A drone security inspection method, system and device

Abstract

The present invention belongs to the field of public security management technology, and relates to a method, system and device for public security inspection using drones. In public security inspections, traditional methods are limited by the environment and have difficulty in accurately extracting human targets and processing dynamic distance relationships between people. The present invention uses drones to collect images of target inspection areas, uses feature extraction methods to construct grid images, and extracts human targets in combination with a collection of human target feature images; uses capsule neural networks to establish a human vector field to analyze distance relationships and track their changes, and then uses long and short-term memory network analysis and processing to obtain a constant distance interval; finally, it determines whether the distance relationship between people is normal and gives a prompt. The present invention can accurately identify people, dynamically analyze distance relationships and determine anomalies, effectively improving public security inspection efficiency and prevention and control capabilities.

Description

Technical Field

[0001] The present invention belongs to the technical field of public security management, and relates to a method, system and device for public security inspection by an unmanned aerial vehicle (UAV). Background Art

[0002] Compared to traditional patrol vehicles, helicopters, and other equipment, drones are relatively inexpensive to purchase and maintain. They also have a longer service life and lower operating costs. In urban security patrols, drones can be integrated with existing surveillance systems, reducing the need for additional equipment.

[0003] In complex environments and backgrounds, it is difficult for traditional drone inspection methods to accurately extract and locate human targets from images.

[0004] In dynamic environments, the relative positions of multiple people constantly change. Traditional static image analysis methods cannot effectively handle this dynamic change and tend to overlook abnormal conditions such as human disturbances. In real-world situations, it is necessary to promptly detect and report abnormal crowds or abnormal distances between people to avoid safety hazards. Summary of the Invention

[0005] In order to solve the problems existing in the background technology, the present invention proposes a drone security inspection method, system and device.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] In one aspect, the present invention provides a method for drone security inspection, comprising:

[0008] The drone collects images of the target inspection area and extracts human targets from the images of the target inspection area based on feature extraction methods;

[0009] The capsule neural network is used to analyze the distance relationship between people in the target inspection area image. The long short-term memory network is used to analyze the distance relationship between any two people and obtain the constant distance interval between the two people.

[0010] Based on the distance relationship between any human targets, it is determined whether the distance between the characters is within the normal range, and a prompt or no prompt is given based on the distance between the corresponding two human targets.

[0011] Furthermore, the specific method of extracting human targets from the target inspection area image based on the feature extraction method is:

[0012] The target inspection area image is constructed as a first grid image, where the side length of the grid is a preset pixel value, and the characteristic gradient change relationship between each grid in the first grid image and its eight neighboring grids is extracted, and the characteristic gradient change relationship of each grid is extracted to generate a second grid image;

[0013] Set a collection of pre-trained character target feature images, each character target feature image corresponds to a different action of the character target;

[0014] Each person target feature image in the person target feature image collection generates a window;

[0015] The window is set in the second grid image and shifted. The similarity between the image in the window and the human target feature image corresponding to the window is compared by Euclidean distance. The corresponding image in the window whose image similarity is higher than the similarity threshold is set as the human target.

[0016] Furthermore, the specific method of using the capsule neural network to analyze the distance relationship between people in the target inspection area image is as follows:

[0017] The capsule neural network establishes a vector field based on all the human targets in the target inspection area image. Each vector represents the distance between a certain human target and other human targets.

[0018] The capsule network tracks the change data of each vector in the vector field and records the distance relationship between all human targets in the target inspection area image.

[0019] Furthermore, the specific method for analyzing the distance relationship between any two person targets through the long short-term memory network is:

[0020] The long short-term memory network receives the distance relationship and changes between any two human targets analyzed by the capsule neural network;

[0021] The long short-term memory network determines the calculated value of the forget gate for each distance relationship by the probability of each distance relationship appearing; each distance relationship not processed by the forget gate is added to the input gate;

[0022] Based on the processing results of the forget gate and the input gate, the unit state of each current distance relationship is calculated. The specific formula is:

[0023] ;

[0024] in is the unit state at the previous moment, is the output of the forget gate, is the output of the input gate, is the candidate memory generated by the current input;

[0025] At each moment, the current output is generated based on the current cell state and the output gate. The current output contains the cell state of the previous moment.

[0026] Based on all possible outputs of the cell state of each distance relation, a constant distance interval corresponding to the distance relation is obtained.

[0027] In another aspect, the present invention provides a drone security inspection system that implements the above-mentioned drone security inspection method, including:

[0028] Data acquisition module, which collects images of the target inspection area through drones;

[0029] Feature extraction module, which extracts human targets from the target inspection area image based on feature extraction method;

[0030] The data processing module uses a capsule neural network to analyze the distance relationship between people in the target inspection area image, and uses a long short-term memory network to analyze the distance relationship between any two people, and obtains the constant distance interval between the two people.

[0031] The prompt module determines whether the distance between any human targets is within a normal range based on the distance relationship between them, and gives a prompt or not based on the distance between the two corresponding human targets.

[0032] On the other hand, the present invention provides a drone security patrol device, which includes a memory and a processor. The memory stores the above-mentioned drone security patrol system, and the processor calls the drone security patrol system in the memory when working.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention constructs a grid image and uses the feature gradient change relationship to enhance the significant features in the image. Combined with a pre-trained collection of human target feature images, the person in the image can be identified more accurately.

[0035] This invention uses a capsule neural network to establish a vector field between people and track the changing data of these vectors, thereby enabling real-time, dynamic analysis of interpersonal distance relationships. It also uses a long-short-term memory network to perform serialized analysis of these distance relationships, determine whether they fall within the normal range, and issue necessary prompts or warnings based on the analysis results. Furthermore, a probabilistic factor is incorporated into the long-short-term memory network to reduce the error rate of activation function judgments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the method of the present invention;

[0037] Figure 2It is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] like Figure 1-Figure 2 As shown, the technical solution adopted by the present invention is as follows: a drone security inspection method, comprising:

[0040] The drone collects images of the target inspection area and extracts human targets from the images of the target inspection area based on feature extraction methods;

[0041] The capsule neural network is used to analyze the distance relationship between people in the target inspection area image. The long short-term memory network is used to analyze the distance relationship between any two people and obtain the constant distance interval between the two people.

[0042] Based on the distance relationship between any human targets, it is determined whether the distance between the characters is within the normal range, and a prompt or no prompt is given based on the distance between the corresponding two human targets.

[0043] The drone captures images of the target inspection area and uses feature extraction methods to extract human subjects from these images. This capture provides the foundational data for subsequent person recognition and distance analysis. Image extraction is the first step in public security inspections, helping to monitor personnel and identify potential safety hazards or anomalies, such as identifying dangerous individuals or missing persons in public places.

[0044] The target inspection area image is constructed as a first grid image, with the grid side length being a preset pixel value. The characteristic gradient variation relationship between each grid in the first grid image and its eight neighboring grids is extracted, and the characteristic gradient variation relationship of each grid is extracted to generate a second grid image. By gridding the image and analyzing the characteristic gradient variation relationship between grids, local features in the image can be better captured, providing more detailed information for the extraction of human targets, and helping to accurately distinguish the characteristics represented by people in complex scenes.

[0045] Set up a collection of pre-trained person feature images, each corresponding to a different action. This pre-trained collection of person feature images covers a wide range of actions, allowing the system to recognize people in a variety of states. This enhances the diversity and accuracy of person recognition and avoids missed detections or misjudgments due to varying actions.

[0046] Each person target feature image in the person target feature image collection generates a window.

[0047] Each character feature image corresponds to the action state of a character. This allows for detailed matching of character features at different positions and angles, improving the ability to find and match characters and increasing the flexibility of character recognition.

[0048] Each person's target feature image generates a corresponding window. The window is set in the second grid image and shifted. The image in the window is compared with the corresponding person's target feature image using Euclidean distance. Images in the window with a similarity above a similarity threshold are set as person targets. This operation, using Euclidean distance comparison, precisely identifies areas with high similarity to person target features. The recognition results for these areas are then marked as person targets, ensuring more precise extraction of the target and improving the accuracy of person recognition.

[0049] Capsule neural networks are used to analyze the distance relationships between people in images of the target inspection area. This analysis helps understand the spatial distribution of people and their relative positions, providing important evidence for identifying unusual individuals. For example, if the distance between two people is shortened due to one person's sudden acceleration, there is a high risk of collision. Alternatively, if someone in a crowd slows down and moves in the opposite direction away from the group, there is a high risk of them getting lost. This can effectively identify potential risks in scenarios such as large-scale events or busy streets.

[0050] The capsule neural network builds a vector field based on all human targets in the target inspection area image. Each vector value represents the distance between a specific human target and other human targets. The vector relationship between a specific human target and other human targets represents the changing relationship between the distance between them, i.e., whether the distance between the specific human target and the other human targets remains unchanged, increases, or decreases. The vector field clearly and intuitively represents the distance relationship between human targets. This vector representation allows the distance relationship between any two target human targets to be locked, facilitating subsequent distance analysis and anomaly detection. By quantifying the distance relationship between human targets as a vector, it provides structured data for subsequent calculations and analysis.

[0051] The capsule network tracks the changing data of each vector in the vector field, recording the distance relationships between all human targets in the target inspection area image. Tracking vector changes allows us to understand the dynamic changes in distance between people, which is crucial for identifying the flow, gathering, or dispersion of people, and can promptly detect possible anomalies, such as sudden gatherings or separations of people.

[0052] By analyzing the distance relationship between any two human targets using a long-short-term memory network (LSTM), a constant distance interval is obtained between any two or more individuals. Using LSTM networks, the time series characteristics of distance relationships are considered. This not only focuses on the current distance relationship but also integrates historical information. This makes it more scientific to determine the long-term stability and normal range of the distance between any two individuals, improving the accuracy of abnormal situation judgments. The relationship between any two human targets can be determined solely based on their distance, which is sufficient for inspection tasks and reduces the computational complexity of subsequent analysis.

[0053] The LSTM network receives the distance relationship and changes between any two human targets analyzed by the capsule neural network. Integrating the data from the capsule neural network makes the distance analysis more comprehensive. The LSTM network's characteristics are used to process time series information, fully leveraging historical data for deeper analysis.

[0054] The LSTM network uses the probability of each distance relationship to determine the value calculated by the forget gate for each distance relationship. Each distance relationship not processed by the forget gate is added to the input gate. The forget gate and input gate mechanisms filter and select information, focusing on important distance relationships, improving analysis efficiency and accuracy, and avoiding interference from irrelevant information. Furthermore, the forget gate's probabilistic approach allows it to better reflect actual conditions, preventing the activation function from allowing an unrealistic result to pass through the activation function.

[0055] Based on the processing results of the forget gate and the input gate, the unit state of each current distance relationship is calculated. The specific formula is:

[0056] ;

[0057] in is the unit state at the previous moment, is the output of the forget gate, is the output of the input gate, is the candidate memory generated by the current input. By calculating the unit state through this formula, we can comprehensively consider the information at different times and obtain a more representative and reliable distance relationship state, providing an accurate basis for the subsequent constant distance interval calculation.

[0058] At each moment, the current output is generated based on the current cell state and the output gate. The current output includes the cell state at the previous moment. This output includes the cell state at the previous moment, which can more comprehensively reflect the time series characteristics of the distance relationship and help to more accurately determine whether the distance relationship is normal.

[0059] Based on all possible outputs of the cell state for each distance relationship, a constant distance interval for the corresponding distance relationship is derived. This constant distance interval provides a clear quantitative standard for determining whether the distance between people is normal. When the actual distance exceeds this interval, it is considered an abnormality, allowing the system to take appropriate measures such as prompts.

[0060] Based on the distance relationship between arbitrary human targets, the system determines whether the distance between them is within the normal range and issues a warning or non-warning based on the distance between the two corresponding human targets. By determining whether the distance between people is normal, it can promptly detect crowds or abnormal spacing, and issue corresponding warnings or non-warnings. This enables automated security patrol monitoring, improves patrol efficiency, and enhances security prevention and control capabilities. For example, it can promptly detect potential conflicts or illegal gatherings in crowded areas.

[0061] In another aspect, the present invention provides a drone security inspection system that implements the above-mentioned drone security inspection method, including:

[0062] The data acquisition module collects images of the target inspection area through drones.

[0063] The feature extraction module extracts human targets from the target inspection area image based on the feature extraction method.

[0064] The data processing module uses a capsule neural network to analyze the distance relationship between people in the target inspection area image, and uses a long short-term memory network to analyze the distance relationship between any two human targets to obtain a constant distance interval between the two human targets.

[0065] The prompt module determines whether the distance between any human targets is within a normal range based on the distance relationship between them, and gives a prompt or not based on the distance between the two corresponding human targets.

[0066] On the other hand, the present invention provides a drone security patrol device, which includes a memory and a processor. The memory stores the above-mentioned drone security patrol system, and the processor calls the drone security patrol system in the memory when working.

[0067] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A drone security inspection method, characterized in that: Includes: The drone collects images of the target inspection area and extracts human targets from the images of the target inspection area based on feature extraction methods; The capsule neural network is used to analyze the distance relationship between people in the target inspection area image. The long short-term memory network is used to analyze the distance relationship between any two people and obtain the constant distance interval between the two people. Based on the distance relationship between any two human targets, determine whether the distance between them is within the normal range, and give or take a prompt based on the distance between the two corresponding human targets; The specific method for analyzing the distance relationship between any two character targets through the long short-term memory network is: The long short-term memory network receives the distance relationship and changes between any two human targets analyzed by the capsule neural network; The long short-term memory network determines the calculated value of the forget gate for each distance relationship by the probability of each distance relationship appearing; each distance relationship not processed by the forget gate is added to the input gate; Based on the processing results of the forget gate and the input gate, the unit state of each current distance relationship is calculated. The specific formula is: ； in is the unit state at the previous moment, is the output of the forget gate, is the output of the input gate, is the candidate memory generated by the current input; At each moment, the current output is generated based on the current cell state and the output gate. The current output contains the cell state of the previous moment. Based on all possible outputs of the cell state of each distance relation, a constant distance interval corresponding to the distance relation is obtained.

2. The method for public security inspection using a drone according to claim 1, characterized in that: The specific method of extracting human targets from the target inspection area image based on the feature extraction method is: The target inspection area image is constructed as a first grid image, where the side length of the grid is a preset pixel value, and the characteristic gradient change relationship between each grid in the first grid image and its eight neighboring grids is extracted, and the characteristic gradient change relationship of each grid is extracted to generate a second grid image; Set a collection of pre-trained character target feature images, each character target feature image corresponds to a different action of the character target; Each person target feature image in the person target feature image collection generates a window; The window is set in the second grid image and shifted. The similarity between the image in the window and the human target feature image corresponding to the window is compared by Euclidean distance. The corresponding image in the window whose image similarity is higher than the similarity threshold is set as the human target.

3. The method for public security inspection using an unmanned aerial vehicle according to claim 1, characterized in that: The specific method of using capsule neural network to analyze the distance relationship between people in the target inspection area image is as follows: The capsule neural network establishes a vector field based on all the human targets in the target inspection area image. Each vector represents the distance between a certain human target and other human targets. The capsule network tracks the change data of each vector in the vector field and records the distance relationship between all human targets in the target inspection area image.

4. A drone security inspection system, characterized by: The method for executing the drone security inspection method according to claim 1 comprises: Data acquisition module, which collects images of the target inspection area through drones; Feature extraction module, which extracts human targets from the target inspection area image based on feature extraction method; The data processing module uses a capsule neural network to analyze the distance relationship between people in the target inspection area image, and uses a long short-term memory network to analyze the distance relationship between any two people, and obtains the constant distance interval between the two people. The prompt module determines whether the distance between any human targets is within a normal range based on the distance relationship between them, and gives a prompt or not based on the distance between the two corresponding human targets.

5. A drone security inspection device, comprising a memory and a processor, characterized in that: The memory stores the UAV security inspection system according to claim 4, and the processor calls the UAV security inspection system in the memory when working.

Images