Park intelligent video monitoring system and method based on deep learning

Abstract

The invention relates to the technical field of park intelligent management, and discloses a park intelligent video monitoring system and method based on deep learning, and the system comprises a data collection module and an intelligent management module. According to the park intelligent video monitoring system and method based on deep learning, a data collection module collects a data set, an intelligent management module establishes a two-dimensional coordinate system and marks a coordinate point corresponding to each monitoring device, whenever a strange person or a strange vehicle enters and exits from a park, whether the strange person or the strange vehicle continuously appears in a monitoring video of the monitoring device is judged, and the strange person or the strange vehicle enters and exits from the park. When an abnormal signal is generated, the intelligent management module analyzes and generates a delay time difference, then judges whether there is a strange person or a strange vehicle in a patrol video of the unmanned aerial vehicle equipment, identifies appearance characteristics and behavior characteristics through a deep learning model, is high in intelligent monitoring safety, and when an early warning signal is generated, the intelligent management module analyzes and generates a tracking mileage. And the unmanned aerial vehicle equipment with the shortest tracking mileage is screened for tracking, so that the efficient tracking capability is high.

Description

Technical Field

[0001] The present invention relates to the field of park intelligent management technology, and specifically to a park intelligent video surveillance system and method based on deep learning. Background Art

[0002] A deep learning-based intelligent video surveillance system for campuses utilizes artificial intelligence (AI) to improve monitoring efficiency and accuracy. The system captures real-time video streams through cameras and analyzes the video content using deep learning models to implement various intelligent monitoring functions. In industrial parks, science and technology parks, and other locations, intelligent video surveillance systems can provide 24-hour security monitoring, enabling timely detection and resolution of potential safety hazards. With the continuous development and improvement of deep learning technology, these intelligent video surveillance systems for campuses will become even more intelligent and precise. Future systems may integrate additional features, such as sentiment analysis and voice recognition, and operate stably in more complex environments. Furthermore, with the advancement of edge computing technology, the system's response speed will be further accelerated, significantly improving real-time performance. This will not only enhance campus security but also provide managers with powerful data support, helping them better operate and manage the campus.

[0003] Currently, traditional deep learning-based intelligent video surveillance systems for campuses have a relatively simple method of collecting video footage. Blind spots in monitoring can easily lead to major security loopholes in campus management. In addition, when suspicious strangers or vehicles are discovered, manual investigations take a long time to track them down, resulting in a high risk of misjudgment. Summary of the Invention

[0004] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides a campus intelligent video surveillance system and method based on deep learning, which has the advantages of high intelligent monitoring security and strong efficient tracking capabilities. It solves the problems that the traditional campus intelligent video surveillance system based on deep learning has a relatively single way of collecting video data, a long investigation time and a high risk of misjudgment in tracking.

[0005] (2) Technical solution To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a campus intelligent video surveillance system based on deep learning, comprising a data acquisition module and an intelligent management module; The data acquisition module is composed of a monitoring data unit, a patrol data unit and an instruction data unit. The monitoring data unit is connected to the monitoring device through the network to collect a monitoring data set. The monitoring data set includes the monitoring videos taken in real time by multiple monitoring devices in the park. The patrol data unit is connected to the drone equipment through the network to collect a patrol data set. The patrol data set includes patrol videos of all time points in the park. The instruction data unit is connected to the central control console through the network to collect an instruction data set. The instruction data set includes all control instructions entered by the park management personnel through the central control console. The data acquisition module transmits the monitoring data set, patrol data set and instruction data set to the intelligent management module through the network; The intelligent management module consists of an investigation and analysis unit, an abnormality assessment unit and an early warning management unit. The investigation and analysis unit generates a target trajectory based on the monitoring data set. The investigation and analysis unit is based on the target trajectory Determine whether the strangers or vehicles entering the park continue to appear in the surveillance video. If the strangers or vehicles entering the park do not continue to appear in the surveillance video, a corresponding abnormal signal is generated and transmitted to the abnormality assessment unit. The abnormality assessment unit analyzes and generates a delay time difference based on the patrol data set. , and then according to the delay time difference Determine whether there are strangers or strange vehicles in the patrol video shot by the drone equipment. If there are no strangers or strange vehicles in the patrol video shot by the drone equipment, generate a corresponding warning signal and transmit it to the warning management unit. The warning management unit obtains the deep learning model through the network connection central console , and combined with the instruction data set, analyze and generate tracking mileage , the deep learning model Based on tracking mileage , filter out tracked mileage The shortest drone equipment tracks strangers or vehicles.

[0006] Preferably, the expression of the monitoring data set is , to The corresponding surveillance videos are taken in real time by each monitoring device in the park. Indicates the area name corresponding to the monitoring device, 1 to Indicates the number of monitoring devices installed in the park indivual.

[0007] Preferably, the patrol data set is expressed as , to The patrol videos corresponding to each time point in the park, Indicates the time point when the drone device shoots the patrol video, 1 to Said that the patrol video captured by the drone equipment has Group.

[0008] Preferably, the expression of the instruction data set is , to Corresponding to the control instructions input by the park management personnel through the central control console each time, Indicates the time when the park manager inputs the control command, from 1 to Indicates that the control instructions entered by the park management personnel are Second-rate.

[0009] Preferably, the target trajectory The calculation process is as follows: Take the park entrance and exit as the origin Establish a two-dimensional coordinate system , and combined with the number of monitoring devices installed in the park in the monitoring data set in the two-dimensional coordinate system Set the corresponding number of coordinate points in Monitoring devices corresponding to coordinate points, and then The coordinate points are marked as ; Whenever a stranger or vehicle enters or leaves the park, it will be marked as ; ; In the formula, represents the target trajectory, Represents a two-dimensional coordinate system Middle coordinate points, Indicates based on The surveillance video corresponding to the coordinate point is used to determine whether the strangers or vehicles entering the park continue to appear in the surveillance video. If the strangers or vehicles entering the park continue to appear in the surveillance video, the two-dimensional coordinate system is used to determine whether the strangers or vehicles entering the park continue to appear in the surveillance video. Generate the corresponding target trajectory.

[0010] Preferably, the delay time difference The calculation process is as follows: In the two-dimensional coordinate system In the video, mark the coordinate point where the stranger or strange vehicle disappears in the surveillance video as , and then collect the patrol data captured by the drone equipment Substitute patrol video into the two-dimensional coordinate system Filter out drone equipment photography Patrol video of the scene and marked as ; ; In the formula, Indicates the delay time difference, Indicates the current time point, Indicates the current time point minus the time taken by the drone. The time difference obtained from the patrol video is the delay time difference.

[0011] Preferably, if the delay time difference When it is equal to 0, it means the drone equipment is shooting If there are strangers or strange vehicles in the patrol video at the time, When it is not equal to 0, it means the drone equipment is shooting If there are no strangers or vehicles in the patrol video at the time, a corresponding warning signal will be generated and transmitted to the warning management unit.

[0012] Preferably, after receiving the warning signal, the warning management unit substitutes the control instructions at the corresponding time point in the instruction data set into the deep learning model , and analyze and generate tracking mileage , the calculation process is as follows: According to the control instructions of the instruction data set at the corresponding time point, the coordinate point of the corresponding drone device at the current time point is marked as ; ; In the formula, Indicates tracking mileage, The tracking mileage is calculated by calculating the straight-line distance between the current coordinate point of the drone device and the coordinate point where the stranger or vehicle disappears, based on the Euclidean distance calculation formula.

[0013] Preferably, the drone equipment shoots If there are strange people or strange vehicles in the patrol video, the early warning management unit will use the deep learning model to Identify the appearance and behavior characteristics of unfamiliar people or vehicles and transmit them to the central console via the network.

[0014] The deep learning-based intelligent video surveillance method for a campus includes the following steps: Step 1: Connect the monitoring devices to the network to obtain real-time monitoring videos captured by multiple monitoring devices in the park, and form a monitoring data set with the acquired data; Step 2: Establish a two-dimensional coordinate system based on the monitoring data set The coordinate points corresponding to the installation location of each monitoring device are marked. Whenever a stranger or a strange vehicle enters or leaves the park, it is determined whether the stranger or strange vehicle entering the park continues to appear in the monitoring video of the monitoring device. If it continues to appear in the monitoring video, the corresponding target trajectory is generated. , if it does not appear continuously in the surveillance video, a corresponding abnormal signal is generated; Step 3: Connect the drone device to the network to obtain patrol videos of the park at all time points, and form the obtained data into a patrol dataset; Step 4: Whenever an abnormal signal is generated, according to the patrol data set and the two-dimensional coordinate system , analyze and generate delay time difference Then determine whether there are strangers or strange vehicles in the patrol video shot by the drone equipment. If there are strangers or strange vehicles in the patrol video, the deep learning model Identify the appearance and behavior characteristics of strangers or vehicles and transmit them to the central control console via the network. If there are no strangers or vehicles in the patrol video, a corresponding warning signal will be generated; Step 5: Connect to the central control console via the network to obtain all control instructions input by the park management personnel, and form the obtained data into an instruction data set; Step 6: Whenever a warning signal is generated, according to the instruction data set and the two-dimensional coordinate system , analyze and generate tracking mileage , and through deep learning models Filter tracked mileage The shortest drone equipment tracks strangers or vehicles.

[0015] Compared with the existing technology, the present invention provides a campus intelligent video surveillance system and method based on deep learning, which has the following beneficial effects: 1. The present invention connects the monitoring device through the data acquisition module to obtain the monitoring video captured by multiple monitoring devices in the park in real time, and forms a monitoring data set. The intelligent management module establishes a two-dimensional coordinate system based on the monitoring data set. The coordinate points corresponding to the installation location of each monitoring device are marked. Whenever a stranger or a strange vehicle enters or leaves the park, it is determined whether the stranger or strange vehicle entering the park continues to appear in the monitoring video of the monitoring device. If it continues to appear in the monitoring video, the corresponding target trajectory is generated. If it does not appear continuously in the surveillance video, a corresponding abnormal signal is generated. The data acquisition module connects to the drone equipment to obtain the patrol video of the park at all time points and forms a patrol data set. Whenever an abnormal signal is generated, the intelligent management module generates a corresponding abnormal signal based on the patrol data set and the two-dimensional coordinate system. , analyze and generate delay time difference Then determine whether there are strangers or strange vehicles in the patrol video shot by the drone equipment. If there are strangers or strange vehicles in the patrol video, the deep learning model Identify the appearance and behavior characteristics of strangers or vehicles and transmit them to the central control console via the network. If there are no strangers or vehicles in the patrol video, a corresponding early warning signal will be generated, and the intelligent monitoring has high security.

[0016] 2. The present invention connects the central control console through the data acquisition module to obtain all the control instructions input by the park management personnel and form an instruction data set. Whenever an early warning signal is generated, the intelligent management module will generate an instruction data set based on the instruction data set and the two-dimensional coordinate system. , analyze and generate tracking mileage , and through deep learning models Filter tracked mileage The shortest drone device tracks strangers or vehicles with high efficiency and strong tracking capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the system flow of the present invention; Figure 2 It is a step diagram of the method of the present invention. DETAILED DESCRIPTION

[0018] 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.

[0019] Because traditional campus intelligent video surveillance systems based on deep learning have a relatively simple way of collecting video images, blind spots in monitoring can easily lead to major security loopholes in campus management. In addition, when suspicious strangers or vehicles are found, manual investigation takes a long time to track them down, which increases the risk of misjudgment. Therefore, a campus intelligent video surveillance system and method based on deep learning is provided. Please refer to Figure 1-Figure 2 ,The system includes a data acquisition module and an intelligent management module; The data acquisition module consists of a monitoring data unit, a patrol data unit, and an instruction data unit. The monitoring data unit collects monitoring data sets through a network connection to the monitoring device. The monitoring data set includes real-time monitoring videos taken by multiple monitoring devices in the park. The expression of the monitoring data set is: , to The corresponding surveillance videos are taken in real time by each monitoring device in the park. Indicates the area name corresponding to the monitoring device, 1 to Indicates the number of monitoring devices installed in the park Comprehensive collection of surveillance videos from different areas is the basis for safe management of the park; The patrol data unit collects patrol data sets by connecting to drone devices through the network. The patrol data sets include patrol videos at all time points in the park. The expression of the patrol data set is , to The patrol videos corresponding to each time point in the park, Indicates the time point when the drone device shoots the patrol video, 1 to Said that the patrol video captured by the drone equipment has The team uses drone patrol videos to solve the blind spots in traditional surveillance and further reduce security vulnerabilities in the park. The instruction data unit is connected to the central control console through the network to collect the instruction data set. The instruction data set includes all the control instructions input by the park management personnel through the central control console. The expression of the instruction data set is , to Corresponding to the control instructions input by the park management personnel through the central control console each time, Indicates the time when the park manager inputs the control command, from 1 to Indicates that the control instructions entered by the park management personnel are Second, in actual use, the control commands input by managers are usually to control the drone equipment to strengthen patrols; The data acquisition module transmits the monitoring data set, patrol data set and instruction data set to the intelligent management module through the network; The intelligent management module consists of an investigation and analysis unit, an abnormality assessment unit and an early warning management unit. The investigation and analysis unit generates target trajectories based on the monitoring data set. , the calculation process is as follows: Take the park entrance and exit as the origin Establish a two-dimensional coordinate system , and combined with the number of monitoring devices installed in the park in the monitoring data set in the two-dimensional coordinate system Set the corresponding number of coordinate points in Monitoring devices corresponding to coordinate points, and then The coordinate points are marked as ; Whenever a stranger or vehicle enters or leaves the park, it will be marked as ; ; In the formula, represents the target trajectory, Represents a two-dimensional coordinate system Middle coordinate points, Indicates based on The surveillance video corresponding to the coordinate point is used to determine whether the strangers or vehicles entering the park continue to appear in the surveillance video. If the strangers or vehicles entering the park continue to appear in the surveillance video, the two-dimensional coordinate system is used to determine whether the strangers or vehicles entering the park continue to appear in the surveillance video. Generate the corresponding target trajectory in the target trajectory, which has a coherent and complete target trajectory Higher safety for people and vehicles; The investigation and analysis unit is based on the target trajectory Determine whether the stranger or vehicle entering the park continues to appear in the surveillance video. If the stranger or vehicle entering the park does not continue to appear in the surveillance video, it indicates that the stranger or vehicle is suspected of being abnormal, and a corresponding abnormal signal is generated and transmitted to the abnormality assessment unit; The abnormality assessment unit analyzes and generates delay time difference based on the patrol data set , the calculation process is as follows: In the two-dimensional coordinate system In the video, mark the coordinate point where the stranger or strange vehicle disappears in the surveillance video as , and then collect the patrol data captured by the drone equipment Substitute patrol video into the two-dimensional coordinate system Filter out drone equipment photography Patrol video of the scene and marked as ; ; In the formula, Indicates the delay time difference, Indicates the current time point, Indicates the current time point minus the time taken by the drone. The time difference obtained from the patrol video is the delay time difference; The abnormality assessment unit is based on the delay time difference Determine whether there are strangers or vehicles in the patrol video shot by the drone equipment. If the delay time difference When it is equal to 0, it means the drone equipment is shooting If there are strangers or strange vehicles in the patrol video at the time, When it is not equal to 0, it means the drone equipment is shooting If there are no strangers or vehicles in the patrol video at the time, the corresponding warning signal will be generated and transmitted to the warning management unit. Subsequently, drone equipment will be dispatched to track the target, and the intelligent monitoring is highly secure. The early warning management unit obtains the deep learning model through the network connection to the central console , deep learning model Specifically, it is a convolutional neural network (CNN), which mainly has the function of image recognition. After the warning management unit receives the warning signal, it substitutes the control instructions at the corresponding time point in the instruction data set into the deep learning model. , and analyze and generate tracking mileage , the calculation process is as follows: According to the control instructions of the instruction data set at the corresponding time point, the coordinate point of the corresponding drone device at the current time point is marked as ; ; In the formula, Indicates tracking mileage, The tracking mileage is calculated by calculating the straight-line distance between the current coordinate point of the drone and the coordinate point where the stranger or vehicle disappears, based on the Euclidean distance calculation formula. Deep Learning Models Based on tracking mileage , filter out tracked mileage The shortest drone device tracks strangers or vehicles with high efficiency and strong tracking capabilities.

[0020] The deep learning-based intelligent video surveillance method for campuses includes the following steps: Step 1: Connect the monitoring devices to the network to obtain real-time monitoring videos captured by multiple monitoring devices in the park, and form a monitoring data set with the acquired data; Step 2: Establish a two-dimensional coordinate system based on the monitoring data set The coordinate points corresponding to the installation location of each monitoring device are marked. Whenever a stranger or a strange vehicle enters or leaves the park, it is determined whether the stranger or strange vehicle entering the park continues to appear in the monitoring video of the monitoring device. If it continues to appear in the monitoring video, the corresponding target trajectory is generated. , if it does not appear continuously in the surveillance video, a corresponding abnormal signal is generated; Step 3: Connect the drone device to the network to obtain patrol videos of the park at all time points, and form the obtained data into a patrol dataset; Step 4: Whenever an abnormal signal is generated, according to the patrol data set and the two-dimensional coordinate system , analyze the generated delay time difference Then determine whether there are strangers or strange vehicles in the patrol video shot by the drone equipment. If there are strangers or strange vehicles in the patrol video, the deep learning model Identify the appearance and behavioral characteristics of strangers or vehicles. For example, appearance characteristics refer to the stranger's face, height, and body shape, while behavioral characteristics refer to the length of time the stranger has stayed and whether the license plate is obscured. These characteristics are then transmitted to the central control console via the network. If no strangers or vehicles are present in the patrol video, a corresponding warning signal is generated, providing high security for intelligent monitoring. Step 5: Connect to the central control console via the network to obtain all control instructions input by the park management personnel, and form the obtained data into an instruction data set; Step 6: Whenever a warning signal is generated, according to the instruction data set and the two-dimensional coordinate system , analyze and generate tracking mileage , and through deep learning models Filter tracked mileage The shortest drone device tracks strangers or vehicles with high efficiency and strong tracking capabilities.

[0021] Example 1: In this experiment, the petrochemical park was selected as the experimental object. After detection, the target A did not persist in the surveillance video captured by all monitoring devices in the park. The coordinate point where the target A disappeared in the surveillance video was marked as , and filter out drone equipment shooting Real-time patrol video, shot by drone equipment The patrol video was recorded at 14:00 in the afternoon, and the current time is 14:12 in the afternoon. The calculation formula is as follows: ; In the formula, Indicates the delay time difference, Indicates the current time point minus the time taken by the drone. The time difference obtained by the patrol video This is the delay time difference.

[0022] Example 2: In this experiment, the logistics park was selected as the experimental object. After testing, the target B did not persist in the surveillance video captured by all the monitoring devices in the park. The coordinate point where the target B disappeared in the surveillance video was marked as There are two UAVs responsible for tracking target B. The coordinates of these two UAVs at the current time are: 、 , the tracking mileage of these two drone devices The calculation formula is as follows: ; In the formula, Indicates tracking mileage, Indicates that the straight-line distance between the coordinate point of the first drone device at the current time point and the coordinate point where the stranger or strange vehicle disappears is calculated according to the Euclidean distance calculation formula. To track mileage, Indicates that the straight-line distance between the coordinate point of the second drone device at the current time point and the coordinate point where the stranger or strange vehicle disappears is calculated based on the Euclidean distance calculation formula. That is, to track mileage, it was determined that the first drone device was more suitable for tracking strangers or vehicles.

[0023] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The park intelligent video surveillance system based on deep learning is characterized by: Including data acquisition module and intelligent management module; The data acquisition module is composed of a monitoring data unit, a patrol data unit and an instruction data unit. The monitoring data unit is connected to the monitoring device through the network to collect a monitoring data set. The monitoring data set includes monitoring videos taken in real time by multiple monitoring devices in the park. The patrol data unit is connected to the drone equipment through the network to collect a patrol data set. The patrol data set includes patrol videos of all time points in the park. The instruction data unit is connected to the central control console through the network to collect an instruction data set. The instruction data set includes all control instructions input by the park management personnel through the central control console. The data acquisition module transmits the monitoring data set, patrol data set and instruction data set to the intelligent management module through the network; The intelligent management module consists of an investigation and analysis unit, an abnormality assessment unit and an early warning management unit. The investigation and analysis unit generates target trajectories based on the monitoring data set. The investigation and analysis unit is based on the target trajectory Determine whether the strangers or vehicles entering the park continue to appear in the surveillance video. If the strangers or vehicles entering the park do not continue to appear in the surveillance video, a corresponding abnormal signal is generated and transmitted to the abnormality assessment unit. The abnormality assessment unit analyzes and generates a delay time difference based on the patrol data set. , and then according to the delay time difference Determine whether there are strangers or strange vehicles in the patrol video taken by the drone equipment. If there are no strangers or strange vehicles in the patrol video taken by the drone equipment, generate a corresponding warning signal and transmit it to the warning management unit. The warning management unit obtains the deep learning model through the network connection central control station , and combined with the instruction data set, analyze and generate tracking mileage , the deep learning model According to tracking mileage , filter out tracked mileage The shortest drone equipment tracks strange people or strange vehicles.

2. The campus intelligent video surveillance system based on deep learning according to claim 1 is characterized in that: The expression of the monitoring data set is: , to Corresponding to the surveillance video captured in real time by each surveillance device in the park, Indicates the area name corresponding to the monitoring device, 1 to Indicates the number of monitoring devices installed in the park indivual.

3. The park intelligent video surveillance system based on deep learning according to claim 2 is characterized in that: The expression of the patrol data set is: , to Corresponding to the patrol videos at each time point in the park, Indicates the time when the drone device shoots the patrol video, 1 to Said that the patrol video captured by the drone equipment has Group.

4. The park intelligent video surveillance system based on deep learning according to claim 3 is characterized in that: The expression of the instruction data set is: , to Corresponding to the control commands input by the park managers through the central control console each time, Indicates the time when the park manager inputs the control command, from 1 to Indicates that the control instructions entered by the park management personnel are Second-rate.

5. The park intelligent video surveillance system based on deep learning according to claim 4 is characterized in that: The target trajectory The calculation process is as follows: Take the park entrance as the origin Establishing a 2D coordinate system , and combined with the number of monitoring devices installed in the park in the monitoring data set, the two-dimensional coordinate system Set the corresponding number of coordinate points in Monitoring devices correspond to coordinate points, and then The coordinate points are marked as ; Whenever a stranger or vehicle enters or leaves the park, it is marked as ; ； In the formula, represents the target trajectory, Represents a two-dimensional coordinate system Middle coordinate points, According to The surveillance video corresponding to the coordinate point is used to determine whether the strangers or vehicles entering the park continue to appear in the surveillance video. If the strangers or vehicles entering the park continue to appear in the surveillance video, the two-dimensional coordinate system Generate the corresponding target trajectory.

6. The campus intelligent video surveillance system based on deep learning according to claim 5 is characterized in that: The delay time difference The calculation process is as follows: In the two-dimensional coordinate system In the video, the coordinates of the points where the strangers or vehicles disappear in the surveillance video are marked as , and then collect the patrol data captured by the drone equipment Substitute patrol video into the two-dimensional coordinate system Filter out drone equipment photography Patrol video of the incident, marked as ; ； In the formula, Indicates the delay time difference, Indicates the current time point. Indicates the current time minus the time taken by the drone The time difference obtained from the patrol video is the delay time difference.

7. The campus intelligent video surveillance system based on deep learning according to claim 6 is characterized in that: If the delay time difference When it is equal to 0, it means the drone equipment is shooting If there are unfamiliar people or vehicles in the patrol video at the time, the delay When it is not equal to 0, it means the drone equipment is shooting If there are no strangers or vehicles in the patrol video at that time, a corresponding warning signal will be generated and transmitted to the warning management unit.

8. The campus intelligent video surveillance system based on deep learning according to claim 7 is characterized in that: After receiving the warning signal, the warning management unit substitutes the control instruction at the corresponding time point in the instruction data set into the deep learning model , and analyze and generate tracking mileage , the calculation process is as follows: According to the control instructions at the corresponding time point of the instruction data set, the coordinate point corresponding to the current time point of the UAV equipment is marked as ; ； In the formula, Indicates tracking mileage, It means that according to the Euclidean distance calculation formula, the straight-line distance between the coordinate point of the drone device at the current time point and the coordinate point where the stranger or strange vehicle disappears is calculated, which is the tracking mileage.

9. The campus intelligent video surveillance system based on deep learning according to claim 8, characterized in that: Said drone equipment filming If there are strangers or strange vehicles in the patrol video, the early warning management unit uses the deep learning model to Identify the appearance and behavior characteristics of unfamiliar people or vehicles and transmit them to the central console via the network.

10. A campus intelligent video surveillance method based on deep learning, applied to a campus intelligent video surveillance system based on deep learning according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: obtain the monitoring videos taken by multiple monitoring devices in the park in real time through the network connection monitoring device, and form the obtained data into a monitoring data set; Step 2: Establish a two-dimensional coordinate system based on the monitoring data set The coordinate points corresponding to the installation location of each monitoring device are marked. Whenever a stranger or a strange vehicle enters or leaves the park, it is determined whether the stranger or the strange vehicle entering the park continues to appear in the monitoring video of the monitoring device. If it continues to appear in the monitoring video, the corresponding target trajectory is generated. , if it does not appear continuously in the surveillance video, a corresponding abnormal signal is generated; Step 3: Connect the drone device to the network to obtain patrol videos of the park at all time points, and form a patrol data set with the acquired data; Step 4: Whenever an abnormal signal is generated, according to the patrol data set and the two-dimensional coordinate system , analyze the generated delay time difference , and then determine whether there are strangers or strange vehicles in the patrol video taken by the drone equipment. If there are strangers or strange vehicles in the patrol video, the deep learning model Identify the appearance and behavior characteristics of strangers or strange vehicles, and transmit them to the central control console through the network. If there are no strangers or strange vehicles in the patrol video, generate corresponding warning signals; Step 5: Connect to the central control console through the network to obtain all control instructions input by the park management personnel, and form the obtained data into an instruction data set; Step 6: Whenever a warning signal is generated, according to the instruction data set and the two-dimensional coordinate system , analyze and generate tracking mileage , and through the deep learning model Filter to track mileage The shortest drone equipment tracks strange people or strange vehicles.

Images