Video caching and pushing method and device of automobile data recorder and medium

Abstract

The invention discloses a video caching and pushing method and device of an automobile data recorder and a medium. The video caching and pushing method comprises the steps of firstly obtaining a video viewing interaction log recorded by a terminal device; then, performing feature extraction on the video viewing interaction log to obtain multi-dimensional feature vectors of a plurality of cached videos; and then, according to the video viewing interaction log and the multi-dimensional feature vector, performing preference analysis to obtain user preference information. After external environment data corresponding to the cached videos are obtained, dynamic cache adjustment is further carried out on the multiple cached videos according to the external environment data and the user preference information, and a target video cache sequence is obtained; and pushing a corresponding cached video to the user according to the target video cache sequence. According to the embodiment of the invention, dynamic cache adjustment is carried out on the cached video by combining the user preference information and the external environment data of the cached video, so that the cached video can be distributed more reasonably, and the user can preferentially receive the cached video which most meets the requirements of the user.

Description

Technical Field

[0001] The embodiments of the present application relate to, but are not limited to, the field of video caching technology, and in particular to a video caching and pushing method, device, and medium for a driving recorder. Background Art

[0002] Traditional dashcams usually only have simple video recording and storage functions, and lack intelligent analysis and processing of video content. This results in the video data captured by dashcams being stored in raw form without screening and processing, which not only takes up a lot of storage space, but also requires users to spend a lot of time searching for the required content from massive videos when using them. In addition, traditional dashcams also have limitations in video push. Usually, videos can only be pushed according to chronological order or manual selection by users, and cannot be pushed intelligently according to the actual needs and preferences of users. Although some dashcams have begun to try to introduce simple video classification and retrieval functions, these functions are usually based on fixed rules or templates and lack in-depth exploration of user personalized needs. Summary of the invention

[0003] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0004] The embodiment of the present application provides a video caching and pushing method for a driving recorder, which can dynamically adjust the video cache, thereby ensuring that the video distribution is more reasonable, so that users can preferentially receive the video that best meets their needs.

[0005] In a first aspect, an embodiment of the present application provides a video caching and pushing method for a driving recorder, comprising: obtaining a video viewing interaction log recorded by a terminal device; performing feature extraction on the video viewing interaction log to obtain a multidimensional feature vector of multiple cached videos; performing preference analysis based on the video viewing interaction log and the multidimensional feature vector to obtain user preference information; obtaining external environment data of the cached video; dynamically caching the multiple cached videos based on the external environment data and the user preference information to obtain a target video cache sequence; and pushing the corresponding cached video to the user based on the target video cache sequence.

[0006] In combination with the first aspect, in an embodiment of the present application, the video viewing interaction log includes video logs of multiple cached videos, each of the video logs including the viewing time, event type, video category, user interaction information and video tag information of the corresponding cached video; the feature extraction of the video viewing interaction log to obtain multi-dimensional feature vectors of the multiple cached videos includes: feature extraction of the viewing time of the multiple cached videos to obtain a time feature vector; feature extraction of the event type of the multiple cached videos to obtain an event type weight feature vector; feature extraction of the video category of the multiple cached videos to obtain a viewing frequency feature vector; feature extraction of the user interaction information corresponding to the multiple cached videos to obtain an interaction behavior feature vector; feature extraction of the video tag information of the multiple cached videos to obtain a content relevance feature vector; and obtaining multi-dimensional feature vectors of the multiple cached videos based on the time feature vector, the event type feature vector, the viewing frequency feature vector, the interaction behavior feature vector and the content relevance feature vector.

[0007] In combination with the first aspect, in an embodiment of the present application, the preference analysis is performed based on the video viewing interaction log and the multi-dimensional feature vector to obtain user preference information, including: calculating the weight of each dimensional feature vector in the multi-dimensional feature vector based on the video viewing interaction log; obtaining the standard eigenvalue of each dimensional feature vector, performing weighted calculation based on the standard eigenvalue of each dimensional feature vector and the corresponding weight to obtain a preference score for each dimensional feature vector; calculating a preference probability based on the preference score, and performing a preference analysis based on the preference probability to obtain user preference information.

[0008] In combination with the first aspect, in an embodiment of the present application, the dynamic cache adjustment of the multiple cached videos according to the external environment data and the user preference information to obtain a target video cache sequence includes: preliminarily caching the multiple cached videos according to the user preference information to obtain an initial video cache sequence; dynamically cache adjusting the initial video cache sequence according to the external environment data, the event type and the video tag information to obtain a target video cache sequence.

[0009] In combination with the first aspect, in an embodiment of the present application, the external environment data of the cached video includes weather information and traffic information; the dynamic cache adjustment of the initial video cache sequence according to the external environment data, the event type and the video tag information to obtain a target video cache sequence includes: performing weighted calculation according to the weather information, the traffic information, the event type and the video tag information to obtain an information richness score of the cached video; performing risk assessment on the weather information, the traffic information and the video tag information according to a preset risk level assessment function to obtain the risk level of the cached video; obtaining a cache priority of the cached video according to the information richness score and the risk level; and dynamically sorting the initial video cache sequence according to the cache priority to obtain a target video cache sequence.

[0010] In combination with the first aspect, in an embodiment of the present application, the video tag information includes driving time information; the risk assessment of the weather information, the traffic information and the video tag information according to a preset risk level assessment function to obtain the risk level of the cached video includes: obtaining a first risk index of the weather information, a second risk index of the traffic information and a third risk index of the driving time information; obtaining a first risk factor and a second risk factor of the risk level assessment function; confirming a weather risk value according to the first risk index and the first risk factor, confirming a traffic risk value according to the second risk index and the first risk factor, and confirming a driving time risk value according to the third risk index and the first risk factor; obtaining the risk level of the cached video according to the weather risk value, the driving time risk value, the traffic risk value and the second risk factor.

[0011] In combination with the first aspect, in an embodiment of the present application, pushing the corresponding cached video to the user according to the target video cache sequence includes: encrypting the target video cache sequence to obtain an encrypted video cache sequence; confirming the user's identity information, and pushing the corresponding cached video to the user according to the encrypted video cache sequence.

[0012] In combination with the first aspect, in an embodiment of the present application, after pushing the corresponding cached video to the user according to the target video cache sequence, the method includes: obtaining the current video log recorded by the driving recorder; updating the user preference information according to the current video log to obtain the current preference information; dynamically sorting the target video cache sequence according to the current preference information to obtain an optimized target video cache sequence.

[0013] On the other hand, an embodiment of the present application further provides an electronic device, comprising: at least one processor; at least one memory for storing at least one program; and when at least one of the programs is executed by at least one of the processors, the video caching and pushing method as described above is implemented.

[0014] On the other hand, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program executable by a processor. When the computer program executable by the processor is executed by the processor, it is used to implement the video caching and pushing method as described above.

[0015] The embodiment of the present application provides a method for caching and pushing videos of a driving recorder. First, a video viewing interaction log recorded by a terminal device (such as a driving recorder, etc.) is obtained; then, feature extraction is performed on the video viewing interaction log to obtain multi-dimensional feature vectors of multiple cached videos, which are helpful to distinguish the importance of the video content and the degree of interest that the user may have; then, preference analysis is performed based on the video viewing interaction log and the multi-dimensional feature vector to obtain user preference information. After obtaining the external environment data corresponding to the cached video, the multiple cached videos can be further dynamically cached according to the external environment data and the user preference information to obtain a target video cache sequence; then the corresponding cached video is pushed to the user according to the target video cache sequence. The embodiment of the present application dynamically caches the cached video by combining the user preference information and the external environment data of the cached video, which can ensure that the cached video distribution in the target video cache sequence is more reasonable, so that the user can preferentially receive the cached video that best meets their needs. In summary, the embodiment of the present application realizes a more personalized and intelligent video storage and push service. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of a video caching and pushing method for a driving recorder provided in an embodiment of the present application;

[0017] Figure 2 The embodiment of this application provides Figure 1 Schematic diagram of specific steps of step 120;

[0018] Figure 3 The embodiment of this application provides Figure 1 Schematic diagram of specific steps of step 130;

[0019] Figure 4 is a specific flow chart of obtaining a target video cache sequence provided by an embodiment of the present application;

[0020] Figure 5 The embodiment of this application provides Figure 4 Schematic diagram of specific steps of step 420;

[0021] Figure 6 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0023] It should be noted that, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a different order from that in the flowchart. The terms "first", "second", etc. in the specification, claims and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions that the present invention can be implemented, so they have no technical substantive significance. Any modification of the structure, change of the proportional relationship or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope of the technical content disclosed by the present invention. At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" quoted in this specification are only for the convenience of narration, and are not used to limit the scope of the implementation of the present invention. The change or adjustment of the relative relationship should also be regarded as the scope of the implementation of the present invention without substantially changing the technical content.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0025] With the rapid development of the automotive industry and the advancement of intelligent technology, driving recorders have become one of the indispensable auxiliary devices for modern cars. Driving recorders can not only record video logs during driving, providing an important basis for determining responsibility for traffic accidents, but also assist drivers in improving driving safety. However, traditional driving recorders have many deficiencies in video processing and storage, especially in the efficient management and utilization of massive video data.

[0026] Traditional dashcams usually only have simple video recording and storage functions, and lack intelligent analysis and processing of video content. This results in the video data captured by dashcams being stored in raw form without screening and processing, which not only takes up a lot of storage space, but also requires users to spend a lot of time searching for the required content from massive videos when using them. In addition, traditional dashcams also have limitations in video push. Usually, videos can only be pushed according to chronological order or manual selection by users, and cannot be pushed intelligently according to the actual needs and preferences of users. Although some dashcams have begun to try to introduce simple video classification and retrieval functions, these functions are usually based on fixed rules or templates and lack in-depth exploration of user personalized needs.

[0027] In view of this, the embodiment of the present application provides a video caching and pushing method, electronic device and computer-readable storage medium for a driving recorder. First, the video viewing interaction log recorded by the terminal device is obtained; then, the video viewing interaction log is feature extracted to obtain multi-dimensional feature vectors of multiple cached videos, which are helpful to distinguish the importance of the video content and the degree of interest that the user may have; then, preference analysis is performed based on the video viewing interaction log and the multi-dimensional feature vector to obtain user preference information. After obtaining the external environment data corresponding to the cached video, the multiple cached videos can be further dynamically cached according to the external environment data and the user preference information to obtain the target video cache sequence; then the corresponding cached video is pushed to the user according to the target video cache sequence. The embodiment of the present application dynamically caches the cached video by combining the user preference information and the external environment data of the cached video, which can ensure that the cached video distribution in the target video cache sequence is more reasonable, so that the user can receive the cached video that best meets their needs first. In summary, the embodiment of the present application realizes a more personalized and intelligent video storage and push service.

[0028] The embodiments of the present application are further described below in conjunction with the accompanying drawings.

[0029] like Figure 1 As shown, Figure 1 1 is a flow chart of a method for caching and pushing video of a driving recorder provided in an embodiment of the present application. The method includes but is not limited to steps 110 to 160.

[0030] Step 110: Obtaining a video viewing interaction log recorded by the terminal device;

[0031] Step 120: extracting features from the video viewing interaction log to obtain multi-dimensional feature vectors of multiple cached videos;

[0032] Step 130: performing preference analysis based on the video viewing interaction log and the multi-dimensional feature vector to obtain user preference information;

[0033] Step 140: Obtain external environment data of the cached video;

[0034] Step 150: dynamically adjust the cache of multiple cached videos according to the external environment data and the user preference information to obtain a target video cache sequence;

[0035] Step 160: Push the corresponding cached video to the user according to the target video cache sequence.

[0036] Steps 110 to 160 are described in detail below.

[0037] In a feasible embodiment, the terminal device for recording the video viewing interaction log includes a driving recorder and a smart device (such as a smart phone) connected to the driving recorder. Among them, the driving recorder, as an important part of modern automobile safety equipment, can record video and audio information in real time during the vehicle's driving process. These video logs are extremely valuable for analyzing driving behavior, traffic accident investigation, vehicle safety monitoring, etc. Normally, the driving recorder can store video logs in built-in storage media (such as SD cards, solid-state drives, etc.). The smart terminal connected to the driving recorder provides users with a convenient way to obtain video logs.

[0038] In a feasible embodiment, the video log is usually composed of a series of continuous video frames, each of which can capture the real-time picture of the vehicle during driving. Usually, the dash cam is equipped with built-in sensors that can monitor key data such as acceleration, braking status, steering operation, etc. in real time. Once abnormal driving behavior (such as sudden braking, collision, etc.) is detected, the dash cam can automatically mark the current video clip and record metadata such as event type and timestamp. These video logs not only contain video content, but also integrate rich metadata such as timestamp, vehicle speed, geographic location, etc.

[0039] It should be noted that each dashcam can correspond to a specific user ID. This correspondence is not only based on the design logic of the dashcam itself, but also takes into full consideration multiple dimensions such as the actual needs of the user, data security and privacy protection. On the one hand, dashcams are usually equipped with advanced identity recognition systems that can accurately record and identify the user's identity information. This identity information may exist in the form of user names, passwords, biometrics (such as fingerprints, facial recognition) or physical media (such as USB flash drives, smart cards), ensuring that only authorized users can access and operate the dashcam. On the other hand, each dashcam corresponds to a user ID, which helps to achieve personalized settings and accurate data management. Users can personalize the dashcam according to their driving habits and preferences, such as adjusting the video quality, setting emergency video trigger conditions, etc. At the same time, through the user ID, the dashcam can automatically associate the recorded data with the corresponding user, making it convenient for users to quickly find and review driving records when needed.

[0040] In a feasible embodiment, by assigning a unique user ID to each dashcam, users can quickly locate and access the video logs associated with it. For example, when a vehicle accident occurs and video evidence needs to be retrieved, the user only needs to enter his unique user ID to retrieve the required video log from the records of the relevant dashcam. In addition, when viewing the video log, the user can not only browse the video content, but also slow down, pause and fast forward the video, so as to observe and analyze the details in the video more carefully.

[0041] In a feasible embodiment, the video log recorded by the dash cam is composed of multiple continuous video frames, which can be combined into multiple video clips. When the user views and interacts with these video logs, his behavior will be recorded by the dash cam or the smart device connected to it to form a video viewing interaction log. The video viewing interaction log is a set of logs generated when the user views and interacts with the video content, which may contain multiple cached videos, which are generated when the user views and interacts with specific video clips. Among them, the log of each cached video in the video viewing interaction log records in detail the corresponding viewing time, event type, video category, user interaction information, and video tag information and other key information. Specifically, the viewing time refers to the specific time period or time point when the user watches these cached videos, which reflects the time distribution of the user watching the video. The event type refers to different types of events that occur in these cached videos, which are usually related to the driving status of the vehicle or the user's operating behavior. Common event types include but are not limited to: Start / Shutdown: Record the time when the vehicle starts or shuts down. Collision sensing: When the vehicle collides, the dash cam can automatically detect and record it. Emergency braking: Record the emergency braking of the vehicle. Turning: Records the turning behavior of the vehicle, including left turn, right turn, etc. Speeding: Records the situation where the vehicle speed exceeds the set threshold. Video category refers to the type or theme of these cached videos. Video logs can be classified according to different video categories, for example: Daily driving: Records the video of the vehicle during normal driving. Parking monitoring: When the vehicle is parked, the dashcam can still record the video of the surrounding environment. Scenery recording: Records the scenery or interesting scenes along the way. Accident recording: Records the video when the vehicle has an accident. These videos usually contain key information such as collision and rear-end collision. Video categories help users quickly locate the video content of interest, or filter and view according to specific needs. User interaction information refers to the various behaviors of users interacting with the video during the video viewing process, such as clicking, pausing, fast forward, rewinding, zooming in, zooming out, etc. This information can reflect the user's viewing habits and needs. Video tag information includes the tags of the cached video itself, which can be used to describe the content, type or characteristics of the video. For example, videos can be marked as various types according to driving time periods, such as daytime driving, nighttime driving, daytime parking, and nighttime parking. At the same time, they can also be classified according to the frame rate of the video, such as normal frame rate video and time-lapse video with 1 frame per second.

[0042] See also Figure 2 The specific process of extracting features from the video viewing interaction log in step 120 to obtain multi-dimensional feature vectors of multiple cached videos may include but is not limited to steps 210 to 260.

[0043] Step 210: extracting features of the viewing time of multiple cached videos to obtain a time feature vector;

[0044] Step 220: extracting features of event types of multiple cached videos to obtain event type weight feature vectors;

[0045] Step 230: extracting features from video categories of multiple cached videos to obtain viewing frequency feature vectors;

[0046] Step 240: extracting features from user interaction information corresponding to the plurality of cached videos to obtain an interaction behavior feature vector;

[0047] Step 250: extracting features from video tag information of multiple cached videos to obtain content relevance feature vectors;

[0048] Step 260: Obtain multi-dimensional feature vectors of the plurality of cached videos according to the time feature vector, the event type feature vector, the viewing frequency feature vector, the interaction behavior feature vector, and the content relevance feature vector.

[0049] In a feasible embodiment, after obtaining the video viewing interaction log, the video logs of multiple cached videos contained therein may be traversed to perform feature extraction, thereby obtaining multi-dimensional feature vectors of the multiple cached videos.

[0050] In a feasible embodiment, in step 210, in the process of extracting the viewing time features of multiple cached videos to construct a time feature vector, the timestamps in the video log can be first analyzed, the number of hours each video is watched is counted, and it is taken into consideration as one of the key features. It should be noted that in addition to the number of hours, other feature elements closely related to the time dimension can also be taken into consideration, including but not limited to subdividing the viewing time into specific time periods such as morning, afternoon, and evening, and distinguishing the differences in viewing habits between weekdays and weekends. By combining these time features, a comprehensive time feature vector can be constructed, which can accurately reflect the user's video viewing behavior patterns in different time contexts.

[0051] In a feasible embodiment, in step 220, in the process of extracting features of event types of multiple cached videos, in particular, for event types of multiple cached videos, the following strategy can be adopted to construct an event type weight feature vector: First, by traversing the video logs of these cached videos, the event types and their corresponding viewing times that appear in each video log are recorded and counted in detail. Subsequently, based on these statistical information, the viewing ratio of each event type is further calculated, that is, the percentage of the viewing times of each event type in the total viewing times. This ratio data not only reflects the user's preference for various types of events.

[0052] In a feasible embodiment, in step 230, in the process of extracting features from the video categories of the plurality of cached videos, the video logs of the cached videos can be traversed to record and count the video categories and their corresponding viewing times in detail in each video log. According to these statistical results, after sorting and analysis, they can be finally converted into a feature vector reflecting the viewing frequency of the video category, i.e., the viewing frequency feature vector.

[0053] In a feasible embodiment, in step 240, the user interaction information covers various interactive behaviors generated by the user and the video content during the video viewing process. These behavior types may at least include click to play, pause play, fast forward, slow play, frame by frame play, rewind, zoom in and out, etc. In order to extract the feature vector of the interactive behavior from multiple cached videos, each type of interactive behavior recorded in the video log can be deeply analyzed. Specifically, the interactive behavior type corresponding to each video can be located according to the video log, and then the frequency of each interactive behavior is counted. Through this series of operations, a vector that can fully reflect the characteristics of the user's interactive behavior can be constructed, and the vector can intuitively display the user's preferred interaction mode through the frequency of behavior.

[0054] In a feasible embodiment, after successfully extracting the multi-dimensional feature vectors of each cached video, these feature vectors can be standardized, specifically including the time feature vector, event type feature vector, viewing frequency feature vector, interactive behavior feature vector, and content relevance feature vector. In order to achieve standardization, each dimension can be normalized to ensure that all feature vectors remain consistent in value, which is convenient for subsequent analysis and comparison.

[0055] In a feasible embodiment, after obtaining the multi-dimensional feature vectors of the plurality of cached videos that have been subjected to normalization processing, these feature vectors can be further combined with the video viewing interaction logs for in-depth analysis to obtain user preference information. Figure 3 In step 130, a comprehensive preference analysis is performed based on the video viewing interaction log and the multi-dimensional feature vector to obtain a specific process of user preference information. This process may include but is not limited to steps 310 to 330.

[0056] Step 310: Calculate the weight of each dimensional feature vector in the multi-dimensional feature vector according to the video viewing interaction log;

[0057] Step 320: Obtain the standard eigenvalue of each dimensional feature vector, perform weighted calculation according to the standard eigenvalue of each dimensional feature vector and the corresponding weight, and obtain the preference score of each dimensional feature vector;

[0058] Step 330: Calculate the preference probability according to the preference score, perform preference analysis based on the preference probability, and obtain user preference information.

[0059] In a feasible embodiment, the purpose of step 310 is to extract useful information from the video viewing interaction log to determine the relative importance of each dimensional feature vector in the multidimensional feature vector. In the process of calculating the weight of each dimensional feature vector in the multidimensional feature vector according to the video viewing interaction log, the video viewing interaction log can be first cleaned to remove irrelevant information, erroneous data and duplicate records, and then extract features related to the multidimensional feature vector from the cleaned log, such as viewing time, playback events (such as pause, fast forward, rewind, etc.), viewing frequency, interaction intensity (such as clicks, comments, sharing, etc.), content relevance (such as user preferences for different types or theme videos), etc., and then the extracted features are standardized (such as normalization) to ensure that they are compared on the same scale. Subsequently, statistical methods (such as correlation coefficients, covariance matrices) or machine learning algorithms (such as feature importance evaluation) are used to analyze the correlation between different features and their association with user preferences. On this basis, the weight of each dimensional feature vector can be calculated according to the results of the correlation analysis. In this step, a variety of methods can be flexibly adopted: first, build a user behavior model to accurately evaluate the specific impact of different features on user preferences by simulating the user's viewing and interaction process, so as to derive weights; second, reasonably allocate weights directly based on the strength of the correlation between features and user preferences; third, use machine learning algorithms (such as decision trees, random forests, gradient boosting machines, etc.) to accurately evaluate the importance of features and use this as a basis for allocating weights.

[0060] It is worth noting that if you choose to calculate the weights by building a user behavior model, you can further verify the calculated weights to ensure that they can accurately reflect the user's preferences for different dimensional features. Then, you can make necessary adjustments and optimizations to the weights based on the verification results to improve their accuracy.

[0061] In a feasible embodiment, in step 320, the standard eigenvalue represents the eigenvalue of the dimensional feature vector after standardization. Standardization is usually performed to ensure that all eigenvalues ​​are on the same scale to facilitate subsequent calculations. This process may include normalization, standardization to a specific range (such as 0-1) or other forms of processing to eliminate the influence of different feature dimensions. The weighted calculation is obtained by traversing the standard eigenvalue corresponding to each dimensional feature vector, then multiplying the value by the weight of the dimension, and adding all the products. The result of the weighted calculation is usually a scalar value, which combines the standard eigenvalues ​​and weights of all dimensional feature vectors, and can reflect the user's overall preference for all features. The higher the preference score, the higher the user's preference for all features (or the more it meets a certain standard).

[0062] In a feasible embodiment, in step 330, the preference probability is a quantitative representation that can be used to describe the user's preference for a video content of a specific type or feature. By performing preference analysis based on the preference probability, the user's viewing habits and preferences can be understood more accurately, thereby providing the user with more personalized video recommendations and services.

[0063] In a feasible embodiment, the preference score can be converted into a preference probability through a softmax function. The softmax function is a function that converts the original score (which may be a real number or a negative number) into a probability distribution, ensuring that the sum of all converted probabilities is 1. The expression of the softmax function is as follows:

[0064]

[0065] Where Z is the input feature vector, K is the length of the feature vector, and σ(Z) j It is the value of the jth element in the Softmax function output feature vector.

[0066] Specifically, the softmax function performs an exponential operation on the preference scores of each dimensional feature vector, and then divides these exponential values ​​by the sum of all the scores after the exponential operation to accurately calculate the probability value corresponding to each option. This process ensures that all output probability values ​​are between 0 and 1, and their sum is 1.

[0067] In a feasible embodiment, after obtaining the preference probability, preference analysis can be further carried out. This process mainly covers the interpretation and comparison of probability values, aiming to accurately reveal the user's preference for different options. For example, the option with the highest probability value can be found and regarded as the user's most preferred option; at the same time, the user's preference order for different options can be clearly understood by sorting the probability values. Based on the results of the preference analysis, user preference information can be extracted and sorted. This information may include the user's most preferred options, preference order, preference strength, etc. This information is of great value in understanding user needs and providing personalized recommendations.

[0068] It is understandable that during the driving recording process, the video recording can be integrated and labeled in real time from multiple sources to enhance the label information. This process is mainly achieved by integrating internal and external data sources (i.e. external environmental data), which can include: Weather conditions: extreme weather conditions such as heavy rain, fog, and heavy snow are crucial for evaluating the driving environment. GPS precise location information: not only provides the real-time location of the vehicle, but also further obtains key data such as the speed limit and road complexity of the location. Networked traffic information: including surrounding traffic situation, congestion level, and whether the vehicle is speeding or violating traffic rules, etc. These data help to understand the current road conditions and the compliance of driving behavior. GSensor acceleration sensor data: by monitoring dynamic information such as acceleration, deceleration, and sharp turns of the vehicle, it provides an important basis for driving behavior analysis. Time period characteristics: different time periods such as morning, noon, and evening may affect driving conditions and traffic flow. Geographic location characteristics: covering specific location information such as scenic spots, bridges, and large rivers, adding rich geographical background to the video content. AI video analysis: Use artificial intelligence technology to analyze objects or people of interest in the video and add corresponding labels, such as luxury sports cars, snow-capped mountains, rainbows, etc. These labels greatly enrich the description of the video content. These multi-dimensional labels are designed to describe the driving video content more comprehensively and meticulously. Each cached video corresponds to multiple dimensional labels, which not only helps to improve the richness and accuracy of the video content, but also provides great convenience for calculating the content relevance of multiple cached videos. In this way, video content that is highly relevant to user interests can be more effectively screened, sorted, and displayed.

[0069] In a feasible embodiment, after obtaining the user preference information, multiple cached videos can be further dynamically cached and adjusted according to the external environment data and the user preference information to obtain a target video cache sequence. Specifically, multiple cached videos can be initially cached according to the user preference information to obtain an initial video cache sequence, and then multiple cached videos in the initial video cache sequence can be dynamically cached and adjusted according to the external environment data, event type and video tag information of the cached videos. This process ensures the timeliness and pertinence of the video cache, and finally generates a target video cache sequence that meets the user preferences and external environment changes.

[0070] In a feasible embodiment, the external environment data of the cached video may include weather information and traffic information. Figure 4 The specific process of dynamically adjusting the initial video cache sequence according to the external environment data, event type and video tag information of the cached video to obtain the target video cache sequence may include but is not limited to steps 410 to 440.

[0071] Step 410: Perform weighted calculation based on weather information, traffic information, event type, and video tag information to obtain an information richness score of the cached video;

[0072] Step 420: Perform risk assessment on weather information, traffic information and video tag information according to a preset risk level assessment function to obtain a risk level of the cached video;

[0073] Step 430: Obtaining a cache priority of a cached video according to the information richness score and the risk level;

[0074] Step 440: dynamically sort the initial video cache sequence according to the cache priority to obtain a target video cache sequence.

[0075] It is understood that the video tag information includes the tags of the cached video itself, which can be used to describe the content, type or characteristics of the video. For example, the video can be marked as daytime driving, nighttime driving, daytime parking, nighttime parking, etc. according to the driving time; at the same time, it can also be classified according to the frame rate of the video, such as normal frame rate video and 1 frame per second time-lapse video.

[0076] In a feasible embodiment, the video tag information includes the internal data source tags of the cached video, and these tags can describe the driving time information of the video in detail. For example, the video can be carefully marked as multiple types such as daytime driving, nighttime driving, daytime parking, and nighttime parking according to the driving time period, and these time period information can constitute the time information tag of the cached video. Similarly, weather information, traffic information, and event types can also be included in the corresponding tag categories.

[0077] In a feasible embodiment, in the processing flow of step 410, a basic score can be set according to the number of tag categories contained in the video. Generally speaking, the more the number of tag categories, the richer the information carried by the video. In this embodiment, since the video contains tags of four categories, namely time, weather, traffic and event type, the basic score is set to 4 (i.e., the number of tag categories). Next, the number of tags under each tag category is counted. To this end, a dictionary can be initialized, in which the key is the tag category (such as weather, traffic, etc.), the value is the number of tags in the corresponding category, and the number of all tag categories is initially set to 0. Subsequently, each tag in the video is traversed, and it is classified into the corresponding category according to the content of the tag, and the count of the category is increased. For example, the tag "time_morning" (a tag related to time content) can be classified into the time information tag category, and the count of the category is increased. After completing the classification and counting of the tags, it can be confirmed how many categories have a number of tags greater than 0. For each such category, a unit value (i.e., 1) can be contributed to it in the summation operation. Then, multiply the sum by 0.5 to get the diversity bonus score. Finally, add the diversity bonus score to the base score to get the information richness score of the cached video. This score can comprehensively reflect the amount of information in the video and the diversity of the label categories.

[0078] See also Figure 5 In a feasible embodiment, the specific process of step 420 may include but is not limited to steps 510 to 540.

[0079] Step 510: Obtaining a first risk index of weather information, a second risk index of traffic information, and a third risk index of driving time information according to a preset risk level assessment function;

[0080] Step 520: Obtain a first risk factor and a second risk factor of a risk level assessment function;

[0081] Step 530: confirming a weather risk value according to the first risk index and the first risk factor, confirming a traffic risk value according to the second risk index and the first risk factor, and confirming a driving time risk value according to the third risk index and the first risk factor;

[0082] Step 540: Obtain the risk level of the cached video according to the weather risk value, the driving time risk value, the traffic risk value and the second risk factor.

[0083] In a feasible embodiment, it should be noted that a dictionary containing external environment data is defined in the preset risk level assessment function. This dictionary covers weather information, traffic information and time information. For each category of information, the risk level assessment function also defines a nested dictionary for storing risk factors of different categories and their corresponding risk values. The key of this nested dictionary is the risk factor category (such as weather, traffic, time), and the corresponding value is a smaller dictionary. In this smaller dictionary, the key is a specific risk factor (such as rain, congestion, peak hours), and the value is the risk value corresponding to these risk factors (floating point numbers between 0 and 1). Specifically, in the weather information, weather contents such as rain, snow, fog and normal can be included, and risk values ​​are set for them respectively, for example, the risk value of rain is 0.7, the risk value of snow is 0.8, the risk value of fog is 0.6, and the normal risk value is 0.2. In the traffic information, traffic complexity contents such as congestion, general and smooth traffic can be included, and corresponding risk values ​​can be set, such as the risk value of traffic congestion is 0.6, the risk value of general traffic is 0.4, and the risk value of smooth traffic is 0.2. For time information, time contents such as night, peak hours and normal can be included, and risk values ​​can be set for them respectively, such as the risk value of night is 0.5, the risk value of peak hours is 0.4, and the risk value of normal is 0.2. In the processing flow of step 510, each label category in the dictionary of the function can be traversed, and the corresponding risk value can be obtained from the nested dictionary. This value is the risk index of the corresponding label category. It is worth noting that if the risk value of a certain category does not exist in the dictionary, the function can use the first risk factor (such as 0.2) as the risk value of the category by default. Then, the function can add the risk values ​​of each category to obtain the total risk level of the cached video. It should be noted that if the calculated total risk value exceeds the second risk factor (such as 1), the function can return 1 as the final risk level; otherwise, it returns the actual calculated total risk value.

[0084] In a feasible embodiment, when calculating the cache priority of a video, a comprehensive method can be used, which combines the preset basic priority, the information richness score of the video and the risk level. Specifically, the calculation formula of the cache priority can be expressed as:

[0085] a_prior ity=base_prior ity*(1+in_r ich)*(1+ri sk_score).

[0086] Wherein, a_priority represents the cache priority of the video. base_priority represents the preset basic priority, which is a fixed value used to determine the cache order of the video when there is no interference from other factors. in_rich represents the information richness score of the video, which is a floating point number between 0 and 1, used to measure the richness and attractiveness of the video content. The richer the information and the more attractive the video is, the higher its information richness score. risk_score represents the risk level of the video, which is also a floating point number between 0 and 1. It should be noted that this formula is only a possible implementation method, and the specific implementation details may need to be adjusted according to the actual application scenario and requirements. The embodiments of the present application do not limit the implementation method of the formula.

[0087] In a feasible embodiment, in the processing flow of step 440, all videos can be sorted according to the cache priority list of cached videos (including the cache priority of all videos) using a sorting algorithm (such as quick sort, merge sort, etc.). The sorting result is the target video cache sequence. It is worth noting that when a new video is added or the information of an existing video (such as a tag, external data) changes, this step can be re-executed to dynamically adjust the cache sequence.

[0088] It should be noted that, in the process of executing step 410, in order to further improve the comprehensiveness and accuracy of the calculation, more types of data source tags can be introduced, which cover other key information in internal and external data sources. After integrating these additional data source tags, they are carefully weighted. In this process, the weight can be reasonably allocated according to the importance and relevance of each tag, thereby obtaining a more accurate and comprehensive information richness score.

[0089] In a feasible embodiment, after obtaining the target video cache sequence, the video with the highest cache priority can be pushed to the user first according to the order of the videos in the sequence. This strategy is intended to significantly improve the speed and fluency of users watching videos in real time. By ensuring that high-priority videos are cached and pushed first, users can access the content they are interested in or need more quickly, thereby optimizing the overall user experience. Such optimization not only reduces the user's waiting time, but also improves the availability and response speed of video content.

[0090] In a feasible embodiment, when pushing the corresponding cached video to the user according to the target video cache sequence, multiple security measures can be taken to strengthen the security of video transmission and caching. Specifically, the target video cache sequence can be encrypted first to obtain an encrypted video cache sequence. Subsequently, the user's identity information is confirmed to ensure that only verified legitimate users can access the video content. After confirming the user's identity, the corresponding cached video is pushed to the user according to the encrypted video cache sequence. This process ensures both the accurate push of video content and the security of data transmission.

[0091] In a feasible embodiment, during the video transmission and caching process, advanced public-private key encryption technology can be used to fully encrypt the video data, thereby effectively preventing the data from being intercepted or tampered with during the transmission process, ensuring the high security of the video data. In addition, blockchain technology can also be used for distributed storage and user identity authentication. The decentralized and tamper-proof characteristics of blockchain technology can provide guarantees for the security and reliability of video data. Specifically, through blockchain technology, distributed storage of video data can be achieved, and the redundancy and fault tolerance of data can be improved; at the same time, blockchain technology can also be used for user identity authentication to ensure that only legitimate users can access and view video content.

[0092] In a feasible embodiment, after successfully pushing the corresponding cached video to the user according to the target video cache sequence, a dynamic update mechanism can be introduced to further improve the user experience and service personalization. Specifically, first, obtain the current video log recorded in real time by the driving recorder. Then, based on these current video logs and combined with the user's past behavior data, the user's preference information is dynamically updated to obtain the current preference information that is closer to the user's current needs. Subsequently, according to the updated current preference information, the target video cache sequence is dynamically sorted, and an optimized target video cache sequence can be obtained, which can better meet the user's personalized needs.

[0093] Reference Figure 6 The embodiment of the present application further discloses an electronic device, the electronic device 800 comprising:

[0094] at least one processor 601;

[0095] At least one memory 602, used to store at least one program;

[0096] When at least one program is executed by at least one processor, the video caching and pushing method as described above is implemented

[0097] The present application also discloses a computer-readable storage medium.

[0098] A computer program executable by a processor is stored therein, and when the computer program executable by the processor is executed by the processor, it is used to implement the above-mentioned video caching and pushing method.

[0099] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A video caching and pushing method for a driving recorder, characterized in that: include: Obtaining video viewing interaction logs recorded by terminal devices; Performing feature extraction on the video viewing interaction log to obtain multi-dimensional feature vectors of multiple cached videos; Performing preference analysis based on the video viewing interaction log and the multi-dimensional feature vector to obtain user preference information; Acquire external environment data of the cached video; Dynamically adjust the cache of the plurality of cached videos according to the external environment data and the user preference information to obtain a target video cache sequence; The corresponding cached video is pushed to the user according to the target video cache sequence.

2. The video caching and pushing method according to claim 1, characterized in that: The video viewing interaction log includes video logs of multiple cached videos, each of which includes viewing time, event type, video category, user interaction information and video tag information of the corresponding cached video; The feature extraction of the video viewing interaction log to obtain a multi-dimensional feature vector of a plurality of cached videos includes: Extracting features of the viewing time of the plurality of cached videos to obtain a time feature vector; Extracting features of the event types of the plurality of cached videos to obtain an event type weight feature vector; Extracting features of the video categories of the plurality of cached videos to obtain a viewing frequency feature vector; Extracting features of the user interaction information corresponding to the plurality of cached videos to obtain an interaction behavior feature vector; Performing feature extraction on the video tag information of the plurality of cached videos to obtain a content relevance feature vector; Multi-dimensional feature vectors of multiple cached videos are obtained according to the time feature vector, the event type feature vector, the viewing frequency feature vector, the interactive behavior feature vector, and the content relevance feature vector.

3. The video caching and pushing method according to claim 2, characterized in that: The performing preference analysis according to the video viewing interaction log and the multi-dimensional feature vector to obtain user preference information includes: Calculating the weight of each dimensional feature vector in the multi-dimensional feature vector according to the video viewing interaction log; Obtaining a standard eigenvalue of each of the dimensional feature vectors, performing weighted calculation according to the standard eigenvalue of each of the dimensional feature vectors and the corresponding weight, to obtain a preference score of each of the dimensional feature vectors; The preference probability is calculated according to the preference score, and the preference analysis is performed based on the preference probability to obtain the user preference information.

4. The video caching and pushing method according to claim 2, characterized in that: The dynamically caching the plurality of cached videos according to the external environment data and the user preference information to obtain a target video cache sequence includes: Preliminarily caching the plurality of cached videos according to the user preference information to obtain an initial video cache sequence; The initial video cache sequence is dynamically cache-adjusted according to the external environment data, the event type and the video tag information to obtain a target video cache sequence.

5. The video caching and pushing method according to claim 4, characterized in that: The external environment data of the cached video includes weather information and traffic information; The dynamically caching and adjusting the initial video cache sequence according to the external environment data, the event type and the video tag information to obtain a target video cache sequence includes: Performing weighted calculation according to the weather information, the traffic information, the event type, and the video tag information to obtain an information richness score of the cached video; Performing risk assessment on the weather information, the traffic information and the video tag information according to a preset risk level assessment function to obtain a risk level of the cached video; Obtaining a cache priority of the cached video according to the information richness score and the risk level; The initial video cache sequence is dynamically sorted according to the cache priority to obtain a target video cache sequence.

6. The video caching and pushing method according to claim 5, characterized in that: The video tag information includes driving time information; The step of performing risk assessment on the weather information, the traffic information, and the video tag information according to a preset risk level assessment function to obtain the risk level of the cached video includes: Acquire a first risk index of the weather information, a second risk index of the traffic information, and a third risk index of the driving time information; Obtaining a first risk factor and a second risk factor of the risk level assessment function; Determine a weather risk value according to the first risk index and the first risk factor, determine a traffic risk value according to the second risk index and the first risk factor, and determine a driving time risk value according to the third risk index and the first risk factor; A risk level of the cached video is obtained according to the weather risk value, the driving time risk value, the traffic risk value and the second risk factor.

7. The video caching and pushing method according to claim 1, characterized in that: The pushing the corresponding cached video to the user according to the target video cache sequence includes: Encrypting the target video cache sequence to obtain an encrypted video cache sequence; Confirm the user's identity information and push the corresponding cached video to the user according to the encrypted video cache sequence.

8. The video caching and pushing method according to claim 4, characterized in that: After pushing the corresponding cached video to the user according to the target video cache sequence, the method includes: Get the current video log recorded by the dashcam; Updating the user preference information according to the current video log to obtain current preference information; The target video cache sequence is dynamically sorted according to the current preference information to obtain an optimized target video cache sequence.

9. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When at least one of the programs is executed by at least one of the processors, the video caching and pushing method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program executable by a processor is stored therein, and when the computer program executable by the processor is executed by the processor, it is used to implement the video caching and pushing method according to any one of claims 1 to 8.

Images