Method and system for quickly searching and positioning target in surveillance video

Abstract

The invention discloses a method and a system for quickly searching and positioning a target in a surveillance video, and the method comprises the steps: carrying out the large-class marking of the target in a key frame of the surveillance video, and generating first marking data; performing coarse classification marking on the targets of each large class according to the first marking data to generate second marking data; performing fine classification marking on the roughly classified target according to the second marked data to generate third marked data; generating a hierarchical association index; calculating feature saliency; selecting the key frame with the highest feature saliency as a retrieval display picture; generating a retrieval video clip; when the retrieval input is a character, generating a first matching result; when the retrieval input is a picture, generating a second matching result; and retrieving the corresponding retrieval display pictures and the retrieval video clips from the hierarchical association index, and outputting the retrieval display pictures and the retrieval video clips according to a sequence of similarity from high to low. The method and the device are used for improving the accuracy and the efficiency of searching and positioning the target in the monitoring video.

Description

Technical Field

[0001] The present invention relates to the field of target search and positioning, and in particular to a method and system for quickly searching and positioning targets in surveillance videos. Background Art

[0002] With the widespread adoption of video surveillance systems in public safety, the massive amount of video footage has created significant challenges for target search and location. Traditional methods for viewing surveillance footage rely primarily on manual viewing or simple time-skipping. This approach is inefficient when faced with massive amounts of video data, making it difficult to quickly and accurately locate specific targets.

[0003] Existing video target retrieval technologies, using artificial intelligence (AI) object recognition algorithms, can analyze and label video content in real time and support rapid retrieval based on target type (such as people, vehicles, etc.). This technology improves target search efficiency by preprocessing and extracting features from video frames to create a retrieval index.

[0004] However, when it is necessary to simultaneously combine the multi-dimensional features of the target (such as appearance features, dynamic features, etc.) for precise search, the above-mentioned related technologies often lack a hierarchical classification and labeling mechanism, making it difficult to achieve multi-level associated retrieval of features, resulting in insufficient accuracy and practicality of the retrieval results, making it difficult to meet the needs of fast and accurate target positioning in practical applications. Summary of the Invention

[0005] The present application provides a method and system for quickly searching and locating targets in surveillance videos, which are used to improve the accuracy and efficiency of searching and locating targets in surveillance videos.

[0006] In a first aspect, the present application provides a method for quickly searching and locating a target in a surveillance video, marking the target in the key frame of the surveillance video into a large category and generating first marking data; Performing rough classification and labeling on each major category of objects according to the first labeling data to generate second labeling data; Performing fine classification and labeling on the coarsely classified objects according to the second labeling data to generate third labeling data; Extracting feature points of each marked object, and associating the feature points with the first marked data, the second marked data, and the third marked data to generate a hierarchical association index; Calculate the feature saliency of each target in the key frames with the same feature value in the surveillance video. The feature saliency is determined by a weighted score of the target's orientation angle, ambient light intensity, and the target's proportion in the frame. Select the key frame with the highest feature significance as the retrieval display image; Taking the retrieval display image as the reference point, take the video segments of preset length forward and backward and merge them to generate the retrieval video clip; When the search input is text, the feature identifier whose similarity with the text is greater than a preset similarity threshold is retrieved through the hierarchical association index to generate a first matching result; When the search input is an image, the feature value of the image is extracted and the feature value in the feature identifier library is compared to perform similarity calculation, and the feature identifier with a similarity greater than a preset similarity threshold is selected to generate a second matching result; According to the first matching result or the second matching result, the corresponding retrieval display picture and retrieval video clip are retrieved from the hierarchical association index, and the retrieval display picture and retrieval video clip are output in descending order of similarity.

[0007] By adopting the above technical solutions and building a hierarchical tagging system and feature extraction mechanism, the target retrieval process becomes more accurate and efficient. Broad-category tagging establishes a basic classification framework, while coarse and fine classification gradually improve tagging accuracy. Feature point extraction and association with multi-level tag data form an index structure that supports rapid positioning. The introduction of feature saliency takes into account key factors such as target orientation angle, ambient lighting intensity, and screen size, ensuring the quality of the retrieved and displayed images. A mechanism for merging video segments of preset durations preserves the target's complete behavioral characteristics. Dual-mode support for text and image retrieval enhances retrieval flexibility, while a similarity threshold ensures the accuracy of retrieval results. Outputting retrieval results sorted by similarity improves retrieval efficiency, allowing users to quickly locate the most relevant target information. This retrieval method, combining multi-level tagging and feature association, significantly improves the accuracy and efficiency of locating specific targets within massive amounts of surveillance data.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, marking targets in key frames of surveillance video into broad categories to generate first marking data specifically includes: Use deep learning network to detect targets on key frames and obtain the target's rectangular box coordinates and confidence level; When the confidence level is greater than a preset threshold, the target area is extracted based on the coordinates of the rectangular box; Perform semantic segmentation on the target area to obtain the contour boundary of the target; Calculate the target's shape feature parameters based on its contour boundary, the shape feature parameters including the target's aspect ratio, area ratio and contour complexity; A decision tree is constructed based on shape feature parameters to classify the target into one of human, animal, car, and aircraft, obtaining the target's general category labeling information; The rectangular frame coordinates, confidence level, contour boundary, shape feature parameters and major category label information of the target are integrated to generate first label data.

[0009] By adopting the above technical solution, a deep learning network is used for target detection and combined with semantic segmentation technology to achieve accurate extraction of the target area. The setting of the target confidence threshold filters out low-quality detection results and improves the reliability of subsequent processing. The extraction of the contour boundary provides an accurate basis for the calculation of shape feature parameters. Parameters such as aspect ratio, area ratio and contour complexity comprehensively characterize the shape characteristics of the target. The decision tree constructed based on shape feature parameters achieves accurate classification of major categories such as humans, animals, cars and aircraft. The integration of the target's rectangular box coordinates, confidence, contour boundary, shape feature parameters and major category labeling information generates comprehensive labeling data. This labeling method based on multi-dimensional features improves the accuracy of target recognition, reduces the occurrence of misclassification, and lays a solid foundation for subsequent refined classification.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, performing coarse classification labeling on each major category of objects based on the first labeling data to generate second labeling data specifically includes: According to the major category label information in the first labeled data, calling the classification model corresponding to the target major category from the pre-trained deep learning model library; Extract the target image area from the original image based on the target's rectangular frame coordinates and contour boundaries; The classification model is used to extract features from the target image area to obtain the local feature vector and global feature vector of the target; Perform feature fusion on the local feature vector and the global feature vector to generate the fused feature vector of the target; The target is roughly classified based on the fusion feature vector and shape feature parameters to obtain the rough classification label information; All information in the first labeled data is integrated with the fused feature vector of the target and the coarse classification label information to generate the second labeled data.

[0011] By adopting the above technical solution, targeted feature extraction is achieved by calling the classification model corresponding to the target category in the pre-trained deep learning model library. The extraction of local feature vectors and global feature vectors comprehensively captures the detailed features and overall features of the target. The feature fusion mechanism organically combines the two types of feature information to enhance the feature expression capability. The combined use of fused feature vectors and shape feature parameters improves the accuracy of coarse classification and reduces classification errors caused by single features. The integration of the first labeled data with the fused feature vector and coarse classification label information constructs a more complete feature description system. The comprehensive use of this multi-dimensional feature improves the robustness of target classification and enhances the reliability of the classification results.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, after outputting the retrieved display images and the retrieved video clips in descending order of similarity, the method further includes: Map the retrieval display images on the time axis and extract the timestamp and spatial coordinate information of the target in each retrieval display image; Performing temporal clustering on the retrieved display images based on timestamps, and classifying the retrieved display images whose time interval is less than a first preset threshold into the same time window; Calculate the spatial distribution density of the target in the retrieved display image within each time window. The spatial distribution density is determined based on the spatial clustering degree of the target's appearance position. Performing spatial clustering on the retrieval display images within each time window according to the spatial distribution density, and marking the retrieval display images whose spatial distance is less than a second preset threshold as an associated image group; Extracting the motion trajectory features of the target in the associated image group; Perform hotspot analysis on the monitored area based on motion trajectory characteristics, generate an activity heat map of the target, and mark high-frequency activity areas in the heat map.

[0013] By employing the above technical solution, the spatiotemporal distribution patterns of targets are revealed through the mapping analysis of timestamp and spatial coordinate information. Temporal clustering categorizes search results with similar time periods, and spatial distribution density calculation reflects the spatial clustering characteristics of targets. The associated image groups formed by spatial clustering demonstrate the continuous activity characteristics of the target, and the extraction of motion trajectory features depicts the target's movement pattern. The activity heat map generated based on motion trajectory features intuitively displays the target's activity patterns, and the marking of high-frequency activity areas highlights key areas of focus. This spatiotemporal analysis method deepens the understanding of target behavior patterns, enabling the monitoring system to better grasp the target's activity patterns and enhance the depth of monitoring analysis.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, calculating the spatial distribution density of objects in the retrieved display image within each time window specifically includes: The monitoring area is divided into several grid cells, and the size of each grid cell is determined based on the average size of the target; Count the frequency of target occurrence in each grid unit and generate a target occurrence frequency matrix; Perform Gaussian smoothing on the target frequency matrix to obtain a smoothed frequency matrix; The normalized frequency value of each grid cell is calculated based on the smoothed frequency matrix. The normalized frequency value is obtained by dividing the smoothed frequency of the grid cell by the total frequency in the time window. The normalized frequency value is the spatial distribution density.

[0015] By adopting the above technical solution, a spatial analysis framework adapted to the target scale was established by dividing the monitoring area into grid cells based on the average target size. The original spatial distribution characteristics were constructed by statistics on the frequency of target occurrence in each grid cell. Gaussian smoothing eliminated the local noise and discreteness of the frequency distribution, making the spatial distribution representation more continuous and stable. The calculation method of normalized frequency values eliminated the influence of the difference in the number of samples in different time windows and obtained a comparable standardized spatial distribution density index. This adaptive grid division combined with smoothing and normalization method not only retains the spatial detail characteristics of the target distribution, but also provides a reliable density quantification index, so that the spatial aggregation characteristics of the target are accurately expressed numerically, effectively reflecting the frequency of target activities in different areas.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, extracting motion trajectory features of a target in an associated image group specifically includes: Sort the retrieved and displayed images in the associated image group in chronological order according to timestamps; Calculate the position offset of the target in the adjacent retrieval display images to obtain the target displacement vector; Calculating the instantaneous velocity of the target based on the target displacement vector and the corresponding time interval; Perform differential operation on the instantaneous velocity to obtain the acceleration change of the target; The target displacement vector, instantaneous velocity and acceleration changes are integrated into the target's motion trajectory characteristics.

[0017] By adopting the above technical solution, the time-series sorting based on timestamps ensures the temporal consistency of motion trajectory feature extraction, and the displacement information of the target movement is obtained by calculating the position offset of the target in the adjacent retrieval display images. The instantaneous velocity calculated by combining the displacement vector and the time interval reflects the motion state of the target, and the acceleration change obtained by performing a differential operation on the instantaneous velocity reveals the dynamic characteristics of the target during the motion process. The displacement vector, instantaneous velocity and acceleration change are integrated into motion trajectory features to construct a complete description system of the target's motion behavior. This multi-dimensional motion feature extraction method not only describes the spatial position changes of the target, but also characterizes the speed changes and acceleration characteristics during the motion process, so that the target's motion behavior characteristics are comprehensively characterized and can effectively distinguish different types of motion patterns.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, marking high-frequency activity areas in a heat map specifically includes: Calculate the heat value of each pixel in the heat map, which represents the frequency of the target activity; Identify connected areas with heat values greater than a preset heat threshold as candidate high-frequency activity areas; Calculate the area and duration of each candidate high-frequency activity region; Candidate high-frequency activity areas are screened based on area and duration, and areas that meet preset conditions are marked as high-frequency activity areas.

[0019] By adopting the above technical solution, the spatial distribution intensity of the target activity is quantitatively expressed by calculating the heat value of each pixel in the heat map. The candidate high-frequency activity areas obtained by screening with a preset heat threshold highlight the key areas where the target activity is frequent. The area and duration of the candidate areas are calculated to provide the spatial range and time scale characteristics of the activity areas. The candidate areas are screened according to area and duration, and the identified high-frequency activity areas have significant spatiotemporal clustering. This multiple screening mechanism based on heat value, regional area and duration eliminates the interference of short-term or local random activities, accurately marks the key areas of the target's continuous high-frequency activities, and reliably identifies the spatial distribution patterns of the target activities in the monitoring area.

[0020] In second aspect, an embodiment of the present application provides a system for quickly searching and locating targets in surveillance videos, the system for quickly searching and locating targets in surveillance videos comprising: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code comprises computer instructions, and the one or more processors call the computer instructions to enable the system to execute the method described in the first aspect and any possible implementation of the first aspect.

[0021] In a third aspect, an embodiment of the present application provides a computer-readable storage medium comprising instructions, which, when executed on a system, enables the system to execute the method described in the first aspect and any possible implementation of the first aspect.

[0022] In a fourth aspect, an embodiment of the present application provides a computer program product, which, when executed on a system, enables the system to execute the method described in any possible implementation manner in the first aspect.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. This application provides a method for quickly searching and locating targets in surveillance videos. By constructing a hierarchical labeling system and feature extraction mechanism, the target retrieval process is made more accurate and efficient. The large-scale labeling establishes the basic classification framework, and the coarse classification and fine classification gradually improve the labeling accuracy. The index structure formed by feature point extraction and association with multi-level labeling data supports rapid positioning. The introduction of feature saliency takes into account key factors such as the target orientation angle, ambient light intensity, and screen ratio, ensuring the quality of the retrieved and displayed images. The merging mechanism of preset length video segments retains the complete behavioral characteristics of the target. The dual-mode support of text retrieval and image retrieval enhances the flexibility of the retrieval method, and the setting of the similarity threshold ensures the accuracy of the retrieval results. The output method of sorting the retrieval results by similarity improves the retrieval efficiency, allowing users to quickly locate the most relevant target information. This retrieval method of multi-level labeling and feature association significantly improves the accuracy and efficiency of locating specific targets in massive surveillance data.

[0024] 2. This application provides a method for quickly searching and locating targets in surveillance videos. Through the mapping analysis of timestamp and spatial coordinate information, the spatiotemporal distribution pattern of the target is revealed. Time series clustering classifies retrieval results with similar time, and the calculation of spatial distribution density reflects the spatial aggregation characteristics of the target. The associated image group formed by spatial clustering shows the continuous activity characteristics of the target, and the extraction of motion trajectory features depicts the movement pattern of the target. The activity heat map generated based on the motion trajectory features intuitively shows the activity pattern of the target, and the marking of high-frequency activity areas highlights the key areas of focus. This spatiotemporal analysis method deepens the understanding of the target behavior pattern, enables the monitoring system to better grasp the target activity pattern, and enhances the depth of monitoring analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a method for quickly searching and locating a target in a surveillance video according to an embodiment of the present application.

[0026] Figure 2 This is another flowchart of a method for quickly searching and locating a target in a surveillance video in an embodiment of the present application.

[0027] Figure 3 This is a schematic diagram of the physical device structure of a system for quickly searching and locating targets in surveillance videos provided by an embodiment of the present application. DETAILED DESCRIPTION

[0028] The terms used in the following examples of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and appended claims of this application, the singular expressions "a," "an," "said," "above," "the," and "this" are intended to include plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in this application refers to any or all possible combinations comprising one or more of the listed items.

[0029] In the following, the terms "first" and "second" are used for descriptive purposes only and should not be understood to imply or suggest relative importance or implicitly indicate the number of the technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of this application, unless otherwise specified, "plurality" means two or more.

[0030] The following uses an embodiment and combines Figure 1 , a method for quickly searching and locating a target in a surveillance video according to an embodiment of the present application is described: See also Figure 1 , which is a flow chart of a method for quickly searching and locating a target in a surveillance video in an embodiment of the present application.

[0031] S101, marking targets in key frames of surveillance video into broad categories to generate first marking data; The system labels the targets in the key frames of the surveillance video into broad categories and generates first label data, which specifically includes: Use deep learning network to detect targets on key frames and obtain the target's rectangular box coordinates and confidence level; When the confidence level is greater than a preset threshold, the target area is extracted based on the coordinates of the rectangular box; Perform semantic segmentation on the target area to obtain the contour boundary of the target; Calculate the target's shape feature parameters based on its contour boundary, the shape feature parameters including the target's aspect ratio, area ratio and contour complexity; A decision tree is constructed based on shape feature parameters to classify the target into one of human, animal, car, and aircraft, obtaining the target's general category labeling information; The rectangular frame coordinates, confidence level, contour boundary, shape feature parameters and major category label information of the target are integrated to generate first label data.

[0032] In this step, the system labels the objects in the keyframes of the surveillance video into broad categories, generating the first labeled data. Keyframes are representative, independent frames in a video. By analyzing keyframes, we can obtain the main content information of the video. Object broad category labeling involves dividing the objects in the surveillance video into several broad categories, such as people, vehicles, and animals, to facilitate subsequent classification and retrieval.

[0033] The system can use a deep learning network to perform target detection on key frames and obtain the target's rectangular frame coordinates and confidence. Deep learning networks such as YOLO, SSD, etc. can quickly and accurately detect targets in images. Then, the system can extract the target area based on the target's rectangular frame coordinates, and perform semantic segmentation on the target area to obtain the target's contour boundary. Semantic segmentation can be implemented using models such as FCN and U-Net. Next, the system calculates the target's shape feature parameters, such as aspect ratio, area ratio, and contour complexity, based on the target's contour boundary, and constructs a decision tree based on the shape feature parameters to classify the target into one of humans, animals, cars, and aircraft, and obtain the target's major category labeling information. Finally, the system integrates the target's rectangular frame coordinates, confidence, contour boundary, shape feature parameters, and major category labeling information to generate the first labeling data.

[0034] S102, performing rough classification and labeling on each major category of objects based on the first labeling data to generate second labeling data; The system roughly classifies and labels the targets of each major category based on the first labeled data to generate second labeled data, which specifically includes: According to the major category label information in the first labeled data, calling the classification model corresponding to the target major category from the pre-trained deep learning model library; Extract the target image area from the original image based on the target's rectangular frame coordinates and contour boundaries; The classification model is used to extract features from the target image area to obtain the local feature vector and global feature vector of the target; Perform feature fusion on the local feature vector and the global feature vector to generate the fused feature vector of the target; The target is roughly classified based on the fusion feature vector and shape feature parameters to obtain the rough classification label information; All information in the first labeled data is integrated with the fused feature vector of the target and the coarse classification label information to generate the second labeled data.

[0035] In this step, the system performs a coarse classification on each broad category of objects based on the first labeled data, generating the second labeled data. Coarse classification involves further dividing the objects into more detailed subcategories based on the broad categories. For example, people can be divided into pedestrians and cyclists, and cars can be divided into cars and trucks.

[0036] The system first calls the classification model corresponding to the target category from the pre-trained deep learning model library based on the major category label information in the first labeled data. Then, based on the target's rectangular frame coordinates and contour boundaries, the system extracts the target image area from the original image, and uses the classification model to extract features from the target image area to obtain the local feature vector and global feature vector of the target. The local feature vector can represent the local detail information of the target, and the global feature vector can represent the overall shape and texture information of the target. Next, the system fuses the local feature vector and the global feature vector to generate a fused feature vector of the target, and performs coarse classification on the target based on the fused feature vector and shape feature parameters to obtain coarse classification label information. Finally, the system integrates all the information in the first labeled data with the target's fused feature vector and coarse classification label information to generate the second labeled data.

[0037] S103, performing fine classification labeling on the coarsely classified objects according to the second labeling data to generate third labeling data; In this step, the system uses the second labeled data to further classify the coarsely classified objects into more detailed categories, generating third labeled data. Fine classification refers to further dividing the objects into more detailed categories based on the coarse classification. For example, pedestrians can be divided into male, female, and children, and cars can be divided into sedans and SUVs.

[0038] The system uses a similar approach to coarse classification, calling a fine-classification model corresponding to the target's coarse classification from a pre-trained deep learning model library. It then extracts features and classifies the target image region to generate fine-classification labeling information. The system can also combine high-level semantic information, such as the target's motion and interaction information, to further refine the target's category. For example, by analyzing a person's motion trajectory and interaction behavior, pedestrians can be subdivided into categories such as ordinary pedestrians, athletes, and people who have fallen. Finally, the system integrates all information in the second labeled data with the target's fine-classification labeling information to generate third labeled data.

[0039] During the sub-classification process, sample imbalance may occur, meaning that some sub-classifications have fewer samples, leading to classifier overfitting. To address this issue, the system can employ data augmentation methods, increasing the amount of data for the less-sampled categories by performing transformations such as rotation, translation, and scaling. Furthermore, the system can incorporate transfer learning methods, leveraging labeled data from other scenarios to assist in classifier training for the target scenario, alleviating the problem of insufficient samples.

[0040] S104, extracting feature points of each marked object, and associating the feature points with the first mark data, the second mark data, and the third mark data to generate a hierarchical association index; In this step, the system extracts feature points for each labeled object and associates them with the first, second, and third labeled data to generate a hierarchical association index. Feature points are key points that represent the characteristics of an object, such as a person's head or hand, or a car's headlights or tires.

[0041] The system can use feature point detection algorithms such as SIFT and SURF to extract stable feature points from the target image area. Then, the system can use feature descriptors such as HOG, HOF, etc. to describe the feature points and obtain the feature vectors of the feature points. Next, the system associates the coordinates and feature vectors of the feature points with the first label data, the second label data, and the third label data to generate a hierarchical association index. The hierarchical association index is a multi-level tree structure. The top-level nodes represent the major categories of the targets, the middle-level nodes represent the coarse classification of the targets, the bottom-level nodes represent the fine classification of the targets, and the leaf nodes store the feature point information of the targets. Through the hierarchical association index, rapid retrieval and positioning of the targets can be achieved.

[0042] S105, calculating the feature saliency of each target in the key frames with the same feature value in the surveillance video; The system calculates the feature saliency of each target in the key frames with the same feature value in the surveillance video. The feature saliency is determined based on the weighted score of the target's orientation angle, ambient light intensity, and the target's proportion in the picture.

[0043] In this step, the system calculates the feature saliency of each target in keyframes with the same feature value in the surveillance video. Feature saliency indicates the prominence of the target feature and can reflect the importance of the target in the image.

[0044] The system can calculate the target's feature salience by considering multiple factors, such as the target's orientation angle, ambient lighting intensity, and the target's proportion within the frame. The more forward the orientation angle, the more moderate the lighting intensity, and the larger the target's proportion, the higher the feature salience. The system can design a weighted summation function to calculate the target's feature salience by combining the scores of each factor.

[0045] During feature saliency calculation, some factors may be difficult to accurately capture. For example, the target's orientation requires estimating head pose, and ambient lighting requires detecting the light source's position. To address this, the system can employ an end-to-end deep learning-based approach to learn the target's feature saliency directly from the image, eliminating the need to explicitly calculate scores for each factor. Furthermore, the system can incorporate an attention mechanism to automatically adjust the feature saliency weights for each target by learning the image's salient regions, improving calculation accuracy.

[0046] S106, selecting the key frame with the highest feature significance as the retrieval display image; In this step, the system selects the keyframe with the highest feature significance as the retrieval display image. The retrieval display image refers to the representative image displayed to the user in the search results, which can intuitively reflect the characteristics of the target.

[0047] The system compares the feature saliency of all keyframes related to the target in the surveillance video and selects the keyframe with the highest feature saliency as the retrieval display image. This ensures that the displayed image has the best visual quality and feature expression.

[0048] When selecting images for search display, the system also considers the image's aesthetics and integrity, avoiding blurry, dark, or incomplete images. Furthermore, to increase the diversity of search display, the system can select multiple keyframes with high feature significance as search display images, showcasing the target's features at different times and angles.

[0049] S107: Taking the retrieval display image as a reference point, taking the video segments of preset lengths forward and backward and merging them to generate a retrieval video segment; In this step, the system uses the search display image as a reference point, extracts and merges video segments of preset durations forward and backward, and generates a search video clip. The search video clip is the video clip displayed to the user in the search results, dynamically displaying the target's movement trajectory and interactive behavior.

[0050] The system first determines the timestamp of the image being retrieved in the surveillance video. It then extracts video segments of a preset length (e.g., 5 seconds) forward and backward, centered around that timestamp. The system then concatenates the two segments to create a complete retrieval video clip. To ensure video continuity and fluidity, the system can create smooth transitions at the boundaries of the segments, using techniques such as gradient effects or keyframe interpolation.

[0051] When generating retrieval video segments, the system may encounter incomplete or occluded target trajectories, impacting video quality. To address this issue, the system can employ a trajectory completion method to automatically fill in the missing target portions of the video segment by tracking and predicting the target's trajectory. Furthermore, the system can incorporate multi-target tracking technology to simultaneously track multiple targets in the video, preventing retrieval failures caused by occlusion or disappearance of a single target.

[0052] S108: When the search input is text, a feature identifier having a similarity with the text greater than a preset similarity threshold is retrieved through the hierarchical association index to generate a first matching result; In this step, when the user inputs a search request in text form, the system searches for feature identifiers whose similarity with the text is greater than a preset similarity threshold through the hierarchical association index to generate a first matching result.

[0053] The system first performs a semantic analysis on the text entered by the user, extracting keywords and subject terms from the text. The system then calculates similarity between the keywords and subject terms and the feature identifiers of each node in the hierarchical association index, using algorithms such as TF-IDF and Word2Vec. The system then selects the feature identifier with a similarity greater than a preset threshold (e.g., 0.8) as the first matching result. This first matching result contains target categories and feature point information that are semantically relevant to the text entered by the user.

[0054] During text retrieval, users may enter text that is too simple or too complex, resulting in inaccurate or insufficient search results. To address this issue, the system can employ a semantic retrieval method based on knowledge graphs. By constructing a domain knowledge graph, it expands the semantic information of the user's input text, improving the accuracy and richness of the search. Furthermore, the system can incorporate a user feedback mechanism to dynamically adjust the parameters and strategies of the retrieval algorithm based on user evaluation and selection of search results, continuously optimizing search results.

[0055] S109: When the search input is a picture, extract the feature value of the picture and calculate the similarity with the feature value in the feature identifier library, select the feature identifier with a similarity greater than a preset similarity threshold, and generate a second matching result.

[0056] This step details the specific operation process performed by the system when the user searches in the form of an image. First, the system extracts features from the input image to obtain the feature value of the image. The feature value may include a combination of multiple visual features such as the color histogram, texture features, shape features, etc. of the image. Then, the system calculates the similarity between the extracted image feature value and the feature value stored in the pre-built feature identification library. The feature identification library stores the feature values of a large number of pre-labeled and classified images, and each feature value corresponds to a specific semantic label or category identifier. The system calculates the similarity between the feature value of the input image and each feature value in the library to find the candidate feature identifier that is most similar to the input image. Finally, the system selects the feature identifier whose similarity is greater than the preset threshold as the matching result of the retrieval and generates a second matching result.

[0057] In the implementation of this step, the system can use a variety of feature extraction and similarity calculation algorithms. For feature extraction, traditional hand-designed features can be used, such as local feature descriptors such as SIFT and SURF, or global feature descriptors such as color histograms and LBP. Deep learning methods can also be used to extract features from images using pre-trained convolutional neural networks (such as VGG, ResNet, etc.) to obtain high-level semantic features. For similarity calculation, traditional metrics such as Euclidean distance and cosine similarity can be used, or learned metrics can be used to train a metric learning network to learn the similarity measurement function between image features. In addition, to improve retrieval efficiency, the system can also optimize the feature identification library, such as using inverted indexing, hash indexing and other technologies to accelerate the search and matching process of similar features.

[0058] In the above embodiment, by constructing a hierarchical labeling system and feature extraction mechanism, the target retrieval process is made more accurate and efficient. The large-scale labeling establishes the basic classification framework, the coarse classification and fine classification gradually improve the labeling accuracy, and the index structure formed by feature point extraction and association with multi-level labeling data supports rapid positioning. The introduction of feature saliency takes into account key factors such as the target orientation angle, ambient light intensity and screen ratio, ensuring the quality of the retrieved and displayed pictures. The merging mechanism of preset length video segments retains the complete behavioral characteristics of the target. The dual-mode support of text retrieval and image retrieval enhances the flexibility of the retrieval method, and the setting of the similarity threshold ensures the accuracy of the retrieval results. The output method of sorting the retrieval results by similarity improves the retrieval efficiency, allowing users to quickly locate the most relevant target information. This retrieval method of multi-level labeling and feature association significantly improves the accuracy and efficiency of locating specific targets in massive monitoring data.

[0059] In the above embodiment, the rapid positioning of the target and the accurate acquisition of the video clips are achieved through the hierarchical labeling and dual-modal retrieval. However, in order to more deeply analyze the activity patterns and spatial distribution characteristics of the target, the system can further perform spatiotemporal analysis on the retrieval results after completing the basic retrieval function. Figure 2 , another method for quickly searching and locating a target in a surveillance video in an embodiment of the present application is described: See also Figure 2 , is another flow chart of a method for quickly searching and locating a target in a surveillance video in an embodiment of the present application.

[0060] S201, mapping the search display images on the time axis, and extracting the timestamp and spatial coordinate information of the target in each search display image; This step further analyzes the retrieved display images in terms of their spatiotemporal dimensions. The system associates each retrieved display image with its corresponding timestamp, forming a timeline mapping. This mapping allows the system to clearly visualize the target's appearance at different points in time. The system also extracts the spatial coordinates of the target in each retrieved display image, recording its specific location within the monitored area. These timestamps and spatial coordinates serve as the foundational data for subsequent spatiotemporal analysis.

[0061] To implement this step, the system leverages image processing and computer vision techniques to automatically identify and locate objects within the retrieved display images. Using object detection and tracking algorithms, the system accurately extracts the bounding box coordinates of the objects. Furthermore, the system integrates the monitoring system's time synchronization mechanism to associate an accurate timestamp with each retrieved display image. By combining timestamp and spatial coordinate information with the retrieved display images, the system constructs a spatiotemporal mapping data structure, laying the foundation for subsequent analysis.

[0062] S202: performing temporal clustering on the retrieved display images based on the timestamps, and grouping the retrieved display images whose time interval is less than a first preset threshold into the same time window; This step processes the extracted timestamp information and groups temporally close retrieved display images into the same time window. By setting a time interval threshold, the system can adaptively adjust the size of the time window, ensuring that images within each time window are sufficiently close in time while maintaining sufficient temporal variance between windows. This temporal clustering approach can divide the target's continuous activity segments into several discrete time windows, facilitating subsequent analysis of the target's behavioral patterns over different time periods.

[0063] To implement this step, the system can dynamically generate time windows using a sliding window approach. By setting a fixed time step and window size, the system can continuously slide along the timeline, generating a series of continuous time windows. For each time window, the system calculates the timestamp differences between the images within it and groups those images together when the differences are less than a preset threshold. By adjusting the time step and window size, the system can control the number and granularity of time windows to suit different analysis needs.

[0064] S203, calculating the spatial distribution density of the objects in the retrieved display image within each time window; The system calculates the spatial distribution density of the objects in the retrieved display images within each time window. The spatial distribution density is determined based on the spatial clustering of the objects' locations, specifically including: The monitoring area is divided into several grid cells, and the size of each grid cell is determined based on the average size of the target; Count the frequency of target occurrence in each grid unit and generate a target occurrence frequency matrix; Perform Gaussian smoothing on the target frequency matrix to obtain a smoothed frequency matrix; The normalized frequency value of each grid cell is calculated based on the smoothed frequency matrix. The normalized frequency value is obtained by dividing the smoothed frequency of the grid cell by the total frequency in the time window. The normalized frequency value is the spatial distribution density.

[0065] This step performs spatial analysis on the retrieved display images within each time window and calculates the spatial distribution density of the target within the monitored area. Spatial distribution density reflects the frequency of target appearance at different locations and can be used to identify target activity hotspots and clusters. By calculating spatial distribution density, the system can quantify the spatial patterns of target activity, providing important evidence for subsequent behavioral analysis and anomaly detection.

[0066] When implementing this step, the system first divides the monitoring area into several grid cells. The size of each grid cell can be determined based on the average size of the target to ensure that the scale of the grid cell is adapted to the target's activity range. Then, the system counts the frequency of target appearance in each grid cell and generates a target appearance frequency matrix. In order to smooth the frequency matrix, the system can apply a Gaussian smoothing filter to it to eliminate the influence of noise and local outliers. Finally, the system calculates the normalized frequency value of each grid cell, that is, divides the smoothed frequency of the grid cell by the total frequency in the time window to obtain the spatial distribution density. The normalization process makes the spatial distribution density in different time windows comparable.

[0067] S204: spatially cluster the search display images within each time window according to spatial distribution density, and mark the search display images whose spatial distance is less than a second preset threshold as a related image group; After calculating the spatial distribution density, this step further spatially clusters the searched and displayed images within each time window. By setting a spatial distance threshold, the system can group spatially similar images into a group of related images. This spatial clustering approach can identify continuous segments of a target's spatial activity, forming a series of related image groups, each representing the target's spatial trajectory over a specific period of time.

[0068] When implementing this step, the system can use spatial clustering algorithms, such as density-based clustering (DBSCAN) or hierarchical clustering, to cluster the retrieved display images. By calculating the spatial distance matrix between images, the system can identify spatially adjacent images and group them into the same cluster. The spatial distance between images within a cluster is less than a preset threshold, while the spatial distance between images between clusters is greater than the threshold. By adjusting the spatial distance threshold, the system can control the granularity and compactness of the clustering to meet different analysis needs. For each generated cluster, the system marks it as a group of associated images, representing a segment of the target's spatial trajectory over a period of time.

[0069] S205, extracting motion trajectory features of the target in the associated image group; The system extracts the motion trajectory features of the target in the associated image group, specifically including: sorting the retrieval display images in the associated image group in time sequence according to the timestamp; calculating the position offset of the target in adjacent retrieval display images to obtain the target displacement vector; calculating the instantaneous speed of the target based on the target displacement vector and the corresponding time interval; performing differential operation on the instantaneous speed to obtain the acceleration change of the target; and integrating the target displacement vector, instantaneous speed and acceleration change into the motion trajectory features of the target.

[0070] This step further analyzes each associated image group to extract the target's motion trajectory features within that group. Motion trajectory features describe the target's spatial movement patterns and dynamic changes, including information such as displacement, velocity, and acceleration. By extracting motion trajectory features, the system can characterize the target's movement patterns over different time periods, providing important evidence for subsequent behavior recognition and anomaly detection.

[0071] When implementing this step, the system first sorts the retrieved display images in the associated image group according to the timestamp to obtain a temporally continuous sequence of images. Then, the system calculates the position offset of the target in adjacent images to obtain a series of target displacement vectors. The displacement vector reflects the spatial movement of the target between two consecutive frames of images. Based on the displacement vector and the corresponding time interval, the system can further calculate the instantaneous velocity of the target, which represents the movement rate of the target at each time point. Finally, the system performs a differential operation on the instantaneous velocity to obtain the acceleration change of the target, reflecting the dynamic change trend of the target's motion state. By integrating the displacement vector, instantaneous velocity and acceleration change together, the system obtains the complete target motion trajectory characteristics.

[0072] S206: Perform hot spot analysis on the monitored area based on the motion trajectory characteristics, generate an activity heat map of the target, and mark high-frequency activity areas in the heat map.

[0073] The system performs hotspot analysis on the monitored area based on motion trajectory features, generates an activity heat map of the target, and marks high-frequency activity areas in the heat map. Specifically, it includes: calculating the heat value of each pixel in the heat map, which indicates the frequency of the target's activity; identifying connected areas with heat values greater than a preset heat threshold as candidate high-frequency activity areas; calculating the area and duration of each candidate high-frequency activity area; screening the candidate high-frequency activity areas based on area and duration, and marking areas that meet the preset conditions as high-frequency activity areas. This step uses the extracted motion trajectory features to perform hotspot analysis on the entire monitored area and generate an activity heat map of the target. The activity heat map uses color coding to intuitively display the frequency of the target's activities in different areas. By generating an activity heat map, the system can identify the high-frequency areas and hotspot distribution of target activities in the monitored area, providing an important reference for optimizing monitoring layout, resource scheduling, and abnormal event detection.

[0074] To implement this step, the system first divides the monitored area into a grid, with each grid cell corresponding to a pixel on the heat map. The system then calculates the heat value of each pixel, which indicates the frequency of the target's appearance and activity within the area. The system comprehensively assesses the activity heat of each area by combining factors such as the target's spatial distribution density, motion trajectory characteristics, and duration. The higher the heat value, the more frequent the target's activity within the area. By color-mapping the heat values, the system generates an intuitive activity heat map, with different colors representing varying degrees of activity frequency. Finally, the system analyzes the activity heat map and identifies high-frequency activity areas. By setting a heat threshold, the system extracts connected areas with high heat values as candidate high-frequency activity areas. The system then calculates the area and duration of each candidate area, screens the candidate areas based on preset conditions, and ultimately marks the true high-frequency activity areas.

[0075] In the above embodiment, the spatiotemporal distribution pattern of the target is revealed through the mapping analysis of timestamp and spatial coordinate information. Time series clustering classifies retrieval results with similar time, and the calculation of spatial distribution density reflects the spatial aggregation characteristics of the target. The associated image group formed by spatial clustering shows the continuous activity characteristics of the target, and the extraction of motion trajectory features depicts the movement pattern of the target. The activity heat map generated based on the motion trajectory features intuitively shows the activity pattern of the target, and the marking of high-frequency activity areas highlights the key areas of focus. This spatiotemporal analysis method deepens the understanding of the target behavior pattern, enables the monitoring system to better grasp the target activity pattern, and enhances the depth of monitoring analysis.

[0076] The following describes the system in the embodiment of the present invention from the perspective of hardware processing. Figure 3, which is a schematic diagram of the physical device structure of a system for quickly searching and locating targets in surveillance videos provided in an embodiment of the present application.

[0077] It should be noted that Figure 3 The structure of the system shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0078] like Figure 3 As shown, the system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes, such as the methods described in the above embodiments, based on programs stored in a read-only memory (ROM) 302 or programs loaded from a storage unit 308 into a random access memory (RAM) 303. RAM 303 also stores various programs and data required for system operation. CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to bus 304.

[0079] The following components are connected to the I / O interface 305: an input section 306 including a camera, infrared sensor, and the like; an output section 307 including a liquid crystal display (LCD) and speakers; a storage section 308 including a hard disk and the like; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. Removable media 311, such as a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, is installed in the drive 310 as needed, so that computer programs read from the media can be installed in the storage section 308 as needed.

[0080] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for executing the methods illustrated in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 309 and / or installed from removable media 311. When executed by the central processing unit (CPU) 301, the computer program performs the various functions defined in the present invention.

[0081] It should be noted that the computer-readable medium described in the embodiments of the present invention may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium may include a data signal transmitted in baseband or as part of a carrier wave, which carries a computer-readable computer program. Such a propagated data signal may take any of a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof.

[0082] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0083] As another aspect, the present invention further provides a computer-readable storage medium, which may be included in the system described in the above embodiments, or may exist independently and not incorporated into the system. The storage medium carries one or more computer programs, and when executed by a processor of a system, the system implements the methods provided in the above embodiments.

[0084] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0085] As used in the above embodiments, the term “when…” may be interpreted as “if…” or “after…” or “in response to determining…” or “in response to detecting…”, depending on the context. Similarly, the phrases “upon determining…” or “if (stated condition or event) is detected” may be interpreted as “if determining…” or “in response to determining…” or “upon detecting (stated condition or event)” or “in response to detecting (stated condition or event)”, depending on the context.

[0086] In the above embodiments, all or part of the embodiments can be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, hard disk, tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive).

[0087] Those skilled in the art will appreciate that all or part of the process steps in the above-described method embodiments can be implemented by a computer program instructing the relevant hardware. The program can be stored in a computer-readable storage medium, and when executed, the program can include the process steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for quickly searching and locating a target in a surveillance video, characterized in that: include: Marking the targets in the key frames of the surveillance video into broad categories to generate first labeled data; Performing rough classification and labeling on each major category of objects according to the first labeling data to generate second labeling data; Performing fine classification and labeling on the coarsely classified objects according to the second labeling data to generate third labeling data; Extracting feature points of each marked object, and associating the feature points with the first mark data, the second mark data, and the third mark data to generate a hierarchical association index; Calculating the feature saliency of each target in the key frames with the same feature value in the surveillance video, wherein the feature saliency is determined based on a weighted score of the target's orientation angle, ambient light intensity, and the target's proportion in the frame; Select the key frame with the highest feature significance as the retrieval display picture; Taking the retrieval display image as a reference point, taking the video segments of preset lengths forward and backward and merging them to generate a retrieval video clip; When the search input is text, the hierarchical association index is used to search for feature identifiers whose similarity to the text is greater than a preset similarity threshold, and a first matching result is generated; When the search input is an image, extracting the feature value of the image and calculating the similarity with the feature value in the feature identifier library, selecting the feature identifier with a similarity greater than the preset similarity threshold, and generating a second matching result; According to the first matching result or the second matching result, the corresponding retrieval display picture and the retrieval video clip are retrieved from the hierarchical association index, and the retrieval display picture and the retrieval video clip are output in descending order of similarity.

2. The method according to claim 1, characterized in that The step of marking the targets in the key frames of the surveillance video into broad categories to generate first marking data specifically includes: Using a deep learning network to perform target detection on the key frame to obtain the rectangular frame coordinates and confidence level of the target; When the confidence level is greater than a preset threshold, extracting the target area based on the rectangular frame coordinates; Performing semantic segmentation on the target area to obtain the contour boundary of the target; Calculating shape feature parameters of the target according to the contour boundary of the target, wherein the shape feature parameters include the aspect ratio, area ratio and contour complexity of the target; Building a decision tree based on the shape feature parameters to classify the target into one of a person, an animal, a car, and an aircraft, thereby obtaining broad category labeling information of the target; The rectangular frame coordinates of the target, the confidence level, the contour boundary, the shape feature parameters and the major category label information are integrated to generate first label data.

3. The method according to claim 1, characterized in that The step of roughly classifying and labeling the objects of each major category according to the first labeling data to generate second labeling data specifically includes: Based on the major category label information in the first label data, calling a classification model corresponding to the target major category from a pre-trained deep learning model library; Extracting a target image region from the original image based on the rectangular frame coordinates and contour boundaries of the target; Using the classification model to perform feature extraction on the target image area to obtain a local feature vector and a global feature vector of the target; Performing feature fusion on the local feature vector and the global feature vector to generate a fused feature vector of the target; Performing rough classification on the target based on the fused feature vector and the shape feature parameter to obtain rough classification label information; All information in the first labeled data is integrated with the fused feature vector of the target and the coarse classification label information to generate second labeled data.

4. The method according to claim 1, wherein After outputting the search display image and the search video clip in descending order of similarity, the method further includes: Mapping the search display pictures on the time axis, and extracting the timestamp and spatial coordinate information of the target in each search display picture; Performing temporal clustering on the retrieved display images based on the timestamps, and classifying the retrieved display images whose time interval is less than a first preset threshold into the same time window; Calculating the spatial distribution density of the objects in the retrieved display image within each of the time windows, wherein the spatial distribution density is determined based on the spatial clustering degree of the locations where the objects appear; Performing spatial clustering on the retrieved display pictures within each of the time windows according to the spatial distribution density, and marking the retrieved display pictures whose spatial distance is less than a second preset threshold as an associated picture group; Extracting motion trajectory features of the target in the associated image group; Hot spot analysis is performed on the monitored area based on the motion trajectory characteristics to generate an activity heat map of the target, and high-frequency activity areas are marked in the heat map.

5. The method according to claim 4, characterized in that The calculating of the spatial distribution density of the target in the retrieved display picture within each time window specifically includes: Dividing the monitoring area into a plurality of grid cells, wherein the size of each grid cell is determined based on the average size of the target; Counting the frequency of occurrence of the target in each grid unit to generate a target occurrence frequency matrix; Performing Gaussian smoothing on the target occurrence frequency matrix to obtain a smoothed frequency matrix; The normalized frequency value of each grid unit is calculated according to the smoothed frequency matrix, wherein the normalized frequency value is obtained by dividing the smoothed frequency of the grid unit by the total frequency in the time window, and the normalized frequency value is the spatial distribution density.

6. The method according to claim 4, characterized in that The extracting the motion trajectory features of the target in the associated picture group specifically includes: sorting the retrieved display pictures in the associated picture group in time sequence according to the timestamp; Calculating the position offset of the target in the adjacent search display image to obtain a target displacement vector; Calculating the instantaneous velocity of the target based on the target displacement vector and the corresponding time interval; performing a differential operation on the instantaneous velocity to obtain an acceleration change of the target; The target displacement vector, the instantaneous velocity and the acceleration change are integrated into the motion trajectory characteristics of the target.

7. The method according to claim 4, characterized in that Marking high-frequency activity areas in the heat map specifically includes: Calculating a heat value for each pixel in the heat map, where the heat value indicates the frequency of the target activity; Identify the connected areas whose heat value is greater than a preset heat threshold as candidate high-frequency activity areas; Calculating the area and duration of each candidate high-frequency activity region; The candidate high-frequency activity regions are screened according to the area and the duration, and regions meeting preset conditions are marked as high-frequency activity regions.

8. A system for quickly searching and locating targets in surveillance video, characterized in that: The system comprises: One or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the system to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium comprising instructions, characterized in that: When the instructions are executed on a system, the system is caused to perform the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that When the computer program product is run on a system, the system is caused to perform the method according to any one of claims 1 to 7.

Images