Method and device for identifying high-altitude parabolic object

High-definition camera data is obtained through real-time transmission protocol, and high-altitude thrown objects are combined with SORT and SOM network algorithms to identify objects thrown at high altitudes, solving the problem of false detection in the existing technology, realizing an efficient and automated identification process, and adaptive lighting and scene changes.

CN118967743APending Publication Date: 2024-11-15CHINA NORTHWEST WATER CONSERVANCY & HYDROPOWER ENG CONSULTING +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410882160.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-03
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The identification method of high-altitude thrown objects in the prior art requires a large amount of data to solve the problem of false detection caused by background and flying birds, and it is difficult to adapt to lighting and scene changes.

Method used

The real-time transmission protocol is used to obtain high-definition camera data, track and classify moving objects through SORT target tracking algorithm and SOM network algorithm, background modeling is performed by combining Gaussian hybrid model and morphological opening operations, and detection results are transmitted using the MQTT protocol.

Benefits of technology

It realizes efficient high-altitude object throw recognition, reduces false detection, automates all analysis processes, and adapts scene changes in background modeling, without manual parameters adjustment, saving time and effort.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118967743A_ABST
    Figure CN118967743A_ABST
Patent Text Reader

Abstract

The invention discloses a high-altitude parabolic object identification method and device. The method comprises the following steps: acquiring image data of an external environment of a building shot by a high-definition camera installed on the building in real time based on a real-time stream transmission protocol; receiving and processing the real-time video stream; background modeling is carried out to distinguish a background and a moving object, and interference terms are removed through open operation; tracking the moving object based on an SORT target tracking algorithm, and obtaining a movement track of the moving object; based on an SOM network algorithm, classifying the motion trail of the moving object so as to distinguish whether the moving object is a high-altitude parabolic object; and transmitting the detected position information of the high-altitude parabolic object to an alarm platform through an MQTT protocol. According to the method, a moving object is tracked through an SORT target tracking algorithm and an SOM network algorithm, and the trajectory of the moving object is classified, so that a large amount of data does not need to be collected for training. In addition, all analysis processes are fully automatic, manual parameter adjustment is not needed, and time and labor are saved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of Internet of Things and computer vision technology, and in particular to a method and device for identifying objects thrown from high altitudes. Background Art

[0002] In order to effectively monitor, identify and warn of high-altitude dropping behaviors, most methods currently use deep learning methods to detect and track suspicious targets of high-altitude dropping. However, such methods often require a large amount of data to resolve false detections caused by background, flying birds, etc. Therefore, there is an urgent need for a simple and effective method to identify high-altitude dropping. Summary of the invention

[0003] The main purpose of the present invention is to provide a method and device for identifying objects thrown from high altitudes, so as to solve the technical problem in the prior art that a large amount of data is needed to solve the false detection caused by background, flying birds, etc.

[0004] To achieve the above-mentioned purpose, the present invention provides a method for identifying objects thrown from high altitudes, the method comprising: S10, acquiring in real time based on a real-time transmission protocol image data of an external environment of a building captured by a high-definition camera installed on the building; S20, receiving and processing a real-time video stream; S30, performing background modeling to distinguish between the background and moving objects and removing interference items through an open operation; S40, tracking the moving object based on a SORT target tracking algorithm and deriving a motion trajectory of the moving object; S50, classifying the motion trajectory of the moving object based on a SOM network algorithm to distinguish whether it is an object thrown from high altitudes; S60, transmitting the position information of the detected object thrown from high altitudes to an alarm platform through an MQTT protocol.

[0005] Optionally, the step S20 includes the following steps: S210, receiving a real-time video stream; S220, performing video de-jitter processing on the real-time video stream based on a bit plane matching algorithm.

[0006] Optionally, the step S220 includes the following steps: S221, taking the first frame of the real-time video stream as a reference image, converting the RGB images of subsequent video frames into grayscale images, performing binary encoding processing, and decomposing them into different bit planes; S222, extracting the fourth bit plane image; S223, setting four image blocks in the four corner areas of the reference image, and in the current video frame, obtaining the best matching block through Boolean XOR operation around the image blocks, and then obtaining the local motion vector; S224, performing median filtering on the obtained four local motion vectors and the global motion vector of the previous frame to obtain the global motion vector of the current frame, and then performing reverse compensation on the current video frame; S225, outputting a relatively stable video sequence image.

[0007] Optionally, the step S30 includes the following steps: S310, performing background modeling based on the MOG2 algorithm of the Gaussian mixture model; S320, distinguishing the background and moving objects; S330, removing interference items based on morphological opening operation.

[0008] Optionally, the step S310 includes the following steps: S311, initializing the background model, initializing a mixed model containing K Gaussian distributions for each pixel; S312, inputting the video frame image acquired within a preset time for training to obtain an initial background model; S313, inputting a new video frame image, calculating the matching degree between each pixel and its corresponding background model, if the matching degree between the pixel color and the Gaussian distribution of the background model is lower than a preset threshold, the pixel is considered to be the background, otherwise it is the foreground; S314, updating the model, using the foreground pixels to update the parameters of the corresponding Gaussian distribution in the background model, so that the background model can adapt to scene changes and reduce false detections.

[0009] Optionally, the step S40 includes the following steps: S410, obtaining the circumscribed rectangular frame of the moving object in the foreground image, and numbering each moving object; S420, constructing a Kalman filter for modeling and predicting the motion state of each tracked target; S430, inputting the rectangular frames of all moving objects and their corresponding numbers into the Kalman filter, and estimating the position and speed of the target in the next frame in combination with the current state of the target and the motion model; S440, performing data association based on the Hungarian algorithm to determine the correspondence between the target detected in the current frame and the tracked target in the previous frame; S450, matching the target in the current frame with the tracked target in the previous frame by minimizing the association cost, thereby determining the target number and trajectory; S460, updating the target tracking state, matching the detected target with the tracked target according to the matching result, and updating the state of the Kalman filter; S470, outputting the position and speed of each moving object in each frame.

[0010] Optionally, in step S460, the matching results are divided into unmatched tracking targets, unmatched detection results, and successfully matched tracking targets.

[0011] Optionally, after step S460, the method may further include the following steps: S461, removing the continuously unmatched tracking target from the tracking list; S462, initializing the unmatched detection result as a new tracking target.

[0012] Optionally, the step S50 includes the following steps: S510, constructing a SOM network classifier suitable for high-altitude parabolic objects; S520, using the position information, velocity information and acceleration information of high-altitude parabolic objects and non-high-altitude parabolic objects in all video frames within a preset time to construct a feature data set of high-altitude parabolic objects and a feature data set of non-high-altitude parabolic objects, and normalizing the data set, and inputting it into the SOM network for training; S530, the SOM network automatically adjusts the weights of neurons through a competitive learning mechanism to adapt to the differences and similarities between input data, and maps the input trajectory to neurons according to the characteristics of the input trajectory; S540, receiving the position and velocity of the moving object output by the SORT network algorithm within a preset time, calculating the acceleration of each frame, and constructing it into a feature vector and inputting it into the SOM network classifier; S550, judging whether each of the moving objects is a high-altitude parabolic object.

[0013] In addition, to achieve the above-mentioned purpose, the embodiment of the present application also provides a device for identifying objects thrown from high altitudes, and the device includes: an image acquisition module, which is used to obtain image data of the external environment of a building captured by a high-definition camera installed on the building in real time based on a real-time transmission protocol; a video stabilization module, which is used to receive and process real-time video streams; a background modeling module, which is used to perform background modeling, distinguish between background and moving objects, and remove interference items through open operations; a target tracking module, which is used to track the moving object based on the SORT target tracking algorithm and derive the motion trajectory of the moving object; a trajectory classification module, which is used to classify the motion trajectory of the moving object based on the SOM network algorithm to distinguish whether it is an object thrown from high altitudes; an alarm reporting module, which is used to transmit the location information of the detected object thrown from high altitudes to the alarm platform through the MQTT protocol.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] The method and device for identifying high-altitude objects provided in the embodiment of the present application obtains image data of the external environment of the building in real time through a real-time transmission protocol, and tracks moving objects and classifies their trajectories through the SORT target tracking algorithm and the SOM network algorithm, so there is no need to collect a large amount of data for training, and can greatly avoid false detections caused by background, flying birds, etc. In addition, all analysis processes are fully automated, and the background modeling continuously adapts to changes in the scene over time, and adapts to changes in lighting and scenes, without the need for manual parameter adjustment, saving time and effort. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart of a method for identifying objects thrown from a high altitude provided in an embodiment of the present application;

[0017] Figure 2Another flow chart of the method for identifying objects thrown from a high altitude provided in an embodiment of the present application;

[0018] Figure 3 This is a structural block diagram of the device for identifying objects thrown from high places provided in an embodiment of the present application.

[0019] Description of reference numerals:

[0020] 300. A device for identifying objects thrown from high altitudes; 310. An image acquisition module; 320. A video anti-shake module; 330. A background modeling module; 340. A target tracking module; 350. A trajectory classification module; 360. An alarm reporting module.

[0021] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0022] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0023] In order to solve the above technical problems, the present application provides a method for identifying objects thrown from a high altitude, which can be executed by a computer, such as Figure 1 and Figure 2 As shown, the method may include the following steps:

[0024] S10, acquiring image data of the external environment of the building captured by a high-definition camera installed on the building in real time based on a real-time transmission protocol.

[0025] Specifically, RTP (Realtime Transport Protocol) provides end-to-end transmission services with real-time characteristics for data, such as interactive video audio or analog data under multicast or unicast network services. Applications usually run RTP on UDP (User Datagram Protocol) to use its multi-path node and verification services; both protocols provide the functions of transport layer protocols. However, RTP can be used with other suitable underlying networks or transport protocols. If the underlying network provides multicast mode, then RTP can use the multicast table to transmit data to multiple destinations.

[0026] In an exemplary embodiment, step S20 may include the following steps:

[0027] S210, receiving a real-time video stream;

[0028] S220, performing video de-jittering processing on the real-time video stream based on a bit plane matching algorithm.

[0029] Further, in an exemplary embodiment, step S220 may include the following steps:

[0030] S221, taking the first frame of the real-time video stream as a reference image, converting the RGB images of the subsequent video frames into grayscale images, performing binary encoding processing on them, and decomposing them into different bit planes;

[0031] S222, extracting the fourth bit plane image;

[0032] S223, setting four image blocks in four corner areas of the reference image, and obtaining the best matching block by performing a Boolean XOR operation around the image blocks in the current video frame, and then obtaining a local motion vector;

[0033] S224, performing median filtering on the obtained four local motion vectors and the global motion vector of the previous frame to obtain the global motion vector of the current frame, and then performing reverse compensation on the current video frame;

[0034] S225, outputting relatively stable video sequence pictures.

[0035] In an exemplary embodiment, step S30 may specifically include the following steps:

[0036] S310, background modeling based on the MOG2 algorithm of Gaussian mixture model;

[0037] S320, distinguishing the background from the moving object;

[0038] S330, removing interference items based on morphological opening operation.

[0039] Specifically, the Gaussian mixture model uses the Gaussian probability density function (normal distribution curve) to accurately quantify things and decomposes one thing into several models based on the Gaussian probability density function (normal distribution curve).

[0040] In this exemplary embodiment, since the high-altitude parabolic objects scene captured is usually a fixed scene with little change, but the lighting conditions are often unstable, the MOG2 (Mixtore of Gauss ian2) algorithm uses a Gaussian mixture model to model the color of each pixel. By learning the distribution of pixel colors, it can adaptively establish a background model and update the model according to new observation data. It can better cope with scene changes caused by changes in brightness. Therefore, the MOG2 algorithm is selected for background modeling to separate the background and moving objects.

[0041] In an exemplary embodiment, step S310 may specifically include the following steps:

[0042] S311, initializing the background model, and initializing a mixture model containing K Gaussian distributions for each pixel;

[0043] S312, inputting video frame images acquired within a preset time for training to obtain an initial background model, wherein the preset time is 3 seconds;

[0044] S313, input a new video frame image, calculate the matching degree between each pixel and its corresponding background model, if the matching degree of the Gaussian distribution of the pixel color and the background model is lower than a preset threshold, the pixel is considered to be the background, otherwise it is the foreground;

[0045] S314, updating the model, using the foreground pixels to update the parameters of the corresponding Gaussian distribution in the background model, so that the background model can adapt to scene changes and reduce false detection.

[0046] In an exemplary embodiment, step S40 may specifically include the following steps:

[0047] S410, obtaining the circumscribed rectangular frame of the moving objects in the foreground image, and numbering each moving object;

[0048] S420, constructing a Kalman filter for modeling and predicting the motion state of each tracked target;

[0049] S430, inputting the rectangular frames of all moving objects and their corresponding numbers into the Kalman filter, and estimating the position and speed of the target in the next frame in combination with the current state of the target and the motion model;

[0050] S440, performing data association based on the Hungarian algorithm to determine a correspondence between a target detected in the current frame and a tracked target in a previous frame;

[0051] S450, matching the target in the current frame with the tracked target in the previous frame by minimizing the association cost, thereby determining the number and trajectory of the target;

[0052] S460, updating the target tracking state, matching the detected target with the tracking target according to the matching result, and updating the state of the Kalman filter, wherein the matching result is divided into an unmatched tracking target, an unmatched detection result, and a successfully matched tracking target;

[0053] S470, output the position and speed of each moving object in each frame.

[0054] Specifically, since the moving objects of high-altitude parabolic objects have strong regularity, there is basically an IOU (intersection-over-union) overlap between each frame, and the SORT (Simple On line And Real time Tracking) algorithm uses the current frame position to predict the position of the next frame, and can better track objects with strong regularity. Therefore, this embodiment adopts the SORT algorithm to track moving objects.

[0055] Further, in step S420, the process of constructing the Kalman filter is as follows: Assume that the two variables p and velocity v are both random and follow a Gaussian distribution. Each variable has a mean μ, which is the center of the random distribution, and a variance σ representing uncertainty. 2 ,The position of an object is related to its speed. The faster the speed, the farther its next position will be. Kalman filtering uses a series of previous observations to predict the next observation.

[0056] In an exemplary embodiment, after step S460, the method may specifically include the following steps:

[0057] S461, removing the continuously unmatched tracking target from the tracking list;

[0058] S462, initializing the unmatched detection result as a new tracking target.

[0059] S50, based on a SOM (Self-organizing Maps, self-organizing map neural network) network algorithm, classify the motion trajectory of the moving object to distinguish whether it is a high-altitude parabolic object.

[0060] In an exemplary embodiment, the step S50 includes the following steps:

[0061] S510, constructing a SOM network classifier suitable for high-altitude objects;

[0062] S520, constructing a feature data set of high-altitude parabolic objects and a feature data set of non-high-altitude parabolic objects using the position information, velocity information and acceleration information of high-altitude parabolic objects and non-high-altitude parabolic objects in all video frames within a preset time, normalizing the data set, and inputting it into the SOM network for training, wherein the preset time is 3 seconds;

[0063] S530, the SOM network automatically adjusts the weights of neurons through a competitive learning mechanism to adapt to the differences and similarities between input data, and maps the input trajectory to neurons according to its characteristics;

[0064] S540, receiving the position and velocity of the moving object output by the SORT network algorithm within a preset time, calculating the acceleration of each frame thereof, and constructing a feature vector to be input into the SOM network classifier;

[0065] S550, determining whether each of the moving objects is a high-altitude parabolic object.

[0066] Specifically, the SOM self-organizing map neural network trains the input data by unsupervised learning, generates a low-dimensional discrete mapping, can identify and distinguish different motion modes, and organizes them into different categories. First, a SOM network classifier suitable for high-altitude parabolic objects is constructed, and the position information, velocity information and acceleration information of the high-altitude parabolic objects and non-high-altitude parabolic objects in all video frames within 3 seconds are used to construct the feature data set of the high-altitude parabolic objects and the feature data set of non-high-altitude parabolic objects, and the data set is normalized and input into the SOM network for training. The SOM network automatically adjusts the weight of the neuron through a competitive learning mechanism to adapt to the differences and similarities between the input data, and maps it to the neuron according to the characteristics of the input trajectory. These neurons are arranged according to a certain topological structure, and adjacent neurons correspond to similar motion trajectories. In the present embodiment, the moving object is divided into four categories: high-altitude parabolic objects, flying birds, leaves, and clothes. Next, the position and velocity of the moving object output by the SORT algorithm within three seconds are received, and the acceleration of each frame is calculated, and the feature vector is constructed and input into the high-altitude parabolic SOM network classifier to determine whether each moving object is a high-altitude parabolic object.

[0067] Furthermore, the feature vector is constructed by integrating the motion position coordinates, velocity, and acceleration values ​​and normalizing them.

[0068] S60, transmitting the position information of the detected high-altitude parabolic object to the alarm platform via MQTT (MQ TelemetryTransport, message queue telemetry transmission protocol).

[0069] Specifically, MQTT is a client-server based message publish / subscribe transport protocol. The MQTT protocol is lightweight, simple, open and easy to implement, which makes it very widely applicable. In many cases, including restricted environments, such as machine-to-machine (M2M) communication and the Internet of Things (IoT). It has been widely used in sensors communicating via satellite links, medical devices that occasionally dial up, smart homes, and some miniaturized devices.

[0070] The method for identifying high-altitude objects provided in the embodiment of the present application obtains image data of the external environment of the building in real time through a real-time transmission protocol, and tracks moving objects and classifies their trajectories through the SORT target tracking algorithm and the SOM network algorithm, so there is no need to collect a large amount of data for training, and can greatly avoid false detections caused by background, flying birds, etc. In addition, all analysis processes are fully automated, and the background modeling continuously adapts to changes in the scene over time, and adapts to changes in lighting and scenes, without the need for manual parameter adjustment, saving time and effort.

[0071] Based on the above embodiments, Figure 3 Another embodiment of the present application further provides a device for identifying objects thrown from a high altitude. The device 300 for identifying objects thrown from a high altitude may include the following modules:

[0072] The image acquisition module 310 is used to acquire image data of the external environment of the building captured by a high-definition camera installed on the building in real time based on a real-time streaming protocol;

[0073] The video anti-shake module 320 is used to receive and process the real-time video stream;

[0074] Background modeling module 330, used to perform background modeling, distinguish between background and moving objects and remove interference items through opening operation;

[0075] A target tracking module 340 is used to track the moving object based on a SORT target tracking algorithm and obtain a moving trajectory of the moving object;

[0076] The trajectory classification module 350 is used to classify the motion trajectory of the moving object based on the SOM network algorithm to distinguish whether it is a high-altitude parabolic object;

[0077] The alarm reporting module 360 ​​is used to transmit the location information of the detected high-altitude projectiles to the alarm platform through the MQTT protocol.

[0078] In the description of the present application, it should be noted that the terms "first", "second" and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0079] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0080] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0081] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0082] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0083] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0084] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

[0085] In addition, the algorithms mentioned in the above embodiments are all prior art; although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

Claims

1. A method for identifying objects thrown from high altitudes, characterized in that: The method for identifying objects thrown from high altitude comprises the following steps: S10, acquiring image data of the external environment of the building captured by a high-definition camera installed on the building in real time based on a real-time transmission protocol; S20, receiving and processing real-time video stream; S30, performing background modeling to distinguish the background from the moving object and removing interference items through an open operation; S40, tracking the moving object based on the SORT target tracking algorithm, and obtaining the moving trajectory of the moving object; S50, based on the SOM network algorithm, classifying the motion trajectory of the moving object to distinguish whether it is a high-altitude parabolic object; S60, transmit the location information of the detected high-altitude projectiles to the alarm platform via the MQTT protocol.

2. The method for identifying objects thrown from high altitude according to claim 1, characterized in that: The step S20 comprises the following steps: S210, receiving a real-time video stream; S220, performing video de-jittering processing on the real-time video stream based on a bit plane matching algorithm.

3. The method for identifying objects thrown from high altitude according to claim 2, characterized in that: The step S220 includes the following steps: S221, taking the first frame of the real-time video stream as a reference image, converting the RGB images of the subsequent video frames into grayscale images, performing binary encoding processing on them, and decomposing them into different bit planes; S222, extracting the fourth bit plane image; S223, setting four image blocks in four corner areas of the reference image, and obtaining the best matching block by performing a Boolean XOR operation around the image blocks in the current video frame, and then obtaining a local motion vector; S224, performing median filtering on the obtained four local motion vectors and the global motion vector of the previous frame to obtain the global motion vector of the current frame, and then performing reverse compensation on the current video frame; S225, outputting relatively stable video sequence pictures.

4. The method for identifying objects thrown from high altitude according to claim 1, characterized in that: The step S30 comprises the following steps: S310, background modeling based on the MOG2 algorithm of Gaussian mixture model; S320, distinguishing the background from the moving object; S330, removing interference items based on morphological opening operation.

5. The method for identifying objects thrown from high altitude according to claim 4, characterized in that: The step S310 includes the following steps: S311, initializing the background model, and initializing a mixture model containing K Gaussian distributions for each pixel; S312, inputting video frame images acquired within a preset time for training to obtain an initial background model; S313, input a new video frame image, calculate the matching degree between each pixel and its corresponding background model, if the matching degree of the Gaussian distribution of the pixel color and the background model is lower than a preset threshold, the pixel is considered to be the background, otherwise it is the foreground; S314, updating the model, using the foreground pixels to update the parameters of the corresponding Gaussian distribution in the background model, so that the background model can adapt to scene changes and reduce false detection.

6. The method for identifying objects thrown from high altitude according to claim 1, characterized in that: The step S40 comprises the following steps: S410, obtaining the circumscribed rectangular frame of the moving objects in the foreground image, and numbering each moving object; S420, constructing a Kalman filter for modeling and predicting the motion state of each tracked target; S430, inputting the rectangular frames of all moving objects and their corresponding numbers into the Kalman filter, and estimating the position and speed of the target in the next frame in combination with the current state of the target and the motion model; S440, performing data association based on the Hungarian algorithm to determine a correspondence between a target detected in the current frame and a tracked target in a previous frame; S450, matching the target in the current frame with the tracked target in the previous frame by minimizing the association cost, thereby determining the number and trajectory of the target; S460, updating the target tracking state, matching the detected target with the tracking target according to the matching result, and updating the state of the Kalman filter; S470, output the position and speed of each moving object in each frame.

7. The method for identifying objects thrown from high altitude according to claim 6, characterized in that: In step S460, the matching results are divided into unmatched tracking targets, unmatched detection results, and successfully matched tracking targets.

8. The method for identifying objects thrown from high altitude according to claim 7, characterized in that: After step S460, the method may further include the following steps: S461, removing the continuously unmatched tracking target from the tracking list; S462, initializing the unmatched detection result as a new tracking target.

9. The method for identifying objects thrown from high altitude according to claim 1, characterized in that: The step S50 comprises the following steps: S510, constructing a SOM network classifier suitable for high-altitude objects; S520, constructing a feature data set of high-altitude parabolic objects and a feature data set of non-high-altitude parabolic objects using the position information, velocity information, and acceleration information of high-altitude parabolic objects and non-high-altitude parabolic objects in all video frames within a preset time, normalizing the data set, and inputting it into the SOM network for training; S530, the SOM network automatically adjusts the weights of neurons through a competitive learning mechanism to adapt to the differences and similarities between input data, and maps the input trajectory to neurons according to its characteristics; S540, receiving the position and velocity of the moving object output by the SORT network algorithm within a preset time, calculating the acceleration of each frame thereof, and constructing a feature vector to be input into the SOM network classifier; S550, determining whether each of the moving objects is a high-altitude parabolic object.

10. A device for identifying objects thrown from high altitudes, characterized in that: include: An image acquisition module is used to acquire image data of the external environment of a building captured by a high-definition camera installed on the building in real time based on a real-time transmission protocol; Video anti-shake module, used to receive and process real-time video stream; Background modeling module, used to perform background modeling, distinguish between background and moving objects and remove interference items through opening operation; A target tracking module, used to track the moving object based on a SORT target tracking algorithm and obtain a moving trajectory of the moving object; A trajectory classification module is used to classify the motion trajectory of the moving object based on the SOM network algorithm to distinguish whether it is a high-altitude parabolic object; The alarm reporting module is used to transmit the location information of detected high-altitude thrown objects to the alarm platform through the MQTT protocol.