WO2021217934A1 - 监控牲畜数量的方法、装置、计算机设备及存储介质 - Google Patents

监控牲畜数量的方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2021217934A1
WO2021217934A1 PCT/CN2020/105770 CN2020105770W WO2021217934A1 WO 2021217934 A1 WO2021217934 A1 WO 2021217934A1 CN 2020105770 W CN2020105770 W CN 2020105770W WO 2021217934 A1 WO2021217934 A1 WO 2021217934A1
Authority
WO
WIPO (PCT)
Prior art keywords
livestock
frame
tracking
detection
redundant
Prior art date
Application number
PCT/CN2020/105770
Other languages
English (en)
French (fr)
Inventor
张玉琪
陈伟杰
Original Assignee
平安国际智慧城市科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安国际智慧城市科技股份有限公司 filed Critical 平安国际智慧城市科技股份有限公司
Publication of WO2021217934A1 publication Critical patent/WO2021217934A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/40Scenes; Scene-specific elements in video content
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/50Context or environment of the image
    • G06V20/52Surveillance or monitoring of activities, e.g. for recognising suspicious objects

Definitions

  • This application relates to the field of artificial intelligence technology, in particular to methods, devices, computer equipment, and storage media for monitoring the number of livestock.
  • the purpose of the embodiments of the present application is to provide a method, device, computer equipment and storage medium for monitoring the number of livestock, so as to reduce the multi-detection rate and the missed-detection rate.
  • the embodiments of the present application provide a method for monitoring the number of livestock, which adopts the following technical solutions:
  • a method for monitoring the number of livestock including the following steps:
  • the pre-trained livestock classification model is used to determine the content of each redundant tracking box in turn. If the content of the redundant tracking box is determined to be a real livestock, the redundant tracking box is used as tracking to compensate for the missed detection frame;
  • an embodiment of the present application also provides a device for monitoring the number of livestock, which adopts the following technical solutions:
  • a device for monitoring the number of livestock including:
  • the acquisition module is used to control the camera installed in the livestock pen to perform video acquisition, and intercept the video according to the frame to obtain the collected picture;
  • the detection module is configured to perform livestock detection on the collected pictures according to the pre-trained target detection AI model to obtain at least one detection frame, wherein the content in the detection frame is an image of a single livestock determined by the model;
  • the acquisition module is used to intercept the image of a single livestock in the detection frame to obtain a sample image of the livestock, and compare all the sample images of the livestock in the current frame with all the sample images of the livestock in the previous frame through the pre-trained deep tracking model , Obtain tracking detection matching box and redundant tracking box;
  • the judging module is used to judge the content of each redundant tracking box in turn through a pre-trained livestock classification model. If the content of the redundant tracking box is determined to be a real animal, then the redundant tracking box As a tracking to make up for missed frames;
  • the calculation module is used to add the number of missing frames for tracking and compensation to the number of matching frames for tracking and detection to obtain the actual number of livestock as a statistical result;
  • the comparison module is used to compare the statistical results with the number of livestock stored in the database.
  • the alarm module is used to determine that the number of livestock is abnormal if the statistical result is inconsistent with the number of livestock stored in the database, and send an abnormal alarm to the user.
  • the embodiments of the present application also provide a computer device, which adopts the following technical solutions:
  • a computer device includes a memory and a processor, wherein computer readable instructions are stored in the memory, and when the processor executes the computer readable instructions, the steps of the method for monitoring the quantity of livestock as described below are implemented:
  • the pre-trained livestock classification model is used to determine the content of each redundant tracking box in turn. If the content of the redundant tracking box is determined to be a real livestock, the redundant tracking box is used as tracking to compensate for the missed detection frame;
  • the embodiments of the present application also provide a computer-readable storage medium, which adopts the following technical solutions:
  • a computer-readable storage medium having computer-readable instructions stored thereon, and when the computer-readable instructions are executed by a processor, the steps of the method for monitoring the quantity of livestock as described below are realized:
  • the pre-trained livestock classification model is used to determine the content of each redundant tracking box in turn. If the content of the redundant tracking box is determined to be a real livestock, the redundant tracking box is used as tracking to compensate for the missed detection frame;
  • This application is based on the target detection AI model, introduces a deep tracking model and a livestock classification model, so that it can make full use of the video information captured by the camera, and compare the current frame with the previous frame of the detected livestock image, and then obtain the
  • the tracking detection matching frame is used as one of the data of livestock statistics, instead of the follow-up calculation directly by the number of detection frames, which effectively reduces the multi-detection rate.
  • this application can automatically count the number of livestock in the breeding farm, automatically report the statistical results, and achieve the purpose of automatic real-time early warning. When the loss of livestock occurs in the later period, by finding the time point when the predicted number of livestock has fallen and playing back the video of the falling time period, it can help the farm find the cause and stop the loss in time.
  • Figure 1 is an exemplary system architecture diagram to which the present application can be applied;
  • FIG. 2 is a flowchart of an embodiment of the method for monitoring the number of livestock according to the present application
  • Figure 3 is a schematic structural diagram of an embodiment of the device for monitoring the number of livestock according to the present application.
  • Fig. 4 is a schematic structural diagram of an embodiment of a computer device according to the present application.
  • the system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105.
  • the network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105.
  • the network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, and so on.
  • the user can use the terminal devices 101, 102, and 103 to interact with the server 105 through the network 104 to receive or send messages and so on.
  • Various communication client applications such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social platform software, can be installed on the terminal devices 101, 102, and 103.
  • the terminal devices 101, 102, 103 may be various electronic devices with display screens and support for web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III, dynamic Video experts compress standard audio layer 3), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic image experts compress standard audio layer 4) players, laptop portable computers and desktop computers, etc.
  • MP3 players Moving Picture Experts Group Audio Layer III, dynamic Video experts compress standard audio layer 3
  • MP4 Moving Picture Experts Group Audio Layer IV, dynamic image experts compress standard audio layer 4
  • laptop portable computers and desktop computers etc.
  • the server 105 may be a server that provides various services, for example, a background server that provides support for pages displayed on the terminal devices 101, 102, and 103.
  • the method for monitoring the number of livestock provided in the embodiments of the present application is generally executed by a server/terminal device, and accordingly, the device for monitoring the number of livestock is generally set in the server/terminal device.
  • terminal devices, networks, and servers in FIG. 1 are merely illustrative. There can be any number of terminal devices, networks, and servers according to implementation needs.
  • the method for monitoring the number of livestock includes the following steps:
  • S1 Control the camera installed in the livestock pen to collect video, and intercept the video according to frames to obtain collected pictures.
  • a surveillance camera is deployed for each livestock pen that needs to be monitored, so as to achieve the purpose of collecting video of the livestock pen.
  • the camera is a wide-angle camera to ensure that images of the entire livestock pen can be collected.
  • the livestock in the livestock pen is not missed.
  • the livestock can be cattle, sheep, pigs and other animals.
  • the video is intercepted by frame to ensure that the time interval between the acquired pictures is small, and to avoid the situation that the time interval is long and the pictures have large differences.
  • the step of controlling the camera installed in the livestock pen to perform video collection includes:
  • the frame rate of the camera is generally 30 frames or 60 frames. Compared with the scheme of collecting multiple times per second, this application collects once per second, which reduces the density of data transmission. The duration of the collected video is 1 second, so that the video is not too large and the transmission speed is improved.
  • S2 Perform livestock detection on the collected pictures according to the pre-trained target detection AI model to obtain at least one detection frame, where the content in the detection frame is an image of a single livestock determined by the model.
  • the target detection AI model is a general object detection model.
  • the target detection AI model is pre-trained based on a data set of livestock types, and the object detection model is pre-built using one of the following algorithms Model: SSD algorithm, Fast RCNN algorithm, Faster RCNN algorithm.
  • the above three algorithms are algorithms in convolutional neural network technology. Which algorithm is used to construct the object detection model can be determined according to the actual needs of object detection.
  • the target detection AI model is used to ensure that the location of the livestock can be initially detected, and the livestock in the picture can be selected.
  • S3 Intercept the image of a single livestock in the detection frame to obtain a sample livestock image, and compare all the livestock sample images in the current frame with all the livestock sample images in the previous frame through a pre-trained deep tracking model to obtain tracking Detect matching boxes and redundant tracking boxes.
  • the depth tracking model is used for comparison, and the tracking detection matching frame and the redundant tracking frame are determined for subsequent livestock determination.
  • the deep tracking model includes a livestock re-identification model and a livestock tracking AI model.
  • the pre-trained deep tracking model compares all the livestock sample images in the current frame with all the livestock sample images in the previous frame,
  • the steps to obtain the tracking detection matching box and the redundant tracking box include:
  • the information of the motion matching degree has a better effect on detecting the number of livestock.
  • the similarity of the motion is first obtained through Kalman filtering. , Through the cascade matching to get the motion matching degree. Then, the apparent matching degree is extracted and calculated by the deep neural network. Through the motion matching degree and the apparent matching degree, the matching degree of the detection frame can be obtained frame by frame, so as to obtain the tracking detection matching frame and the redundant tracking frame of the current frame according to the matching degree.
  • the re-identification model adopts DSA-reID (Densely Aligned Person Re-identification) based on dense semantic alignment, which effectively solves the problem of spatial semantic misalignment that exists widely in re-identification, and significantly improves the re-identification technology.
  • DSA-reID Densely Aligned Person Re-identification
  • dense semantic alignment effectively solves the problem of spatial semantic misalignment that exists widely in re-identification, and significantly improves the re-identification technology.
  • the accuracy of the algorithm Dense semantics better solves the different shooting angles, obstacle occlusion, and large background differences in practical applications.
  • all the feature vectors of the current frame are compared with all the feature vectors of the previous frame to obtain the tracking detection matching frame and
  • the steps for the redundant tracking box include:
  • all the feature vectors of the current frame are compared with all the feature vectors of the previous frame, and the detection frame with the same feature vector of the current frame and the previous frame is compared As a matching box for tracking detection;
  • it also includes comparing all the feature vectors of the current frame with all the feature vectors in the tracking frame of the previous frame, and using the detection frames with different feature vectors of the current frame and the previous frame as redundant detection frames.
  • the tracking detection matching frame, redundant detection frame, and redundant detection frame are all obtained through the comparison between the feature vectors of the upper and lower frames, which ensures the accuracy of the matching result.
  • the step of obtaining the tracking detection matching frame and the redundant tracking frame of the current frame includes:
  • the detection frame with different feature vectors of the current frame and the previous frame is regarded as a redundant detection frame
  • the tracking detection matching frame and the redundant tracking frame are respectively used as the tracking detection matching frame and the redundant tracking frame of the current frame;
  • the sum of the number of tracking detection matching frames and the number of redundant detection frames should be the number of detection frames in the current frame.
  • the addition of q redundant detection frames and l tracking detection matching frames is equal to m detection frames, and the mathematical verification is performed. If the numbers are equal, the tracking detection matching frame and the redundant tracking frame are respectively used as the tracking detection matching frame and the redundant tracking frame of the current frame.
  • the feature vector comparison is not performed again, and the error report is directly sent to the relevant personnel.
  • each redundant tracking box is sequentially determined by the pre-trained livestock classification model, and if the content determination result in the redundant tracking box is a real livestock, the redundant tracking box is used as a tracking compensation Missing box.
  • a livestock classification model is used to determine the content of the redundant tracking frame. When it is determined to be a real animal, it is used as a tracking to compensate for the missed frame, which realizes the retrieval of a single frame. Image of missed livestock.
  • the pre-trained livestock classification model is used to determine the content in each redundant tracking box in turn, and if the content determination result in the redundant tracking box is a real animal, then the redundant tracking box is used as the tracking
  • the steps to make up for missing boxes include:
  • the redundant tracking frame whose classification probability is greater than the preset threshold is determined as the location of the real livestock, and the tracking frame is obtained to compensate for the missing detection frame.
  • the classification model is used to determine whether it is a real livestock, and a threshold is preset. When the classification probability exceeds the threshold, it is determined that the livestock in the current redundant tracking frame belongs to the livestock type of the application, and then it is determined to be a real livestock to prevent Other livestock that do not belong to the livestock type of this application are mixed into the livestock pen, thereby affecting the calculation of the number of livestock and reducing the multi-check rate of the computer output results.
  • the step of determining that the classification probability is greater than a preset threshold and the redundant tracking frame as the location of the real livestock includes:
  • the classification probability is calculated by the classification probability formula, and the redundant tracking frame with the classification probability greater than a preset threshold is determined as the location of the real livestock;
  • i is the category
  • e is the natural index
  • P is the probability
  • Vi is the output value of the classification network of the livestock classification model corresponding to the category i.
  • the classification probability is calculated to determine whether it is a real livestock to prevent misidentification and prevent multiple inspections.
  • the target detection AI model, the depth tracking model, and the livestock classification model are all pre-trained based on a dataset of livestock types.
  • Deep tracking models include livestock re-identification models and livestock tracking AI models.
  • the livestock re-identification model, the livestock tracking AI model and the livestock classification model are all general re-identification, tracking and classification models.
  • the number of tracking and compensation missed detection frames is added to the number of tracking detection matching frames, as the actual number of livestock, which reduces the multi-detection rate and missed detection rate of the computer for livestock detection.
  • the number of livestock that the farm should have is pre-stored in the database, and the statistical result is compared with the pre-stored number to determine whether there is a change in the number of livestock in the farm.
  • the method further includes:
  • the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.
  • this application provides an embodiment of a device for monitoring the quantity of livestock.
  • the device embodiment corresponds to the method embodiment shown in FIG. Specifically, it can be applied to various electronic devices.
  • the device 300 for monitoring the number of livestock in this embodiment includes: an acquisition module 301, a detection module 302, an acquisition module 303, a determination module 304, a calculation module 305, a comparison module 306, and an alarm module 307. in:
  • the collection module 301 is used to control a camera installed in the livestock pen to perform video collection, and intercept the video according to frames to obtain collected pictures;
  • the detection module 302 is configured to perform livestock detection on the collected pictures according to the pre-trained target detection AI model to obtain at least one detection frame, wherein the content in the detection frame is an image of a single livestock determined by the model;
  • the acquisition module 303 is used to intercept the image of a single livestock in the detection frame to obtain a sample image of the livestock, and compare all the sample images of the livestock in the current frame with all the sample images of the livestock in the previous frame through the pre-trained deep tracking model Compare, obtain tracking detection matching box and redundant tracking box;
  • the judging module 304 is used for judging the content in each redundant tracking box in turn through a pre-trained livestock classification model. If the content of the redundant tracking box is determined to be a real livestock, then tracking the redundant The frame is used as tracking to make up for missing frames;
  • the calculation module 305 is configured to add the number of missing frames for tracking compensation and the number of matching frames for tracking detection to obtain the actual number of livestock as a statistical result;
  • the comparison module 306 is used to compare the statistical result with the number of livestock stored in the database.
  • the alarm module 307 is configured to determine that the number of livestock is abnormal when the statistical result is inconsistent with the number of livestock stored in the database, and send an abnormal alarm to the user.
  • this application is based on the target detection AI model, introduces a deep tracking model and a livestock classification model, so that it can make full use of the time information captured by the camera, and compare the current frame with the previous frame of the detected livestock Image, determine the wrongly recognized detection frame to reduce the multi-detection rate; by determining the real livestock in the obtained redundant tracking frame, the missed detection rate is reduced, and the accuracy of the model's detection is improved.
  • the acquisition module 301 is also used to control a camera installed in the livestock pen to collect video on the livestock pen at a frequency of 1 time/second and a duration of 1 second/time.
  • the deep tracking model includes a livestock re-identification model and a livestock tracking AI model
  • the acquisition module 303 includes a re-identification sub-module and an acquisition sub-module.
  • the re-identification sub-module is used to perform livestock re-identification on the livestock sample image through the livestock re-identification model, and obtain feature vectors equal to the number of the detection frames.
  • the obtaining sub-module is used to compare all the feature vectors of the current frame with all the feature vectors of the previous frame through the motion matching degree and the apparent matching degree in the livestock tracking AI model to obtain the tracking detection match of the current frame Boxes and redundant tracking boxes.
  • the obtaining sub-module includes an input unit and a comparison unit.
  • the input unit is used to input collected pictures into the livestock tracking AI model, and the livestock tracking AI model outputs a tracking frame; the comparison unit is used to track the livestock
  • the motion matching degree and apparent matching degree in the AI model compare all the feature vectors of the current frame with all the feature vectors of the previous frame, and use the detection frame with the same feature vector of the current frame and the previous frame as the tracking detection matching frame ,
  • the tracking frame in the tracking frame of the previous frame and the feature vector of the detection frame of the current frame is taken as the redundant tracking frame, and the tracking detection matching frame and the redundant tracking frame of the current frame are obtained.
  • the comparison unit includes a comparison sub-unit, a calculation sub-unit, an output sub-unit, and a re-identification sub-unit, and the comparison sub-unit is used to use a detection frame with different feature vectors of the current frame and the previous frame as a redundant detection frame;
  • the calculation subunit is used to calculate whether the sum of the number of tracking detection matching frames and the number of redundant detection frames is consistent with the number of detection frames in the current frame;
  • the output subunit is used to match the tracking detection when the numbers are consistent
  • the frame and the redundant tracking frame are respectively used as the tracking detection matching frame and the redundant tracking frame of the current frame;
  • the re-identification subunit is used to repeat the feature vector comparison when the numbers are inconsistent to reacquire the tracking detection matching frame, the redundant tracking frame and the redundant tracking frame.
  • the detection frame is recalculated, and when the recalculated quantity is still inconsistent, an error report is sent to the designated person.
  • the determination module 304 includes an acquisition sub-module, a classification sub-module, and a threshold sub-module.
  • the acquisition sub-module is used to acquire the content in the redundant tracking frame of the collected picture, wherein the content in the redundant tracking frame is determined by the model An image of a single livestock;
  • the classification sub-module is used to sequentially classify the content in each of the redundant tracking boxes through the livestock classification model;
  • the threshold sub-module is used to classify the ones whose classification probability is greater than a preset threshold
  • the redundant tracking frame is determined as the location of the real livestock, and the tracking is obtained to make up for the missing frame.
  • FIG. 4 is a block diagram of the basic structure of the computer device in this embodiment.
  • the computer device 200 includes a memory 201, a processor 202, and a network interface 203 that are connected to each other in communication through a system bus. It should be pointed out that the figure only shows the computer device 200 with the components 201-203, but it should be understood that it is not required to implement all the shown components, and more or fewer components may be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculation and/or information processing in accordance with pre-set or stored instructions.
  • Its hardware includes, but is not limited to, a microprocessor, a dedicated Integrated Circuit (Application Specific Integrated Circuit, ASIC), Programmable Gate Array (Field-Programmable Gate Array, FPGA), Digital Processor (Digital Signal Processor, DSP), embedded equipment, etc.
  • ASIC Application Specific Integrated Circuit
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • DSP Digital Processor
  • the computer device may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server.
  • the computer device can interact with the user through a keyboard, a mouse, a remote control, a touch panel, or a voice control device.
  • the memory 201 includes at least one type of readable storage medium, the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static Random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disks, optical disks, etc.
  • the memory 201 may be an internal storage unit of the computer device 200, such as a hard disk or a memory of the computer device 200.
  • the memory 201 may also be an external storage device of the computer device 200, for example, a plug-in hard disk equipped on the computer device 200, a smart media card (SMC), and a secure digital (Secure Digital, SD) card, Flash Card, etc.
  • the memory 201 may also include both an internal storage unit of the computer device 200 and an external storage device thereof.
  • the memory 201 is generally used to store an operating system and various application software installed in the computer device 200, such as computer readable instructions for a method of monitoring the number of livestock.
  • the memory 201 can also be used to temporarily store various types of data that have been output or will be output.
  • the processor 202 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips.
  • the processor 202 is generally used to control the overall operation of the computer device 200.
  • the processor 202 is configured to run computer-readable instructions or process data stored in the memory 201, for example, computer-readable instructions for running the method for monitoring the number of livestock.
  • the network interface 203 may include a wireless network interface or a wired network interface, and the network interface 203 is generally used to establish a communication connection between the computer device 200 and other electronic devices.
  • the time information captured by the camera is fully utilized to determine the incorrectly recognized detection frame by comparing the detected livestock images of the current frame and the previous frame to reduce the multi-detection rate; The content in the real livestock is judged to reduce the missed detection rate, and then improve the accuracy of the model's detection.
  • the computer-readable storage medium may be non-volatile or volatile.
  • the computer-readable storage medium stores a process for monitoring the number of livestock, and the process for monitoring the number of livestock can be executed by at least one processor, so that the at least one processor executes the steps of the method for monitoring the number of livestock as described above.
  • the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • the time information captured by the camera is fully utilized to determine the incorrectly recognized detection frame by comparing the detected livestock images of the current frame and the previous frame to reduce the multi-detection rate; The content in the real livestock is judged to reduce the missed detection rate, and then improve the accuracy of the model's detection.
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to make a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in the various embodiments of the present application.
  • a terminal device which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

本申请实施例属于人工智能技术领域,应用于智慧城市中,具体应用于智慧农业,涉及一种监控牲畜数量的方法,包括根据目标检测AI模型对获得的采集图片进行牲畜检测,获得检测框;截取检测框中单个牲畜的图像,获得牲畜样例图,通过深度跟踪模型和牲畜分类模型对牲畜样例图进行处理,获得跟踪检测匹配框、多余跟踪框和跟踪弥补漏检框;将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;将统计结果与数据库中存储的牲畜数量进行对比;若统计结果与数据库中存储的牲畜数量不一致,判定牲畜数量异常,发送异常警报至用户。其中,获得的统计结果可存储在区块链中,本申请还提供一种监控牲畜数量的装置、计算机设备及存储介质。本申请有效降低了牲畜多检率和漏检率。

Description

监控牲畜数量的方法、装置、计算机设备及存储介质
本申请要求于2020年4月28日提交中国专利局、申请号为202010350129.0,发明名称为“监控牲畜数量的方法、装置、计算机设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域,尤其涉及监控牲畜数量的方法、装置、计算机设备及存储介质。
背景技术
传统的养殖场牲畜数目统计完全依赖人工,人工定时走到养殖场内对各个栏内的牲畜数量进行记录,而采用人工计算的方式容易发生计算错误,造成数量难以对应的情况发生;并且当发现牲畜数量异常时,养殖人员很难及时找到原因,造成养殖牲畜的丢失。
目前现代的养殖场引进自动化设备和智能化软件算法来提高养殖效率、提升养殖质量,发明人意识到现有的大部分方式是采用摄像头和单一的模型结合来进行养殖场牲畜数目统计并监控牲畜,在模型识别过程中可能会误将非牲畜的物体识别为牲畜,进而产生多检率的问题。而在对牲畜的数量进行模型统计的过程中,由于采用的模型不同,导致牲畜漏检的情况发生。
发明内容
本申请实施例的目的在于提出一种监控牲畜数量的方法、装置、计算机设备及存储介质,减小多检率和漏检率的情况。
为了解决上述技术问题,本申请实施例提供一种监控牲畜数量的方法,采用了如下所述的技术方案:
一种监控牲畜数量的方法,包括下述步骤:
控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
将所述统计结果与数据库中存储的牲畜数量进行对比;以及
若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
为了解决上述技术问题,本申请实施例还提供一种监控牲畜数量的装置,采用了如下所述的技术方案:
一种监控牲畜数量的装置,包括:
采集模块,用于控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
检测模块,用于根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
获取模块,用于截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
判定模块,用于通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
计算模块,用于将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
对比模块,用于将所述统计结果与数据库中存储的牲畜数量进行对比;以及
告警模块,用于若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
为了解决上述技术问题,本申请实施例还提供一种计算机设备,采用了如下所述的技术方案:
一种计算机设备,包括存储器和处理器,所述存储器中存储有计算机可读指令,所述处理器执行所述计算机可读指令时实现如下所述的监控牲畜数量的方法的步骤:
控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
将所述统计结果与数据库中存储的牲畜数量进行对比;以及
若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
为了解决上述技术问题,本申请实施例还提供一种计算机可读存储介质,采用了如下所述的技术方案:
一种计算机可读存储介质,所述计算机可读存储介质上存储有计算机可读指令,所述计算机可读指令被处理器执行时实现如下所述的监控牲畜数量的方法的步骤:
控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
将所述统计结果与数据库中存储的牲畜数量进行对比;以及
若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
与现有技术相比,本申请实施例主要有以下有益效果:
本申请以目标检测AI模型为基础,引入深度跟踪模型和牲畜分类模型,使之得以充分利用摄像头捕捉到的视频信息,通过对比当前帧与上一帧的检测到的牲畜图像,进而将获得的跟踪检测匹配框作为牲畜统计的数据之一,而不是直接通过检测框的数量来进行后续计算,有效降低多检率。通过对获得的多余跟踪框中的内容进行真牲畜判定,来降低漏检率,进而提高模型的检测的准确率。同时,本申请能够自动统计养殖场中的牲畜数量,自动上报统计结果,并达到自动实时预警的目的。后期发生牲畜丢失的时间时,通过找到牲畜预测数量下降的时间点和回放下降时间段的视频,能帮助养殖场找到原因,及时止损。
附图说明
为了更清楚地说明本申请中的方案,下面将对本申请实施例描述中所需要使用的附图作一个简单介绍,显而易见地,下面描述中的附图是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1是本申请可以应用于其中的示例性系统架构图;
图2是根据本申请的监控牲畜数量的方法的一个实施例的流程图;
图3是根据本申请的监控牲畜数量的装置的一个实施例的结构示意图;
图4是根据本申请的计算机设备的一个实施例的结构示意图。
附图标记:200、计算机设备;201、存储器;202、处理器;203、网络接口;300、监控牲畜数量的装置;301、采集模块;302、检测模块;303、获取模块;304、判定模块;305、计算模块;306、对比模块;307、告警模块。
具体实施方式
除非另有定义,本文所使用的所有的技术和科学术语与属于本申请的技术领域的技术人员通常理解的含义相同;本文中在申请的说明书中所使用的术语只是为了描述具体的实施例的目的,不是旨在于限制本申请;本申请的说明书和权利要求书及上述附图说明中的术语“包括”和“具有”以及它们的任何变形,意图在于覆盖不排他的包含。本申请的说明书和权利要求书或上述附图中的术语“第一”、“第二”等是用于区别不同对象,而不是用于描述特定顺序。
在本文中提及“实施例”意味着,结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例,也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是,本文所描述的实施例可以与其它实施例相结合。
为了使本技术领域的人员更好地理解本申请方案,下面将结合附图,对本申请实施例中的技术方案进行清楚、完整地描述。
如图1所示,系统架构100可以包括终端设备101、102、103,网络104和服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型,例如有线、无线通信链路或者光纤电缆等等。
用户可以使用终端设备101、102、103通过网络104与服务器105交互,以接收或发送消息等。终端设备101、102、103上可以安装有各种通讯客户端应用,例如网页浏览器应用、购物类应用、搜索类应用、即时通信工具、邮箱客户端、社交平台软件等。
终端设备101、102、103可以是具有显示屏并且支持网页浏览的各种电子设备,包括但不限于智能手机、平板电脑、电子书阅读器、MP3播放器(Moving Picture Experts Group Audio Layer III,动态影像专家压缩标准音频层面3)、MP4(Moving Picture Experts Group Audio Layer IV,动态影像专家压缩标准音频层面4)播放器、膝上型便携计算机和台式计算机等等。
服务器105可以是提供各种服务的服务器,例如对终端设备101、102、103上显示的页面提供支持的后台服务器。
需要说明的是,本申请实施例所提供的监控牲畜数量的方法一般由服务器/终端设备执行,相应地,监控牲畜数量的装置一般设置于服务器/终端设备中。
应该理解,图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要,可以具有任意数目的终端设备、网络和服务器。
继续参考图2,示出了根据本申请的监控牲畜数量的方法的一个实施例的流程图。所述的监控牲畜数量的方法,包括以下步骤:
S1:控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片。
在本实施例中,针对每个需要监控的牲畜圈部署监控摄像头,从而实现采集牲畜圈视频的目的,其中,摄像头为广角摄像头,以保证可以采集到的整个牲畜圈的图像,在采集视频的过程中,不遗漏牲畜圈中的牲畜,其中,牲畜可以为牛、羊、猪等动物。按帧截取视频,以保证获得的采集图片之间的时间间隔小,避免出现时间间隔长,造成图片具有较大差异的情况。
其中,所述控制安装于牲畜圈的摄像头进行视频采集的步骤包括:
控制安装于牲畜圈的摄像头以1次/秒的频率,1秒/次的时长对牲畜圈进行视频采集。
在本实施例中,摄像头的帧率一般是30帧或60帧,相对于每秒采集多次的方案,本申请每秒采集一次,减少了数据传输的密度。采集的所述视频的时长为1秒,使得视频不会过大,提升传输速度。
S2:根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像。
在本实施例中,目标检测AI模型为通用的物体检测模型,作为预训练模型,目标检测AI模型基于牲畜类型的数据集进行预训练,物体检测模型为采用以下算法中的一者预先构建的模型:SSD算法、Fast RCNN算法、Faster RCNN算法。以上三种算法都是卷积神经网络技术中的算法,具体采用哪种算法来构建物体检测模型,可以根据物体检测的实际需求而定。采用目标检测AI模型,保证可以初步检测出牲畜的位置,框选出图片中的牲畜。
S3:截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框。
在本实施例中,通过深度跟踪模型进行对比,确定出跟踪检测匹配框和多余跟踪框,以便后续进行牲畜判定。
具体的,所述深度跟踪模型包括牲畜重识别模型和牲畜跟踪AI模型,所述通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框的步骤包括:
通过所述牲畜重识别模型对所述牲畜样例图进行牲畜重识别,获得与所述检测框数量相等的特征向量;
通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框。
在本实施例中,因为养殖场牲畜的移动速度和摄像头帧率相比不快,运动匹配度的信息对于检测牲畜的数量具有较好的作用,具体的,首先通过卡尔曼滤波得到运动的相似性,通过级联匹配得到运动匹配度。然后通过深度神经网络提取计算表观匹配度。通过运动匹配度和表观匹配度可以逐帧得到检测框的匹配度,从而根据匹配度的情况获得当前帧的跟踪检测匹配框和多余跟踪框。所述重识别模型采用基于密集语义对齐的重识别模型DSA-reID(Densely Semantically Aligned Person Re-identfication),有效地解决了重识别中广泛存在的空间语义不对齐问题,显著地提高了重识别技术的算法精度。密集语义更好的解决了实际应用中存在的拍摄视角各异、障碍物遮挡和背景差异大等。
进一步的,所述通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框的步骤包括:
将采集图片输入所述牲畜跟踪AI模型,所述牲畜跟踪AI模型输出跟踪框;
通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,将当前帧与上一帧的特征向量相同的检测框作为跟踪检测匹配框;
对比当前帧的所有特征向量与上一帧跟踪框中的所有特征向量,将上一帧跟踪框中与当前帧检测框特征向量不同的跟踪框作为多余跟踪框,获得当前帧的跟踪检测匹配框和多余跟踪框。
在本实施例中,还包括,对比当前帧的所有特征向量与上一帧跟踪框中的所有特征向量,将当前帧与上一帧的特征向量不同的检测框作为多余检测框。跟踪检测匹配框、多余检测框、多余检测框都是通过上下帧的特征向量之间的对比,进而获得的,保证了匹配结果的准确性。
其中,所述获得当前帧的跟踪检测匹配框和多余跟踪框的步骤包括:
将当前帧与上一帧的特征向量不同的检测框作为多余检测框;
计算所述跟踪检测匹配框的数量和多余检测框的数量之和与当前帧检测框的数量是否一致;
若数量一致,则将所述跟踪检测匹配框和多余跟踪框分别作为当前帧的跟踪检测匹配框和多余跟踪框;
若数量不一致,则重复进行特征向量对比以重新获取跟踪检测匹配框、多余跟踪框和多余检测框进行重新计算,并在重新计算后的数量仍然不一致时,发送错误报告给指定人员。
在本实施例中,跟踪检测匹配框的个数与多余检测框的个数相加应为当前帧检测框的个数。多余检测框q个和跟踪检测匹配框l个相加等于检测框m个,进行数学上的验算。如果数字相等,则将所述跟踪检测匹配框和多余跟踪框分别作为当前帧的跟踪检测匹配框和多余跟踪框。如果数字不对等,则认为获得的多余检测框q或跟踪检测匹配框l出现错误,重新进行特征向量对比以重新获取跟踪检测匹配框、多余跟踪框和多余检测框进行重新计算,重新计算后的数量仍然不一致时发送错误报告给相关人员,以实现检测过程中的验证,降低计算机的出错率。
在其他实施例中,也可也在数字不对等时,不再重新进行特征向量比对,而直接发送错误报告给相关人员。
S4:通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框。
在本实施例中,能作为的很好的补充,通过一个牲畜分类模型对多余跟踪框中的内容进行判定,判定为真牲畜时,则作为跟踪弥补漏检框,实现了找回针对单帧图像的漏检牲畜。
其中,所述通过预先训练的牲畜分类模型依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框的步骤包括:
获取所述采集图片的多余跟踪框中的内容,其中,多余跟踪框中的内容为模型判定的单个牲畜的图像;
通过所述牲畜分类模型依次对每个所述多余跟踪框中的内容进行分类;
将分类概率大于预设阈值的所述多余跟踪框判定为真牲畜所在位置,获得跟踪弥补漏检框。
在本实施例中,通过分类模型判定是否为真牲畜,预设一阈值,当分类概率超过该阈 值,则判定当前多余跟踪框中的牲畜属于本申请的牲畜类型,则判定为真牲畜,防止其他不属于本申请的牲畜类型的牲畜混入牲畜圈,进而影响牲畜数量的计算,降低计算机输出结果的多检率。
进一步的,所述将分类概率大于预设的阈值所述多余跟踪框判定为真牲畜所在位置的步骤包括:
通过分类概率公式计算分类概率,将分类概率大于预设的阈值所述多余跟踪框判定为真牲畜所在位置;
所述分类概率的计算公式为
Figure PCTCN2020105770-appb-000001
其中,i为类别,e为自然指数,P为概率,Vi为类别i所对应的所述牲畜分类模型的分类网络输出值。
在本实施例中,通过分类概率的计算,以判断是否为真牲畜,防止误识别的情况发生,防止造成多检的情况。
S5:将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果。
在本实施例中,所述目标检测AI模型、深度跟踪模型和牲畜分类模型均是基于牲畜类型的数据集进行预先训练。深度跟踪模型包括牲畜重识别模型和牲畜跟踪AI模型。牲畜重识别模型、牲畜跟踪AI模型和牲畜分类模型均为通用的重识别、跟踪及分类模型。跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,作为实际牲畜数量,降低了计算机对牲畜检测的多检率和漏检率。
S6:将所述统计结果与数据库中存储的牲畜数量进行对比。
在本实施例中,数据库中预存有养殖场应有的牲畜数量,将统计结果哦与预存的数量进行对比,确定养殖场的牲畜数量是否有变化。
S7:若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
在本实施例中,发生牲畜丢失时,及时提醒用户,通过找到牲畜预测数量下降的时间点和回放下降时间段的视频,能帮助养殖场找到原因,及时止损。
其中,在所述将所述统计结果与数据库中存储的牲畜数量进行对比的步骤之后,还包括:
若所述统计结果与数据库中存储的牲畜数量一致,则判定牲畜数量正常。
在本实施例中,数量一致,则判定牲畜没有增加或减少,继续对养殖场进行监测。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机可读指令来指令相关的硬件来完成,该计算机可读指令可存储于一计算机可读取存储介质中,该可读指令在执行时,可包括如上述各方法的实施例的流程。其中,前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory,ROM)等非易失性存储介质,或随机存储记忆体(Random Access Memory,RAM)等。
应该理解的是,虽然附图的流程图中的各个步骤按照箭头的指示依次显示,但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明,这些步骤的执行并没有严格的顺序限制,其可以以其他的顺序执行。而且,附图的流程图中的至少一部分步骤可以包括多个子步骤或者多个阶段,这些子步骤或者阶段并不必然是在同一时刻执行完成,而是可以在不同的时刻执行,其执行顺序也不必然是依次进行,而是可以与其他步骤或者其他步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
进一步参考图3,作为对上述图2所示方法的实现,本申请提供了一种监控牲畜数量的装置的一个实施例,该装置实施例与图2所示的方法实施例相对应,该装置具体可以应用于各种电子设备中。
如图3所示,本实施例所述的监控牲畜数量的装置300包括:采集模块301、检测模 块302、获取模块303、判定模块304、计算模块305、对比模块306和告警模块307。其中:
采集模块301,用于控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
检测模块302,用于根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
获取模块303,用于截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
判定模块304,用于通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
计算模块305,用于将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
对比模块306,用于将所述统计结果与数据库中存储的牲畜数量进行对比;以及
告警模块307,用于当所述统计结果与数据库中存储的牲畜数量不一致时,判定牲畜数量异常,发送异常警报至用户。
在本实施例中,本申请以目标检测AI模型为基础,引入深度跟踪模型和牲畜分类模型,使之得以充分利用摄像头捕捉到的时间信息,通过对比当前帧与上一帧的检测到的牲畜图像,确定错误识别的检测框,来降低多检率;通过对获得的多余跟踪框中的内容进行真牲畜判定,来降低漏检率,进而提高模型的检测的准确率。
其中,所述采集模块301还用于控制安装于牲畜圈的摄像头以1次/秒的频率,1秒/次的时长对牲畜圈进行视频采集。所述深度跟踪模型包括牲畜重识别模型和牲畜跟踪AI模型,所述获取模块303包括重识别子模块和获得子模块。其中,所述重识别子模块用于通过所述牲畜重识别模型对所述牲畜样例图进行牲畜重识别,获得与所述检测框数量相等的特征向量。所述获得子模块用于通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框。
所述获得子模块包括输入单元和对比单元,所述输入单元用于将采集图片输入所述牲畜跟踪AI模型,所述牲畜跟踪AI模型输出跟踪框;所述对比单元用于通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,将当前帧与上一帧的特征向量相同的检测框作为跟踪检测匹配框,将上一帧跟踪框中与当前帧检测框特征向量不同的跟踪框作为多余跟踪框,获得当前帧的跟踪检测匹配框和多余跟踪框。
所述对比单元包括对比子单元、计算子单元、输出子单元和重新识别子单元,所述对比子单元用于将当前帧与上一帧的特征向量不同的检测框作为多余检测框;所述计算子单元用于计算所述跟踪检测匹配框的数量和多余检测框的数量之和与当前帧检测框的数量是否一致;所述输出子单元用于在数量一致时,将所述跟踪检测匹配框和多余跟踪框分别作为当前帧的跟踪检测匹配框和多余跟踪框;所述重新识别子单元用于在数量不一致时,重复进行特征向量对比以重新获取跟踪检测匹配框、多余跟踪框和多余检测框进行重新计算,并在重新计算后的数量仍然不一致时,发送错误报告给指定人员。
所述判定模块304包括获取子模块、分类子模块和阈值子模块,所述获取子模块用于获取所述采集图片的多余跟踪框中的内容,其中,多余跟踪框中的内容为模型判定的单个牲畜的图像;所述分类子模块用于通过所述牲畜分类模型依次对每个所述多余跟踪框中的内容进行分类;所述阈值子模块用于将分类概率大于预设阈值的所述多余跟踪框判定为真牲畜所在位置,获得跟踪弥补漏检框。
为解决上述技术问题,本申请实施例还提供计算机设备。具体请参阅图4,图4为本实施例计算机设备基本结构框图。
所述计算机设备200包括通过系统总线相互通信连接存储器201、处理器202、网络接口203。需要指出的是,图中仅示出了具有组件201-203的计算机设备200,但是应理解的是,并不要求实施所有示出的组件,可以替代的实施更多或者更少的组件。其中,本技术领域技术人员可以理解,这里的计算机设备是一种能够按照事先设定或存储的指令,自动进行数值计算和/或信息处理的设备,其硬件包括但不限于微处理器、专用集成电路(Application Specific Integrated Circuit,ASIC)、可编程门阵列(Field-Programmable Gate Array,FPGA)、数字处理器(Digital Signal Processor,DSP)、嵌入式设备等。
所述计算机设备可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。所述计算机设备可以与用户通过键盘、鼠标、遥控器、触摸板或声控设备等方式进行人机交互。
所述存储器201至少包括一种类型的可读存储介质,所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如,SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中,所述存储器201可以是所述计算机设备200的内部存储单元,例如该计算机设备200的硬盘或内存。在另一些实施例中,所述存储器201也可以是所述计算机设备200的外部存储设备,例如该计算机设备200上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。当然,所述存储器201还可以既包括所述计算机设备200的内部存储单元也包括其外部存储设备。本实施例中,所述存储器201通常用于存储安装于所述计算机设备200的操作系统和各类应用软件,例如监控牲畜数量的方法的计算机可读指令等。此外,所述存储器201还可以用于暂时地存储已经输出或者将要输出的各类数据。
所述处理器202在一些实施例中可以是中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器202通常用于控制所述计算机设备200的总体操作。本实施例中,所述处理器202用于运行所述存储器201中存储的计算机可读指令或者处理数据,例如运行所述监控牲畜数量的方法的计算机可读指令。
所述网络接口203可包括无线网络接口或有线网络接口,该网络接口203通常用于在所述计算机设备200与其他电子设备之间建立通信连接。
在本实施例中,充分利用摄像头捕捉到的时间信息,通过对比当前帧与上一帧的检测到的牲畜图像,确定错误识别的检测框,来降低多检率;通过对获得的多余跟踪框中的内容进行真牲畜判定,来降低漏检率,进而提高模型的检测的准确率。
本申请还提供了另一种实施方式,即提供一种计算机可读存储介质,所述计算机可读存储介质可以是非易失性,也可以是易失性。所述计算机可读存储介质存储有监控牲畜数量的流程,所述监控牲畜数量的流程可被至少一个处理器执行,以使所述至少一个处理器执行如上述的监控牲畜数量的方法的步骤。
需要强调的是,为进一步保证上述统计结果的私密和安全性,上述统计结果还可以存储于一区块链的节点中。本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain),本质上是一个去中心化的数据库,是一串使用密码学方法相关联产生的数据块,每一个数据块中包含了一批次网络交易的信息,用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
在本实施例中,充分利用摄像头捕捉到的时间信息,通过对比当前帧与上一帧的检测到的牲畜图像,确定错误识别的检测框,来降低多检率;通过对获得的多余跟踪框中的内容进行真牲畜判定,来降低漏检率,进而提高模型的检测的准确率。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法。
显然,以上所描述的实施例仅仅是本申请一部分实施例,而不是全部的实施例,附图中给出了本申请的较佳实施例,但并不限制本申请的专利范围。本申请可以以许多不同的形式来实现,相反地,提供这些实施例的目的是使对本申请的公开内容的理解更加透彻全面。尽管参照前述实施例对本申请进行了详细的说明,对于本领域的技术人员来而言,其依然可以对前述各具体实施方式所记载的技术方案进行修改,或者对其中部分技术特征进行等效替换。凡是利用本申请说明书及附图内容所做的等效结构,直接或间接运用在其他相关的技术领域,均同理在本申请专利保护范围之内。

Claims (20)

  1. 一种监控牲畜数量的方法,包括下述步骤:
    控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
    根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
    截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
    通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
    将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
    将所述统计结果与数据库中存储的牲畜数量进行对比;以及
    若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
  2. 根据权利要求1所述的监控牲畜数量的方法,其中,所述深度跟踪模型包括牲畜重识别模型和牲畜跟踪AI模型,所述通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框的步骤包括:
    通过所述牲畜重识别模型对所述牲畜样例图进行牲畜重识别,获得与所述检测框数量相等的特征向量;
    通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框。
  3. 根据权利要求2所述的监控牲畜数量的方法,其中,所述通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框的步骤包括:
    将采集图片输入所述牲畜跟踪AI模型,所述牲畜跟踪AI模型输出跟踪框;
    通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,将当前帧与上一帧的特征向量相同的检测框作为跟踪检测匹配框;
    对比当前帧的所有特征向量与上一帧跟踪框中的所有特征向量,将上一帧跟踪框中与当前帧检测框特征向量不同的跟踪框作为多余跟踪框,获得当前帧的跟踪检测匹配框和多余跟踪框。
  4. 根据权利要求3所述的监控牲畜数量的方法,其中,所述获得当前帧的跟踪检测匹配框和多余跟踪框的步骤包括:
    将当前帧与上一帧的特征向量不同的检测框作为多余检测框;
    计算所述跟踪检测匹配框的数量和多余检测框的数量之和与当前帧检测框的数量是否一致;
    若数量一致,则将所述跟踪检测匹配框和多余跟踪框分别作为当前帧的跟踪检测匹配框和多余跟踪框;
    若数量不一致,则重复进行特征向量对比以重新获取跟踪检测匹配框、多余跟踪框和多余检测框进行重新计算,并在重新计算后的数量仍然不一致时,发送错误报告给指定人员。
  5. 根据权利要求1所述的监控牲畜数量的方法,其中,所述通过预先训练的牲畜分类模型依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框的步骤包括:
    获取所述采集图片的多余跟踪框中的内容,其中,多余跟踪框中的内容为模型判定的 单个牲畜的图像;
    通过所述牲畜分类模型依次对每个所述多余跟踪框中的内容进行分类;
    将分类概率大于预设阈值的所述多余跟踪框判定为真牲畜所在位置,获得跟踪弥补漏检框。
  6. 根据权利要求5所述的监控牲畜数量的方法,其中,所述将分类概率大于预设的阈值所述多余跟踪框判定为真牲畜所在位置的步骤包括:
    通过分类概率公式计算分类概率,将分类概率大于预设的阈值所述多余跟踪框判定为真牲畜所在位置;
    所述分类概率的计算公式为
    Figure PCTCN2020105770-appb-100001
    其中,i为类别,e为自然指数,P为概率,Vi为类别i所对应的所述牲畜分类模型的分类网络输出值。
  7. 根据权利要求1所述的监控牲畜数量的方法,其中,所述控制安装于牲畜圈的摄像头进行视频采集的步骤包括:
    控制安装于牲畜圈的摄像头以1次/秒的频率,1秒/次的时长对牲畜圈进行视频采集。
  8. 一种监控牲畜数量的装置,包括:
    采集模块,用于控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
    检测模块,用于根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
    获取模块,用于截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
    判定模块,用于通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
    计算模块,用于将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
    对比模块,用于将所述统计结果与数据库中存储的牲畜数量进行对比;以及
    告警模块,用于若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
  9. 一种计算机设备,包括存储器和处理器,所述存储器中存储有计算机可读指令,所述处理器执行所述计算机可读指令时实现如下所述的监控牲畜数量的方法的步骤:
    控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
    根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
    截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
    通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
    将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
    将所述统计结果与数据库中存储的牲畜数量进行对比;以及
    若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
  10. 根据权利要求9所述的计算机设备,其中,所述深度跟踪模型包括牲畜重识别模型和牲畜跟踪AI模型,所述通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框的步骤包括:
    通过所述牲畜重识别模型对所述牲畜样例图进行牲畜重识别,获得与所述检测框数量相等的特征向量;
    通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框。
  11. 根据权利要求10所述的计算机设备,其中,所述通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框的步骤包括:
    将采集图片输入所述牲畜跟踪AI模型,所述牲畜跟踪AI模型输出跟踪框;
    通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,将当前帧与上一帧的特征向量相同的检测框作为跟踪检测匹配框;
    对比当前帧的所有特征向量与上一帧跟踪框中的所有特征向量,将上一帧跟踪框中与当前帧检测框特征向量不同的跟踪框作为多余跟踪框,获得当前帧的跟踪检测匹配框和多余跟踪框。
  12. 根据权利要求11所述的计算机设备,其中,所述获得当前帧的跟踪检测匹配框和多余跟踪框的步骤包括:
    将当前帧与上一帧的特征向量不同的检测框作为多余检测框;
    计算所述跟踪检测匹配框的数量和多余检测框的数量之和与当前帧检测框的数量是否一致;
    若数量一致,则将所述跟踪检测匹配框和多余跟踪框分别作为当前帧的跟踪检测匹配框和多余跟踪框;
    若数量不一致,则重复进行特征向量对比以重新获取跟踪检测匹配框、多余跟踪框和多余检测框进行重新计算,并在重新计算后的数量仍然不一致时,发送错误报告给指定人员。
  13. 根据权利要求9所述的计算机设备,其中,所述通过预先训练的牲畜分类模型依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框的步骤包括:
    获取所述采集图片的多余跟踪框中的内容,其中,多余跟踪框中的内容为模型判定的单个牲畜的图像;
    通过所述牲畜分类模型依次对每个所述多余跟踪框中的内容进行分类;
    将分类概率大于预设阈值的所述多余跟踪框判定为真牲畜所在位置,获得跟踪弥补漏检框。
  14. 根据权利要求13所述的计算机设备,其中,所述将分类概率大于预设的阈值所述多余跟踪框判定为真牲畜所在位置的步骤包括:
    通过分类概率公式计算分类概率,将分类概率大于预设的阈值所述多余跟踪框判定为真牲畜所在位置;
    所述分类概率的计算公式为
    Figure PCTCN2020105770-appb-100002
    其中,i为类别,e为自然指数,P为概率,Vi为类别i所对应的所述牲畜分类模型的分类网络输出值。
  15. 根据权利要求9所述的计算机设备,其中,所述控制安装于牲畜圈的摄像头进行视频采集的步骤包括:
    控制安装于牲畜圈的摄像头以1次/秒的频率,1秒/次的时长对牲畜圈进行视频采集。
  16. 一种计算机可读存储介质,所述计算机可读存储介质上存储有计算机可读指令,所述计算机可读指令被处理器执行时实现如下所述的监控牲畜数量的方法的步骤:
    控制安装于牲畜圈的摄像头进行视频采集,并对所述视频按帧截取,获得采集图片;
    根据预先训练的目标检测AI模型对所述采集图片进行牲畜检测,获得至少一个检测框,其中,检测框中的内容为模型判定的单个牲畜的图像;
    截取所述检测框中单个牲畜的图像,获得牲畜样例图,通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框;
    通过预先训练的牲畜分类模型对依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框;
    将获得的跟踪弥补漏检框的数量与跟踪检测匹配框的数量进行相加,获得实际牲畜数量,作为统计结果;
    将所述统计结果与数据库中存储的牲畜数量进行对比;以及
    若所述统计结果与数据库中存储的牲畜数量不一致,则判定牲畜数量异常,发送异常警报至用户。
  17. 根据权利要求16所述的计算机可读存储介质,其中,所述深度跟踪模型包括牲畜重识别模型和牲畜跟踪AI模型,所述通过预先训练的深度跟踪模型将当前帧的所有牲畜样例图与上一帧的所有牲畜样例图进行对比,获得跟踪检测匹配框和多余跟踪框的步骤包括:
    通过所述牲畜重识别模型对所述牲畜样例图进行牲畜重识别,获得与所述检测框数量相等的特征向量;
    通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框。
  18. 根据权利要求17所述的计算机可读存储介质,其中,所述通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,获得当前帧的跟踪检测匹配框和多余跟踪框的步骤包括:
    将采集图片输入所述牲畜跟踪AI模型,所述牲畜跟踪AI模型输出跟踪框;
    通过所述牲畜跟踪AI模型中的运动匹配度和表观匹配度,对当前帧的所有特征向量与上一帧的所有特征向量进行对比,将当前帧与上一帧的特征向量相同的检测框作为跟踪检测匹配框;
    对比当前帧的所有特征向量与上一帧跟踪框中的所有特征向量,将上一帧跟踪框中与当前帧检测框特征向量不同的跟踪框作为多余跟踪框,获得当前帧的跟踪检测匹配框和多余跟踪框。
  19. 根据权利要求18所述的计算机可读存储介质,其中,所述获得当前帧的跟踪检测匹配框和多余跟踪框的步骤包括:
    将当前帧与上一帧的特征向量不同的检测框作为多余检测框;
    计算所述跟踪检测匹配框的数量和多余检测框的数量之和与当前帧检测框的数量是否一致;
    若数量一致,则将所述跟踪检测匹配框和多余跟踪框分别作为当前帧的跟踪检测匹配框和多余跟踪框;
    若数量不一致,则重复进行特征向量对比以重新获取跟踪检测匹配框、多余跟踪框和多余检测框进行重新计算,并在重新计算后的数量仍然不一致时,发送错误报告给指定人员。
  20. 根据权利要求16所述的计算机可读存储介质,其中,所述通过预先训练的牲畜分类模型依次对每个所述多余跟踪框中的内容进行判定,若所述多余跟踪框中的内容判定结果为真牲畜,则将所述多余跟踪框作为跟踪弥补漏检框的步骤包括:
    获取所述采集图片的多余跟踪框中的内容,其中,多余跟踪框中的内容为模型判定的单个牲畜的图像;
    通过所述牲畜分类模型依次对每个所述多余跟踪框中的内容进行分类;
    将分类概率大于预设阈值的所述多余跟踪框判定为真牲畜所在位置,获得跟踪弥补漏检框。
PCT/CN2020/105770 2020-04-28 2020-07-30 监控牲畜数量的方法、装置、计算机设备及存储介质 WO2021217934A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010350129.0 2020-04-28
CN202010350129.0A CN111680551A (zh) 2020-04-28 2020-04-28 监控牲畜数量的方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2021217934A1 true WO2021217934A1 (zh) 2021-11-04

Family

ID=72433890

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/105770 WO2021217934A1 (zh) 2020-04-28 2020-07-30 监控牲畜数量的方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN111680551A (zh)
WO (1) WO2021217934A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114494863A (zh) * 2022-01-12 2022-05-13 北京小龙潜行科技有限公司 基于Blend Mask算法的动物幼崽计数方法以及装置
CN115861644A (zh) * 2022-11-21 2023-03-28 广东鉴面智能科技有限公司 一种生物多样性识别分析方法、系统及介质
CN116363494A (zh) * 2023-05-31 2023-06-30 睿克环境科技(中国)有限公司 一种鱼类数量监测、洄游跟踪的方法和系统

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112101291B (zh) * 2020-09-27 2024-01-30 成都睿畜电子科技有限公司 一种牲畜的护理方法、装置、介质及电子设备
CN112150506A (zh) * 2020-09-27 2020-12-29 成都睿畜电子科技有限公司 一种目标状态的检测方法、装置、介质及电子设备
CN112734730B (zh) * 2021-01-11 2023-07-28 牧原食品股份有限公司 一种牲畜数量识别方法、装置、设备和存储介质
CN112906483B (zh) * 2021-01-25 2024-01-23 中国银联股份有限公司 一种目标重识别方法、装置及计算机可读存储介质
CN113219964A (zh) * 2021-03-30 2021-08-06 广州朗国电子科技有限公司 禽舍环境巡检及调控方法、设备以及介质
CN114120185B (zh) * 2021-11-16 2022-08-09 东莞先知大数据有限公司 一种三鸟档清笼确定方法、电子设备和存储介质
CN113984767A (zh) * 2021-11-24 2022-01-28 牧原肉食品有限公司 牲畜胴体品质检测的系统、方法、设备和计算机存储介质
CN115100807A (zh) * 2022-06-17 2022-09-23 贵州东彩供应链科技有限公司 一种基于摄像头异常监控告警实现动物养殖场监管的系统
CN115063378B (zh) * 2022-06-27 2023-12-05 中国平安财产保险股份有限公司 智能点数方法、装置、设备及存储介质
CN115546192B (zh) * 2022-11-03 2023-03-21 中国平安财产保险股份有限公司 牲畜数量识别方法、装置、设备及存储介质
CN115601401B (zh) * 2022-12-01 2023-04-07 中国平安财产保险股份有限公司 基于牲畜群体运动特性的跟踪计数方法及其相关设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9084411B1 (en) * 2014-04-10 2015-07-21 Animal Biotech Llc Livestock identification system and method
CN108932496A (zh) * 2018-07-03 2018-12-04 北京佳格天地科技有限公司 区域内目标物的数量统计方法及装置
WO2020007363A1 (zh) * 2018-07-06 2020-01-09 京东数字科技控股有限公司 识别目标数量的方法、装置及计算机可读存储介质
CN111008561A (zh) * 2019-10-31 2020-04-14 重庆小雨点小额贷款有限公司 一种牲畜的数量确定方法、终端及计算机存储介质
CN111008560A (zh) * 2019-10-31 2020-04-14 重庆小雨点小额贷款有限公司 一种牲畜的重量确定方法、装置、终端及计算机存储介质

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018223102A1 (en) * 2017-06-02 2018-12-06 Performance Livestock Analytics, Inc. Adaptive livestock growth modeling using machine learning approaches to predict growth and recommend livestock management operations and activities
CN107680080B (zh) * 2017-09-05 2020-02-21 翔创科技(北京)有限公司 牲畜的样本库建立方法和清点方法、存储介质和电子设备
BR112021007269A2 (pt) * 2018-10-17 2022-12-06 Group Ro-Main Inc. monitoração de bens semoventes
CN109801260B (zh) * 2018-12-20 2021-01-26 北京海益同展信息科技有限公司 牲畜个数的识别方法、装置、控制装置及可读存储介质
CN110853076B (zh) * 2019-11-08 2023-03-31 重庆市亿飞智联科技有限公司 一种目标跟踪方法、装置、设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9084411B1 (en) * 2014-04-10 2015-07-21 Animal Biotech Llc Livestock identification system and method
CN108932496A (zh) * 2018-07-03 2018-12-04 北京佳格天地科技有限公司 区域内目标物的数量统计方法及装置
WO2020007363A1 (zh) * 2018-07-06 2020-01-09 京东数字科技控股有限公司 识别目标数量的方法、装置及计算机可读存储介质
CN111008561A (zh) * 2019-10-31 2020-04-14 重庆小雨点小额贷款有限公司 一种牲畜的数量确定方法、终端及计算机存储介质
CN111008560A (zh) * 2019-10-31 2020-04-14 重庆小雨点小额贷款有限公司 一种牲畜的重量确定方法、装置、终端及计算机存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114494863A (zh) * 2022-01-12 2022-05-13 北京小龙潜行科技有限公司 基于Blend Mask算法的动物幼崽计数方法以及装置
CN115861644A (zh) * 2022-11-21 2023-03-28 广东鉴面智能科技有限公司 一种生物多样性识别分析方法、系统及介质
CN116363494A (zh) * 2023-05-31 2023-06-30 睿克环境科技(中国)有限公司 一种鱼类数量监测、洄游跟踪的方法和系统
CN116363494B (zh) * 2023-05-31 2023-08-04 睿克环境科技(中国)有限公司 一种鱼类数量监测、洄游跟踪的方法和系统

Also Published As

Publication number Publication date
CN111680551A (zh) 2020-09-18

Similar Documents

Publication Publication Date Title
WO2021217934A1 (zh) 监控牲畜数量的方法、装置、计算机设备及存储介质
CN109101602B (zh) 图像检索模型训练方法、图像检索方法、设备及存储介质
WO2019120115A1 (zh) 人脸识别的方法、装置及计算机装置
WO2020093694A1 (zh) 生成视频分析模型的方法及视频分析系统
US9607224B2 (en) Entity based temporal segmentation of video streams
US8971591B2 (en) 3D image estimation for 2D image recognition
WO2019033574A1 (zh) 电子装置、动态视频人脸识别的方法、系统及存储介质
WO2021041176A1 (en) Shuffle, attend, and adapt: video domain adaptation by clip order prediction and clip attention alignment
WO2019080881A1 (zh) 行人流量漏斗生成方法及装置、存储介质、电子设备
WO2019033575A1 (zh) 电子装置、人脸追踪的方法、系统及存储介质
WO2021179719A1 (zh) 人脸活体检测方法、装置、介质及电子设备
CN111553234A (zh) 融合人脸特征与Re-ID特征排序的行人跟踪方法及装置
CN109902681B (zh) 用户群体关系确定方法、装置、设备及存储介质
Xu et al. Video analytics with zero-streaming cameras
WO2023019927A1 (zh) 一种人脸识别方法、装置、存储介质及电子设备
CN115103157A (zh) 基于边云协同的视频分析方法、装置、电子设备及介质
CN113379469A (zh) 一种异常流量检测方法、装置、设备及存储介质
WO2018068664A1 (zh) 网络信息识别方法和装置
CN113869115A (zh) 一种人脸图像处理的方法及系统
WO2021174688A1 (zh) 人脸检测方法及系统
CN114155578A (zh) 人像聚类方法、装置、电子设备和存储介质
CN112784691B (zh) 一种目标检测模型训练方法、目标检测方法和装置
CN115810228A (zh) 人脸识别门禁管理方法、装置、电子设备和存储介质
CN114399721A (zh) 人流分析方法、设备及非易失性计算机可读介质
CN112926542A (zh) 一种性别检测方法、装置、电子设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20933452

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 21/02/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20933452

Country of ref document: EP

Kind code of ref document: A1