of the component corresponding to feature points at the extremities is ≤0.3), and satisfies the following conditions: ;

[0080] Y takes the value [0,1]. When X≥0.8, the two contour features are considered to match.

[0081] S2.5 The edge computing gateway fuses video frame data from multiple cameras with motion data from high-speed cameras, and uses a Kalman filter algorithm to eliminate noise in the high-speed camera data to obtain a smooth trajectory of device motion.

[0082] After receiving video frame data and motion data from 15 camera positions, the edge computing gateway uses the Kalman filter algorithm to process the device motion data: it estimates the device attitude at the next moment through the prediction equation, and then corrects the deviation by updating the equation and combining it with the measured data, eliminating noise caused by roller coaster vibration and wind interference, and obtaining a smooth trajectory of device motion.

[0083] S2.6 Based on the pixel coordinates of the user's key feature points and the spatial coordinates of the camera position, the three-dimensional position of the user in a unified spatial coordinate system is calculated by triangulation; combined with the position data of continuous frames, the particle filter algorithm is used to predict the user's motion trajectory, so as to realize the real-time tracking of the user's motion trajectory by multiple cameras.

[0084] Based on the pixel coordinates of key feature points of tourists and the spatial coordinates of corresponding camera positions, the real-time position of tourists in a unified three-dimensional coordinate system is calculated by triangulation (e.g., the position at a certain moment is (325.7m, 188.9m, 722.3m)). Combining the position data of 10 consecutive frames, a particle filtering algorithm (with the number of particles set to 800) is used to predict the movement trajectory of tourists at the next moment, guiding each camera position to dynamically adjust the shooting angle and focal length to ensure that tourists are always in the center area of ​​the image.

[0085] In addition, the ambient light adaptive camera for static scenes automatically adjusts the exposure time, ISO sensitivity, and white balance parameters based on real-time light intensity data collected by the CMOS light sensor. At the same time, it identifies the density of tourists in the current shooting area through the people flow statistics module. When the density is ≥5 people / 10㎡, it automatically switches to wide-angle shooting mode to ensure that the core scene and users are included in the picture. When the density is <5 people / 10㎡, it switches to close-up mode to focus on the interaction between users and scene landmarks, enhancing the sense of depth in the picture.

[0086] For example, the static scene is a cloud sea viewing platform, and the implementation steps are as follows:

[0087] After the camera is started, it automatically completes lens distortion correction and self-calibration. The CMOS light sensor collects ambient light intensity data in real time (e.g., the light intensity is 800 lux at 8 am and 16000 lux at 12 pm). The crowd counting module identifies the number and distribution density of tourists in the observation deck through the frame difference method.

[0088] Low-traffic scene (9:00, tourist density 3 people / 10㎡): light intensity detection is 1200 lux, camera automatically adjusts exposure time to 1 / 300s, ISO sensitivity to 200, white balance to "daylight mode"; due to low traffic density, switch to close-up shooting mode, adjust focal length to 50mm, focus on the interaction between tourists and the sea of ​​clouds, highlighting the integration of people and scene;

[0089] High-traffic scene (11:30, tourist density 6 people / 10㎡): light intensity detection is 15000 lux, camera automatically adjusts exposure time to 1 / 800s, ISO sensitivity to 100, and backlight compensation is enabled; due to the high traffic density, switch to wide-angle shooting mode, adjust focal length to 24mm to ensure that more tourists and the complete sea of ​​clouds are included in the picture, and at the same time optimize the composition of the picture through algorithms to avoid overlapping and obstruction of people.

[0090] During filming, the camera updates light intensity and crowd density data every 500ms, dynamically fine-tuning the shooting parameters: when sudden cloud cover causes the light intensity to drop from 15,000 lux to 8,000 lux, the system adjusts the exposure time to 1 / 500s within 300ms to maintain stable image brightness; all captured video frame data are transmitted to the edge computing gateway after lightweight preprocessing to provide materials for subsequent cloud AI synthesis.

[0091] S3, the cloud-based AI engine extracts the user's behavioral, emotional, and basic characteristics, and combines them with scene characteristics to select suitable video templates from the template library through a collaborative filtering algorithm to generate exclusive short videos;

[0092] S3.1 The cloud-based AI engine receives encrypted video clips uploaded by the edge computing gateway and performs structured extraction of user features: it identifies key user actions based on a human pose estimation model, counts the duration of each action, and marks core behavioral segments; it analyzes key facial expressions of the user through a facial expression recognition model and outputs emotion labels such as pleasure and calmness.

[0093] For example, the "Folk Culture Village - Bamboo Pole Dance Experience Area" of a 5A-level scenic spot is used as the implementation scenario. The core of this scenario is tourists' participation in the traditional bamboo pole dance.

[0094] Step 1: Data Reception and Decryption: The cloud-based AI engine receives video data packets uploaded by the edge computing gateway via an HTTPS encrypted channel. The data packets are encrypted using AES-256 and contain: a 90-second 4K resolution video clip, basic user characteristic tags (young woman, 28 years old, casual style), and scene tags (folk village - bamboo pole dance). The engine calls the decryption module to decrypt the data packets, ensuring the integrity of the video data.

[0095] Step 2: Behavioral Feature Extraction: Based on the OpenPose human pose estimation model, video segments are analyzed frame by frame to identify key user actions: stepping (lasting 8 seconds), jumping (3 times, each lasting 1.2-1.5 seconds), clapping (5 times, each lasting 0.5 seconds), and swaying the body to the rhythm (lasting 15 seconds); the duration of each action is counted, and actions with a duration ≥ 1 second are marked as core behavioral segments: 3 jumping segments (total 4.2 seconds), 15-second swaying body segment; ordinary behavioral segments: stepping, clapping (as transition segments).

[0096] The third step is emotion feature extraction: The user's facial region is located using the MTCNN face detection algorithm (occlusion rate <10%), and then the key facial features (angle of mouth, width of eye opening, and position of eyebrows) are analyzed using the FER emotion recognition model; the emotion labels are output as "pleasure" and "excitement", with "pleasure" being taken as the primary emotion label.

[0097] S3.2. Based on scene feature labels, match templates of the same scene in the template library to initially select several candidate templates;

[0098] For example, combining the scene tags uploaded by the edge computing gateway with the video content (bamboo poles, staff in ethnic costumes, traditional architectural background), the scene tag was ultimately determined to be "Folk Village - Bamboo Pole Dance - Interactive Participation"; based on the scene tag "Folk Village - Bamboo Pole Dance - Interactive Participation" and the main emotion tag "Joy", the system matched candidate templates from the template library:

[0099] Template A: "Joyful Folk Custom Interaction" (Scene matching 100%, Emotion matching 95%)

[0100] Template B "Vibrant Ethnic Style" (Scene matching 100%, Emotion matching 90%)

[0101] Template D: "Fun Interactive Moments" (Scene matching 90%, Emotion matching 85%).

[0102] S3.3. Based on historical template selection data of users with the same scenario and characteristics, sort the remaining templates and select the optimal template;

[0103] For example, a collaborative filtering algorithm is used for adaptation scoring, with the following scoring dimensions and weights:

[0104] Historical user preference with similar characteristics (weight 0.5): Filtered from historical data users with "young women, 28 years old, casual style, and cheerful mood", a total of 120 records were obtained, of which 65% chose template A, 25% chose template B, and 10% chose template D;

[0105] Scene-emotion fit (weight 0.3): Template A (95 points), Template B (90 points), Template D (85 points);

[0106] Segment duration matching degree (weight 0.2): The fit between the candidate template duration and the total duration of the core segment (4.2s) + the duration of the transition segment (10s), template A (15s, fit 98%), template B (20s, fit 85%), template D (12s, fit 90%).

[0107] Calculate the final adaptation score (out of 100):

[0108] Template A: 65% × 0.5 × 100 + 95 × 0.3 + 98 × 0.2 = 32.5 + 28.5 + 19.6 = 80.6 points;

[0109] Template B: 25% × 0.5 × 100 + 90 × 0.3 + 85 × 0.2 = 12.5 + 27 + 17 = 56.5 points;

[0110] Template D: 10% × 0.5 × 100 + 85 × 0.3 + 90 × 0.2 = 5 + 25.5 + 18 = 48.5 points;

[0111] Sort by rating in descending order, and select Template A, "Joyful Folk Custom Interaction," as the best template.

[0112] S3.4 Edit the video according to the optimal template, and automatically synthesize a short video by matching the user's core behavior segments with the template slow motion or special effects nodes and ordinary segments with the regular editing nodes.

[0113] For example, the synthesis steps include:

[0114] Clip alignment: The user's three core jump clips (1.2-1.5s each) are precisely aligned with the two slow-motion nodes of template A (3s-4.5s and 8s-9.5s), with the peak jump time and the slow-motion start time deviating by ≤0.1s; the stepping and clapping transition clips are aligned with the template's regular editing nodes (transitions, background music accents);

[0115] Music matching: Automatically loads the upbeat ethnic music (BPM 120) built into Template A. The rhythm of the music is synchronized with the sound of the bamboo pole being struck and the rhythm of the user's movements. The volume ratio is 70% for the music and 30% for ambient sound.

[0116] Subtitle generation: Based on scene and emotion tags, automatically generate subtitles such as "Successfully checked in at the folk village bamboo pole dance~" and "Happiness is dancing to the rhythm!";

[0117] Output format: Creates a 15-second custom short video with a default resolution of 4K (preserving the original image quality), and supports users to switch to 1080P / 720P when downloading later.

[0118] S4. The system will associate and bind the generated exclusive short videos with user feature data to establish a one-to-one or many-to-one short video ownership relationship, so that users can preview and download them online.

[0119] S4.1 Users may initiate an acquisition request through at least one of the following channels: QR code signs set up within the scenic area, the official mini-program of the scenic area, or the self-service terminal at the visitor center;

[0120] S4.2 After receiving the request, the system will display a verification interface. The user enters the order number or ID card number, and the system will compare the data with the scenic area's ticketing system. If the match is successful, the system will be able to verify the information.

[0121] S4.3 After successful verification, the user enters the short video preview interface. After the user selects the target resolution, the system immediately generates a short video of the corresponding resolution for the user to download to their local device or share to a social media platform.

[0122] It also involves the secondary optimization steps for short videos:

[0123] After users obtain short videos, they can initiate optimization requests through the official mini-program to change background music, adjust transition styles, add personalized text, change filters, and trim clips.

[0124] The cloud-based AI engine re-invokes template library resources or adjusts generation parameters based on user optimization needs to generate optimized short videos;

[0125] Users can preview the optimized short videos through the original identity verification channel and can initiate optimization requests again or download and share them directly.

[0126] On the other hand, the present invention also provides a travel photography short video generation system, comprising:

[0127] The scene camera deployment module deploys high-speed cameras with inertial measurement units in dynamic scenes of the scenic area, and ambient light-sensing adaptive cameras in static scenes. All cameras are connected to the edge computing gateway.

[0128] The scene-based shooting module allows the camera to activate the corresponding shooting mode based on the scene type. In dynamic scenes, the camera tracks the motion trajectory through a multi-camera spatiotemporal synchronization algorithm, while in static scenes, the shooting parameters are adjusted according to the light and the flow of people.

[0129] The dedicated video generation module uses a cloud-based AI engine to extract users' behavioral, emotional, and basic characteristics, and combines these with scene features to select suitable video templates from the template library through a collaborative filtering algorithm, generating a personalized short video.

[0130] The user preview and download module associates the generated exclusive short videos with user feature data, establishing a one-to-one or many-to-one short video ownership relationship, so that users can preview and download them online.

[0131] By installing high-speed cameras with inertial measurement units in dynamic scenes within scenic areas, videos can be captured using these high-speed cameras and downloaded to mobile devices, thus avoiding the inconvenience of manually holding the camera. At the same time, the videos captured by the high-speed cameras are free from shaking, ensuring image quality. In addition, the cloud-based AI engine can automatically generate corresponding short videos, reducing the need for manual editing.

[0132] The above description is merely a specific embodiment of the invention, but the scope of protection of the invention is not limited thereto. Any variations or substitutions conceived without inventive effort should be included within the scope of protection of the invention. Therefore, the scope of protection of the invention should be determined by the scope defined in the claims.

[0133] The embodiments described above are merely preferred embodiments of this specification and are not intended to limit the scope of this specification. Any modifications and improvements made by those skilled in the art to the technical solutions of this specification without departing from the spirit of this specification should fall within the protection scope defined by the claims of this specification.

Claims

1. A method for generating short travel videos, characterized in that, Includes the following steps: S1. Deploy high-speed cameras with inertial measurement units in dynamic scenes of the scenic area, and deploy ambient light adaptive cameras in static scenes. All cameras are connected to the edge computing gateway. S2. The camera activates the corresponding shooting mode according to the scene type. In dynamic scenes, the camera tracks the motion trajectory through a multi-camera spatiotemporal synchronization algorithm. In static scenes, the shooting parameters are adjusted according to the light and the flow of people. S3, the cloud-based AI engine extracts the user's behavioral, emotional, and basic characteristics, and combines them with scene characteristics to select suitable video templates from the template library through a collaborative filtering algorithm to generate exclusive short videos; S4. The system will associate and bind the generated exclusive short videos with user feature data to establish a one-to-one or many-to-one short video ownership relationship, so that users can preview and download them online.

2. The method for generating short travel videos according to claim 1, characterized in that, In the S2, the workflow of the dynamic scene shooting mode includes the following steps: After the high-speed camera is started, a multi-camera communication link is established through the edge computing gateway. The multi-camera timestamp calibration is completed by using a dual synchronization mechanism of Network Time Protocol (NTP) and Precision Time Protocol (PTP). Based on GPS positioning data and the scenic area's electronic map, a unified spatial coordinate system was established to determine the spatial coordinates of each camera position and complete the spatial reference calibration. The high-speed camera collects the device's own motion posture data in real time, with the collection frequency consistent with the camera's frame rate, and binds the motion data and video frame data to the edge computing gateway; The system detects video frames captured by each camera position and extracts the user's human body contour and key feature points to achieve cross-view locking of the same user from multiple camera positions. The edge computing gateway fuses video frame data from multiple cameras with motion data from high-speed cameras, and uses a Kalman filter algorithm to eliminate noise in the high-speed camera data to obtain a smooth trajectory of device motion. Based on the pixel coordinates of key user feature points and the spatial coordinates of the camera position, the three-dimensional position of the user in a unified spatial coordinate system is calculated by triangulation. Combined with the position data of continuous frames, the particle filter algorithm is used to predict the user's motion trajectory, realizing real-time tracking of the user's motion trajectory by multiple cameras.

3. The method for generating short travel videos according to claim 2, characterized in that, The method of extracting the user's human body contour and key feature points to achieve cross-view locking of the same user from multiple camera angles includes the following steps: Each camera position uses the YOLOv8 model to detect human targets in real-time video frames and outputs the bounding box coordinates of valid human targets; Extract and compare the global human body contour features, key human body feature points, and clothing features of adjacent frames to confirm that the target in the current frame and the previous frame is the same user. Based on the target features of the high-point camera position, the cosine similarity of the contour feature vectors of other camera positions and the weighted cosine similarity of the local feature vectors of the benchmark target are calculated. When the similarity meets the threshold, the global user ID is bound to achieve cross-view locking of the same user by multiple camera positions.

4. The method for generating short travel videos according to claim 1, characterized in that, S3 specifically includes the following steps: The cloud-based AI engine receives encrypted video clips uploaded by the edge computing gateway and performs structured extraction of user features; Based on scene feature label matching templates in the same scene library, several candidate templates were initially selected. Based on historical template selection data of users with similar scenarios and characteristics, the remaining templates are sorted, and the optimal template is selected. Edit videos using the optimal template, matching core user behavior segments with template slow-motion or special effects nodes, and ordinary segments with regular editing nodes, and automatically synthesize short videos.

5. The method for generating short travel videos according to claim 4, characterized in that, The steps for structured extraction of user features are as follows: Based on the human pose estimation model, key user actions are identified, the duration of each action is counted, and core behavioral segments are marked. By analyzing key facial expressions of users through a facial expression recognition model, emotional labels such as pleasure and calmness are output.

6. The method for generating short travel videos according to claim 1, characterized in that, In step S4, the user's download steps are as follows: Users can initiate an access request through at least one of the following channels: QR code signs set up in the scenic area, the official mini-program of the scenic area, or the self-service terminal in the visitor center; After receiving the request, the system pops up a verification interface. The user enters the order number or ID card number, and the system compares the data with the scenic spot's ticketing system. If the match is successful, the verification is successful. After verification, users enter the short video preview interface. After selecting the target resolution, the system immediately generates a short video of the corresponding resolution for users to download to their local devices or share to social media platforms.

7. The method for generating short travel videos according to claim 6, characterized in that, Step S4 also involves a secondary optimization step for the short video: After users obtain short videos, they can initiate optimization requests through the official mini-program to change background music, adjust transition styles, add personalized text, change filters, and trim clips. The cloud-based AI engine re-invokes template library resources or adjusts generation parameters based on user optimization needs to generate optimized short videos; Users can preview the optimized short videos through the original identity verification channel and can initiate optimization requests again or download and share them directly.

8. A travel photography short video generation system, the system being applicable to the travel photography short video generation method according to any one of claims 1-7, characterized in that, include: The scene camera deployment module deploys high-speed cameras with inertial measurement units in dynamic scenes of the scenic area, and ambient light-sensing adaptive cameras in static scenes. All cameras are connected to the edge computing gateway. The scene-based shooting module allows the camera to activate the corresponding shooting mode based on the scene type. In dynamic scenes, the camera tracks the motion trajectory through a multi-camera spatiotemporal synchronization algorithm, while in static scenes, the shooting parameters are adjusted according to the light and the flow of people. The dedicated video generation module uses a cloud-based AI engine to extract users' behavioral, emotional, and basic characteristics, and combines these with scene features to select suitable video templates from the template library through a collaborative filtering algorithm, generating a personalized short video. The user preview and download module associates the generated exclusive short videos with user feature data, establishing a one-to-one or many-to-one short video ownership relationship, so that users can preview and download them online.

Images