KR102375380B1 - Apparatus for monitoring animals with low power and method thereof - Google Patents

Abstract

The present invention relates to a low-power animal monitoring apparatus and method, wherein when the appearance of an animal is detected, an image of the animal is acquired, and the type of the animal is predicted using the acquired image according to the state of charge of a battery At the same time, when storing or transmitting the result of predicting the type of the animal and the acquired image, the power charged in the battery is efficiently used to continuously operate even in an environment with limited power supply. To a low-power animal monitoring apparatus and method that operate to effectively monitor the animal.

Description

Low-power animal monitoring device and method therefor

The present invention relates to a low-power animal monitoring apparatus and method, and more particularly, when the appearance of an animal is detected, an image of the animal is acquired, and the animal is obtained using the acquired image according to the state of charge of a battery. At the same time, when predicting the type of animal is operated with low power, and when storing or transmitting the result of predicting the type of the animal and the acquired image, the electric power charged in the battery is efficiently used, so that the power supply is limited. To a low-power animal monitoring apparatus and method for effectively monitoring the animal by continuously operating in an environment.

Recently, with the rapid development of information and communication technology and camera technology, an intelligent unmanned camera equipped with a plurality of sensors and communication modules has been developed and commercialized.

Such intelligent unmanned surveillance cameras are installed in buildings, etc. to monitor crime or intrusion from the outside, in the security field, in the habitat of animals or on the movement path of animals to monitor the animals and study the ecology of animals in various fields, such as the field of natural environment. has been applied and used.

In particular, an intelligent unmanned surveillance camera mainly used in the field of natural environment is installed for a long time in the habitat or movement path and is often used for the purpose of studying the ecology of a specific animal or producing a documentary about a specific animal.

It is very important that the intelligent unmanned surveillance camera for monitoring the ecology of the animal is implemented so that it can continuously operate and monitor the animal because it is not known when the animal will appear.

However, since most of the intelligent unmanned surveillance cameras are installed in places where it is difficult to always supply electric power required for operation, there is a problem in that a person needs to periodically inspect them.

Accordingly, in recent years, in order to solve the problem of power supply, a power supply method by a solar panel is widely used, in which a solar panel is provided in an intelligent unmanned surveillance camera, and the power generated through the solar panel is charged in a battery. is being used However, in the case of a solar panel, since power is generated only during the day when there is sunlight, there is no power generation at night, and even in the case of daytime, when it rains or it is cloudy, there is a problem in that it cannot generate much power due to insufficient sunlight.

Therefore, in the case of the intelligent unmanned surveillance camera, which is implemented to capture an image in real time and transmit it to a central server, there is a limitation in continuously operating only with a solar panel because power consumption is very large.

Accordingly, the present invention installs a low-power animal monitoring device in a plurality of animal monitoring areas, and when the appearance of an animal is detected, a visible light camera, a thermal imaging camera, or a camera equipped with a night vision camera to capture an image of the animal Obtained by photographing, input the acquired image to a learning model equipped to predict the type of animal according to the acquired image, recognize the type of the appeared animal according to various accuracy, and store the captured image into a battery It is intended to propose a method for enabling the low-power animal monitoring device to operate with low power and to continuously monitor the animal's ecology by storing or transmitting it according to the amount of electricity charged in the .

In addition, according to the present invention, images must be acquired with a plurality of cameras according to various weather conditions as well as day or night, and the acquired images must be stored or transmitted. After the appearance of the animal is detected, the difference between the photographed image and the photographed image is calculated, and as a result, the type of animal is predicted only from the image of the animal, and the image is stored or transmitted.

In addition, the present invention does not store the monitoring data including the detection result, the recognition result, the image or a combination thereof every time data is generated on a hard disk (HDD) connected by a bus line, but on a PCB board. It is intended to propose a method for efficiently using the power by storing it as much as possible in a chip-level low-power memory to be mounted, and then variably storing or transmitting it according to the amount of power charged in the battery.

Next, the prior art existing in the technical field of the present invention will be briefly described, and then the technical matters that the present invention intends to achieve differently from the prior art will be described.

First, Korea Patent Registration No. 19484983 (2019.05.27.) relates to a wild animal monitoring system and method, and generates image data of wild animals through a monitoring terminal installed in the habitat of wild animals, and the generation Recording conditions including shooting date, time, location, direction, etc. in one image data and meta information based on habitat environment information including altitude, sunlight, moonlight, temperature, humidity, etc. detected in the habitat are stored in a central server to a wild animal monitoring system and method therefor in which an animal recognition model inputs the image data in the server to recognize an individual by animal species.

However, the prior art simply describes the concept of recognizing an individual by animal type by inputting image data into an animal recognition model, and the technical configuration of predicting the type of animal based on the thermal image consisting of the temperature distribution of the animal is not not listed at all. Also, in the prior art, an image of the animal is acquired only when an animal is detected, and the animal is not predicted based on the acquired image according to the amount of charging power of the battery. There is a significant difference between the prior art and the present invention, because a result or a predicted result and a method for storing the obtained image are also not described at all.

In addition, the non-patent document, IoT-based wildlife crime prevention device implementation (2018 Fall Conference Proceedings, Volume 25 No. 2 (November 2018)), is based on the Internet of Things (IoT), which is used to detect animal movements and take pictures. In this regard, when motion is detected through a detection sensor composed of a PIR sensor or a vibration sensor, a picture is taken through a camera and transmitted to a database, and a user can access the database through the web to check the photographed picture It relates to wildlife security devices.

The prior art is to simply take a picture when the motion of an animal is detected, and transmit it to a remote database for storage.

On the other hand, in the present invention, when the appearance of an animal is detected, an image of the animal is acquired through a visible light camera, a thermal imaging camera, or a camera equipped with a night vision goggles, and according to the amount of power charged in the battery, the The acquired image is input to a learning model provided in advance to predict the type of the detected animal, and monitoring data including the detection result, the predicted result, the acquired image or a combination thereof is stored at the chip level. By storing or transmitting the monitoring data stored in the low-power memory on the hard disk according to the amount of power charged in the battery, the power consumption is minimized to enable continuous animal monitoring, the prior art is clearly different from the present invention.

The present invention was created to solve the above problems, and when the appearance of an animal is detected, an image of the animal is acquired through a visible light image, a thermal image, or a night vision image, and according to the acquired image, the animal's To provide a low-power animal monitoring device and method for continuously monitoring animals by minimizing power consumption by applying each to a learning model provided to predict the type and predicting the type of the detected animal. The purpose.

In addition, the present invention provides a low-power animal monitoring device capable of minimizing the power consumption by monitoring the amount of charging power of a battery provided in the low-power animal monitoring device and performing the prediction at least once according to the monitoring result, and It is another object to provide the method.

In addition, the present invention performs only the detection of the appearance of the animal according to the monitoring result and provides the detection result to the animal monitoring server, predicts the type of the animal and transmits the recognized result to the animal monitoring server, or Another object of the present invention is to provide a low-power animal monitoring apparatus and method for providing the obtained image to an animal monitoring server to enable operation with low power. That is, the present invention provides only the sensed result by performing only the sensing according to the amount of power charged in the battery, or provides the predicted result by performing the process for predicting the type of the animal at least once or more, or provides a sufficient amount of power In this case, by transmitting the acquired image, power consumption can be minimized.

In addition, according to the present invention, the power consumption of the animal detection device can be further reduced by changing the storage path for the detection result, the recognition result, the acquired image, or a combination thereof according to the monitoring result. for other purposes. That is, in the present invention, instead of always accessing and storing the monitoring data on a hard disk connected to a bus line whenever the monitoring data is generated, the monitoring data is first stored in a chip-level low-power memory, and the monitoring data is stored in the battery. By storing the sensed data on the hard disk or transmitting it to the animal monitoring server according to the amount of charged electric power, the electric power can be efficiently used.

A low-power animal monitoring apparatus according to an embodiment of the present invention includes an animal detector that senses the appearance of an animal by sensing infrared rays emitted from an animal body through at least one sensor, and a camera image by detecting the appearance of the animal. It is characterized in that it comprises a camera image acquisition unit for obtaining a camera image, a background removal unit for removing the background from the acquired camera image and outputting an animal image, and an animal prediction unit for predicting the type of the animal from the animal image from which the background has been removed. .

In addition, the background removing unit removes the background that does not include the animal by calculating a difference image obtained by subtracting the camera image captured later from the camera image captured at the first time point when the appearance of the animal is detected, thereby removing the animal image. It is characterized in that it includes generating.

In addition, the camera image is at least one of a visible light image, a thermal image, and a night vision image, and the animal prediction unit. In the case of the visible light image, an ROI is set along the outermost contour for each animal, the type of the animal is predicted with the ROI, and in the case of the thermal image, a pattern for a heat map in which heat is detected for each animal Setting the ROI, predicting the type of the animal with the ROI, in the case of the night vision image, setting the ROI according to the pattern of the projected image for each animal, and predicting the type of the animal with the ROI It is characterized in that it further comprises.

In addition, the animal prediction unit includes predicting the type of the detected animal by applying each of the acquired images to a learning model provided to predict the type of the animal according to the acquired image, wherein the learning model includes: A first learning model generated by learning a visible light image for each animal, a second learning model generated by learning a thermal image for each animal, and a third learning generated by learning a night vision image for each animal It includes a model, and is characterized in that it is generated according to the type of the animal.

In addition, the animal prediction unit, when the charging power of the battery is in a good state, by applying the animal image with the background removed to the learning model and selecting the highest probability among the output results of the learning model, the type of the animal Prediction, and when the charging power of the battery is full, the obtained images with the background removed are applied to the learning model in the order in which they were taken, and the median or average value of the output results of the learning model is calculated. By selecting the highest value, the type of the animal is predicted.

In addition, the low-power animal monitoring apparatus records and stores the sensed detection result or the predicted prediction result as metadata, and a metadata processing unit for transmitting the stored metadata to an animal monitoring server and receiving the metadata When there is a request for the image of the detected animal from the animal monitoring server, it characterized in that it further comprises an animal image transmission unit for transmitting the image to the animal monitoring server.

In addition, the low-power animal monitoring method according to an embodiment of the present invention includes an animal detection step of detecting the appearance of an animal by sensing infrared rays emitted from an animal body through at least one sensor, and a camera according to detecting the appearance of the animal. A camera image acquisition step of acquiring an image, a background removal step of removing a background from the acquired camera image and outputting an animal image, and an animal prediction step of predicting the type of the animal from the animal image from which the background is removed do it with

In addition, in the background removal step, the animal image is removed by calculating a difference image obtained by subtracting the camera image captured later from the camera image captured at the first time point when the appearance of the animal is detected, thereby removing the background that does not include the animal. It is characterized in that it includes the creation of

In addition, the camera image is at least one of a visible light image, a thermal image, and a night vision image, and the animal prediction step includes: In the case of the visible light image, an ROI is set along the outermost contour for each animal, the type of the animal is predicted with the ROI, and in the case of the thermal image, a pattern for a heat map in which heat is detected for each animal Setting the ROI, predicting the type of the animal with the ROI, in the case of the night vision image, setting the ROI according to the pattern of the projected image for each animal, and predicting the type of the animal with the ROI It is characterized in that it further comprises.

In addition, the animal prediction step includes predicting the type of the detected animal by applying each of the acquired images to a learning model provided to predict the type of the animal according to the acquired image, wherein the learning model is , a first learning model generated by learning a visible light image for each animal, a second learning model generated by learning a thermal image for each animal, and a third generated by learning a night vision image for each animal It includes a learning model, and is characterized in that it is generated according to the type of the animal.

In addition, in the animal prediction step, when the amount of charging power of the battery is in a good state, the animal image with the background removed is applied to the learning model to select the highest probability among the output results of the learning model, so that the type of the animal is predicts, and when the charging power of the battery is in a full state, the obtained images with the background removed are applied to the learning model in the order in which they were taken to calculate the median or average value of the output results of the learning model to predict the type of the animal by selecting the highest value.

In addition, the low-power animal monitoring method includes a metadata processing step of recording and storing the sensed detection result or the predicted prediction result as metadata, and transmitting the stored metadata to an animal monitoring server, and receiving the metadata When there is a request for the image of the detected animal from an animal monitoring server, the method may further include an animal image transmission step of transmitting the image to the animal monitoring server.

As described above, in the low-power animal monitoring apparatus and method of the present invention, a visible light image, a thermal image or a night vision image of the animal is acquired only when the appearance of the animal is detected, and the obtained visible light image, thermal image Alternatively, there is an effect of minimizing power consumption by applying each of the learning models provided for predicting the type of animal according to the night vision image to predict the type of the detected animal.

In addition, the present invention provides the detection result to the animal monitoring server according to the monitoring result while monitoring the amount of charging power of the battery, or performs the prediction at least once, and transmits the predicted result to the animal monitoring server Alternatively, by transmitting the acquired visible light image, thermal image, or night vision image to the animal monitoring server, there is an effect of minimizing power consumption.

In addition, the present invention stores the detected result, predicted result, and the obtained visible light image, thermal image or night vision image or a combination thereof in a chip-level low-power memory, but the monitoring data is stored according to the monitoring result. By storing it on a hard disk or transmitting it to the animal monitoring server, there is an effect of efficiently using the power.

1 is a conceptual diagram illustrating a low-power animal monitoring apparatus and method according to an embodiment of the present invention.
2 is a diagram illustrating a process of predicting the type of an animal according to an embodiment of the present invention.
3 is a diagram illustrating in more detail a process of predicting the type of an animal according to an embodiment of the present invention.
4 is a diagram illustrating an ROI of an animal image with a background removed according to an embodiment of the present invention.
5 is a diagram illustrating the structure of a learning network according to an embodiment of the present invention.
6 is a diagram illustrating a process of predicting an animal type using a learning model according to an embodiment of the present invention.
7 is a diagram illustrating a process of predicting an animal type using a learning model according to another embodiment of the present invention.
8 is a block diagram illustrating the configuration of a low-power animal monitoring apparatus according to an embodiment of the present invention.
9 is a block diagram showing the configuration of an animal monitoring server according to an embodiment of the present invention.
10 is a flowchart illustrating a procedure for monitoring an animal with low power according to an embodiment of the present invention.

Hereinafter, a preferred embodiment of the low-power animal monitoring apparatus and method of the present invention will be described in detail with reference to the accompanying drawings. Like reference numerals in each figure indicate like elements. In addition, specific structural or functional descriptions of the embodiments of the present invention are only exemplified for the purpose of describing the embodiments according to the present invention, and unless otherwise defined, all used herein, including technical or scientific terms, are Terms have the same meanings as commonly understood by those of ordinary skill in the art to which the present invention pertains. Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present specification. It is preferable not to In the present invention, data can be interpreted as digital information.

1 is a conceptual diagram illustrating a low-power animal monitoring apparatus and method according to an embodiment of the present invention.

As shown in FIG. 1 , the low-power animal monitoring apparatus 100 according to an embodiment of the present invention is installed in a plurality of preset animal monitoring zones, respectively, to monitor animals in the animal monitoring zones, It is used for the purpose of monitoring the ecology or the invasion of animals.

In addition, the low-power animal monitoring apparatus 100 is provided with a solar panel (not shown), and operates using power generated through the solar panel during the daytime. In addition, the low-power animal monitoring device 100 charges the power produced by the solar panel in a rechargeable battery (not shown), and in a situation in which the solar panel cannot produce power (eg, at night or in rainy weather) It operates using the power charged in the battery.

That is, since the low-power animal monitoring device 100 is installed in a location where it is difficult to receive power at all times, it receives power using the solar panel and charges the battery, and is implemented to operate based on the charged power. It is designed to allow continuous animal monitoring with minimal consumption.

In addition, the low-power animal monitoring apparatus 100 generally performs only a function of detecting the appearance of an animal in order to minimize power consumption.

Detecting the appearance of the animal is performed by sensing the movement of the animal by sensing infrared rays emitted from the body of the animal through at least one sensor (not shown). In addition, the range of the wavelength for the infrared rays is set in advance.

On the other hand, the low-power animal monitoring apparatus 100 of the present invention includes at least one PIR (Passive Infrared Sensor) sensor (not shown) as the sensor, and senses infrared rays emitted from the body of the animal through the PIR sensor. , It is preferable to detect the appearance of the stone object, but is not limited thereto, and the sensor may be configured as an infrared sensor, an ultrasonic sensor, RADAR, or a combination thereof to detect the appearance of the animal. . Hereinafter, detecting the appearance of the animal through the PIR sensor will be described as an example.

In addition, the low-power animal monitoring apparatus 100 is configured as a camera including a visible light camera, a thermal imaging camera, a camera equipped with a night vision goggles, or a combination thereof, and when the appearance of the animal is detected, the The detected animal is photographed to obtain a visible light image, a thermal image, or a night vision image of the detected animal.

In addition, the low-power animal monitoring apparatus 100 predicts the type of the animal by applying each to a learning model provided to predict the type of the animal according to the acquired image.

The learning model is provided for each animal, such as a bear, a lion, a deer, etc., a first learning model learning a visible light image for each animal, a second learning model learning a thermal image of each animal, and each It is configured by including a third learning model that learned the night vision image of the animal.

The learning is performed through a convolutional neural network (CNN), which is a learning network optimized for image processing, the input of the learning model is an image from which the background is removed from the acquired image, and the output of the learning model is the detected image. The probability that an animal is a specific animal. The structure of the learning network will be described in detail with reference to FIG. 5 .

In addition, the learning model is generated by the animal monitoring server 200 and transmitted to each of the power animal monitoring devices 100 to predict the type of animals for the animals detected by each of the low power animal monitoring devices 100 . make it possible

The learning model may be updated in the animal monitoring server 200, and when it is updated, the learning parameters of the learning model are transmitted to the electric power animal monitoring apparatus 100, and each of the animal monitoring apparatuses 100 ), so that each learning model can be kept up to date.

In addition, the low-power animal monitoring device 100 is the camera image (ie, visible light image, thermal image or night vision image) acquired through a camera (ie, a visible light camera, a thermal imaging camera, or a camera equipped with a night vision goggles) Accordingly, the type of the detected animal is predicted by applying each of the acquired images to at least one learning model for predicting the type of the animal.

In this case, the low-power animal monitoring apparatus 100 deletes the background from the acquired image, and applies the animal image composed of only animals to each of the learning models.

In addition, in order to effectively remove the background, the low-power animal monitoring apparatus 100 includes a PIR sensor module composed of a plurality of PIR sensors. In this case, each of the PIR sensors is radially installed in the low-power animal monitoring apparatus 100 and has directivity in a specific direction to detect the appearance of an animal according to the direction.

In addition, the low-power animal monitoring apparatus 100 calculates a difference image obtained by subtracting a camera image obtained after being photographed from a camera image captured at the first time point when the appearance of the animal is detected, and from the image including the animal An image composed only of animals is obtained by deleting the background.

Meanwhile, the prediction is performed according to a charging state set in advance according to the amount of charging power of the battery.

The preset charging state includes: a full state in which the amount of charged electric power of the battery is sufficient; a good state in which the electric charge of the battery is good; It is composed of vulnerabilities that do not exist.

In addition, the full state means a case where the amount of charging power of the battery is 80% to 100%, and the good state means a case where the amount of charging power of the battery is 60% to 79%, and the normal state is, It means a state in which the amount of charging power of the battery is 40% to 59%, and the weak state means a case where the amount of charging power of the battery is less than 40%. In this case, the charging state is variously set according to an installation environment of the low-power animal monitoring device 100 or by a manager of the low-power animal monitoring device 100 .

In addition, the low-power animal monitoring apparatus 100 performs the prediction only when the charging state is greater than or equal to a good state in order to efficiently use the amount of charged electric power of the battery.

On the other hand, the low-power animal monitoring apparatus 100 is set to perform only the basic operation defined and set in advance when the battery charge state is a weak state.

For example, when the amount of charging power of the battery falls below a preset threshold, a report message is set to be transmitted to the animal monitoring server 200, or the low-power animal from the animal monitoring server 200 It is set to receive only messages transmitted from the animal monitoring server 200 such as an on/off control message for the monitoring device 100 .

In addition, the low-power animal monitoring apparatus 100 monitors the amount of charging power of the battery, and when the monitoring result indicates that the charging state for the amount of charging power of the battery is good, any one of the animal images from which the background is removed is displayed. Applied to the learning model, the type of the animal is predicted by selecting the highest probability among the output results of the learning model, and when the state of charge is the state of full, the background is removed from the camera image of the acquired animal By sequentially applying the animal images to the learning models in the order in which they were taken, calculating the median or average value of the probabilities output from each learning model, and selecting the highest value, the detected animal type is predicted.

That is, the low-power animal monitoring device 100 monitors the amount of power of the battery and, when the charging state is normal, transmits only the detected detection result to the animal monitoring server 200, or when the charging state is If it is in a good state or more, the prediction is performed at least once and the predicted prediction result is transmitted to the animal monitoring server 200 . In addition, when the acquired image is requested from the animal monitoring server 200 that has received the detection result or prediction result, when sufficient power is secured to transmit the image according to the monitoring result (that is, abnormal condition) ) to transmit the acquired image only.

Through this, the low-power animal monitoring apparatus 100 operates with a minimum amount of power to continuously perform animal monitoring in the animal monitoring area.

In addition, the low-power animal monitoring apparatus 100 variably sets a storage route for monitoring data including the detection result, the prediction result, the acquired camera image, or a combination thereof in order to further minimize power consumption.

In addition, the low-power animal monitoring apparatus 100 is configured to include a low-power memory (eg, RAM), which is a chip-level memory, and a large-capacity hard disk (HDD) connected by a bus line.

In general, in the case of a hard disk connected to a bus line, when storing specific data, more power is consumed than the chip-level low-power memory. Therefore, the low-power animal monitoring apparatus 100 stores the monitoring data in the low-power memory. First, the monitoring data stored in the low-power memory according to the amount of charging power of the battery is stored in the hard disk or transmitted to the animal monitoring server 200 to efficiently use the power charged in the battery. make it possible

In addition, when the monitoring data is transmitted to the animal monitoring server 200 , the low-power animal monitoring apparatus 100 deletes the corresponding monitoring data stored in the low-power memory or hard disk to secure a storage space.

On the other hand, before storing the monitoring data in the low-power memory, if the amount of power charged in the battery is sufficient (eg, in a good state or more), the monitoring data may be directly stored in the hard disk.

That is, the low-power animal monitoring apparatus 100 variably sets a storage route for storing the monitoring data according to the amount of power charged in the battery so that the power charged in the battery can be used more efficiently.

In addition, the animal monitoring server 200 outputs the detection result or prediction result and the image received from each of the low-power animal monitoring apparatuses 100 to a display provided therein, and the animal through the low-power animal monitoring apparatus 100 . The monitoring result is checked or transmitted to the user terminal 300 so that the animal monitoring result can be checked remotely.

Hereinafter, a process of predicting an animal will be described with reference to FIGS. 2 to 4 .

2 is a diagram illustrating a process of predicting the type of an animal according to an embodiment of the present invention.

As shown in FIG. 2 , in the process of predicting the type of animal in the low-power animal monitoring apparatus 100 according to an embodiment of the present invention, first, the low-power animal monitoring apparatus 100 includes at least one sensor (eg, : PIR sensor) detects the appearance of animals.

Also, when the appearance of the animal is detected through the at least one sensor, the low-power animal monitoring apparatus 100 acquires a camera image captured by the camera until the appearance of the animal is not detected.

In addition, the low-power animal monitoring apparatus 100 removes a background from the acquired camera image, outputs an animal image from which the background is removed, and sets an ROI from the output animal image (red part in FIG. 2), , predicts the type of the animal using the set ROI. In the present invention, the ROI includes what can be set as a rectangular area including the outermost outline of an animal, a pattern of heat distribution, or a pattern of a projected image.

The process of removing the background from the acquired camera image and predicting the animal type using the set ROI will be described with reference to FIG. 3 .

3 is a diagram illustrating in more detail a process of predicting the type of an animal according to an embodiment of the present invention.

As shown in FIG. 3 , the low-power animal monitoring apparatus 100 according to an embodiment of the present invention records the camera image (ie, the animal detection initial image) taken at the first point in time when the appearance of the animal is detected. It is first stored in a buffer provided in the low-power animal monitoring device 100 . In this case, the first animal detection image stored in the buffer becomes a reference image for outputting the animal image by removing the background.

In addition, the low-power animal monitoring apparatus 100 calculates a difference image obtained by subtracting a camera image captured after the first animal detection image from the first animal detection image stored in the buffer, so that the animal detection first image is taken after Outputs an animal image with the background removed from the camera image.

In this case, in order for the low-power animal monitoring apparatus 100 to accurately predict the type of the animal, it is preferable to remove the background from the camera image including the entire part of the animal.

Accordingly, the low-power animal monitoring apparatus 100 uses at least one camera image captured by simultaneously detecting the animal by the preset number of sensors or more after the initial image of animal detection is captured, or After capturing the first detection image, the background is deleted using at least one or more camera images captured after a preset time has elapsed.

In addition, the low-power animal monitoring apparatus 100 sets an ROI in the animal image from which the background is removed, predicts the type of the animal with the set ROI, and outputs the predicted prediction result.

Setting the ROI will be described with reference to FIG. 4 as follows.

4 is a diagram illustrating an ROI of an animal image with a background removed according to an embodiment of the present invention.

As shown in FIG. 4 , when setting the ROI in the animal image from which the background has been removed according to an embodiment of the present invention, the low-power animal monitoring apparatus 100 provides an animal image with the background removed when the animal image is a visible light image. By selecting the outermost contour of the ROI, an ROI is established from the animal image formed as the visible light image.

That is, when the animal image is a visible light image, an ROI for the shape of the animal is set along the outermost contour of each animal of the animal image.

In addition, when the animal image from which the background is removed is a thermal image, an ROI is set from the thermal image by selecting a region having a temperature value greater than or equal to a preset threshold value.

That is, when the animal image is a thermal image, a region having a temperature value greater than or equal to a preset threshold is selected, and a pattern for a heat map in which heat is sensed for each animal is set as an ROI.

In addition, when the animal image from which the background is removed is a night vision image, an ROI is set according to the pattern of the see-through image.

Referring back to FIG. 3 , the low-power animal monitoring apparatus 100 predicts the type of the animal with respect to the animal that has detected the appearance by the set ROI.

In addition, the low-power animal monitoring apparatus 100 selects at least one learning model according to the acquired camera image, and applies the ROI to the selected at least one or more learning models to predict the type of the animal.

In this case, as described above, the prediction is performed when the charging state of the battery is in a good state as a result of monitoring the charging power amount of the battery. In addition, the low-power animal monitoring device 100, when the charging state is in a good state, selects any one of the set ROIs, inputs it to the learning model, and selects the highest probability among the output results of the learning model. , predicts the kind of animal that has sensed the appearance.

In addition, when the charging state is the full state, the set ROI is sequentially applied to the at least one learning model in the order in which the animal image is taken, and the median or average value of the probability output from each learning model is obtained. By calculating and selecting the highest value, the kind of animal that has sensed the appearance is predicted.

Referring back to FIG. 2 , the low-power animal monitoring apparatus 100 records the output prediction result of predicting the type of animal as metadata, stores it in a memory (not shown), and stores the obtained camera image in the Save to memory.

On the other hand, the process of predicting the animal is performed only when the state of charge is above a good state as a result of monitoring the amount of charging power of the battery, and when the state of charge is normal, the detection of the appearance of the animal Only the results are recorded as metadata and stored in the memory.

In addition, when storing the metadata for the prediction result, the metadata for the detection result, and the acquired camera image in the memory, the low-power animal monitoring apparatus 100 variably changes the storage route according to the monitoring result. Setting and storing is the same as described above.

Also, as described with reference to FIGS. 2 to 4 , the camera image is an image captured when the appearance of the animal is detected and means an image including a background, and the animal image has a background in the camera image. It means an image that has been removed, and the ROI means a region of interest formed only by animals in the animal image.

5 is a diagram illustrating the structure of a learning network according to an embodiment of the present invention.

As shown in FIG. 5 , the learning model generated to predict the type of animal according to an embodiment of the present invention is performed through CNN, which is a learning network optimized for image processing.

In addition, the learning network composed of the CNN learns the animal images for learning composed of the visible light image, the thermal image or the night vision image, respectively, and is generated for each type of image and each type of animal.

In this case, the animal image for learning is an image from which the background is removed, and an image in which an ROI is set in the image from which the background is removed.

The learning network is an input layer to which an animal image for learning composed of only animals is input, a convolution layer that performs convolution on the input animal image for learning, and a maximum or average value of the convolution result. It consists of a pooling layer that performs sampling and a fully connected layer.

The convolution layer moves according to a pre-set stride (meaning a movement unit in which the kernel moves in the medical image for training) in which the kernel having a specific weight is set, and the specific part of the animal image for training and the kernel By convolving the weights of , a feature map for the animal image for learning is generated and output.

The pooling layer subsamples the plurality of sub-image data by pooling the feature map to a maximum value or an average value according to a kernel and a stride assigned to the pooling layer.

In this case, the convolution layer and the pooling layer may be configured as a pair and implemented as at least one or more.

The fully associative layer connects the plurality of sub-sampled sub-image data to output a probability that the animal of the animal image for learning is a specific animal. Here, the probability has a value of 0 to 1.

In addition, in the process of learning the animal image for learning, the weight of the learning network is adjusted in a direction to minimize an error with respect to the output of the learning network through a back propagation process.

6 is a diagram illustrating a process of predicting an animal type using a learning model according to an embodiment of the present invention, and FIG. 7 is an animal using a learning model according to another embodiment of the present invention. It is a diagram shown to explain the process of predicting the type of .

As shown in FIG. 6 , a learning model for predicting the type of an animal according to an embodiment of the present invention is generated by the animal monitoring server 200 and provided to each of the animal monitoring apparatuses 100 .

In addition, the learning model includes a first learning model generated by learning a plurality of visible light images prepared for each animal as an animal image for learning, a second learning model generated by learning a plurality of thermal images prepared for each animal as an animal image for learning, and each animal It is configured to include a third learning model generated by learning the prepared plurality of night vision images as animal images for learning.

However, the learning model may be generated by learning various types of images such as a binary image as well as a visible light image, a thermal image, and a night vision image.

That is, the present invention generates a learning model based on the various types of images and provides them in the animal monitoring apparatus 100, so that animals can be accurately predicted from images of animals obtained through various images.

In addition, the animal detection device 100 monitors the amount of electric power charged in the battery, and when the state of charge with respect to the amount of electric power charged in the battery is less than the normal state, the animal is detected without performing a process of predicting the type of the animal. Only the detected detection result is recorded as metadata and stored in the memory.

In addition, as a result of the monitoring, if the state of charge of the battery is good and the obtained camera image is a visible light image, as a result of the monitoring, the animal image with the background removed from the visible light image is displayed. If input to the first learning model, or if the acquired camera image is a thermal image, an animal image with the background removed from the thermal image is input to the second learning model, or if the acquired image is a night vision image, from the night vision image The animal image from which the background has been removed is input to the third learning model, the type of the animal that has detected the appearance is predicted, and the predicted prediction result is recorded as metadata and stored in the memory. In this case, the animal image input to each learning model is an ROI set in the animal image.

In this case, predicting the type of the animal is performed by selecting the highest value among the probabilities of being a specific animal output from each learning model.

On the other hand, as shown in FIG. 7 , the animal monitoring apparatus 100 monitors the amount of electric power charged in the battery. judge to be

At this time, the animal monitoring apparatus 100 outputs at least one animal image from which a background is removed from at least one or more camera images obtained by detecting the animal, and displays the ROI set in the output at least one or more animal images. The detected animal is more accurately predicted by applying each camera image in the shooting order, and the predicted result is recorded as metadata and stored in the memory.

For example, when the obtained image is a visible light image, the ROI is applied to the first learning model in the order in which the visible light image is captured, and when the obtained image is a thermal image or a night vision image, the The ROI is applied to the second learning model or the third learning model, respectively, in the order in which the thermal image or the night vision image is taken.

In this case, predicting the animal is performed by selecting the highest value by calculating a median or average value for the probability of a specific animal output from each learning model.

That is, as described with reference to FIGS. 6 and 7 , the animal detection apparatus 100 records only the sensed result as metadata according to the amount of power charged in the battery or detects the animal using only one image. Power charged in the battery by recording the predicted prediction result as the metadata, or recording the prediction result of predicting the animal using a plurality of images as the metadata and transmitting it to the animal monitoring server 200 is to be consumed efficiently so that the animal can be continuously monitored.

8 is a block diagram illustrating the configuration of a low-power animal monitoring apparatus according to an embodiment of the present invention.

As shown in FIG. 8 , the low-power animal monitoring apparatus 100 according to an embodiment of the present invention uses an animal detection unit 105 for detecting the appearance of an animal, and a camera when the appearance of the animal is detected. A camera image acquisition unit 110 for acquiring a camera image, a background removal unit 115 for removing a background from the acquired camera image, an animal prediction unit for predicting the type of the detected animal based on the acquired camera image ( 120), a metadata processing unit 125 that records the detected detection result or the predicted prediction result as metadata, and each of the low-power animal monitoring apparatus 100 so that the low-power animal monitoring apparatus 100 operates with low power. The control unit 130 for controlling the components, the animal image transmission unit 135 for transmitting the acquired image to the animal monitoring server 200, the low-power animal monitoring device 100 by receiving power from the solar panel 400 ), a power supply unit 140 for supplying power or charging the battery 145 , a battery 145 , a battery monitoring unit 150 for monitoring the amount of power charged in the battery 145 , and a memory 155 ) is comprised of

In addition, the animal detection unit 105 is configured as a sensor module including at least one or more sensors for detecting the appearance of an animal, and the sensor module is configured as a PIR sensor module including at least one or more PIR sensors.

In addition, the animal detection unit 105 senses infrared rays within a preset range through the PIR sensor module, and senses infrared rays emitted from the body of the animal to detect the appearance of the animal.

On the other hand, the animal detection unit 105 of the present invention is preferably composed of a PIR sensor module, but is composed of various sensors for detecting the appearance of an animal, including an infrared sensor, an ultrasonic sensor, a radar, a radar, or a combination thereof. It can be as described above.

In addition, the camera image acquisition unit 110 acquires a camera image of the animal from the first point when the appearance of the animal is detected through the animal detection unit 105 until the appearance of the animal is not detected. It performs a function of storing in the memory 155 .

In addition, the camera image acquisition unit 110 is configured as a camera including a visible light camera, a thermal imaging camera, a camera equipped with a night vision goggles, or a combination thereof, and a visible light image of the animal that detects the appearance through the camera. , to acquire a thermal image or night vision image.

In addition, the camera image acquisition unit 110, when the appearance of the animal is detected through the animal detection unit 105, a camera image taken at the first point in time when the appearance of the animal is detected (that is, the animal detection first image) ) is first obtained and stored in a buffer, so that the background can be simply deleted from the camera image obtained by shooting after the first image of animal detection.

In addition, the background removing unit 115 calculates a difference image obtained by subtracting the camera image captured later from the first animal detection image stored in the buffer through the camera image acquiring unit 110 to calculate a background that does not include the animal. Outputs an animal image from which .

At this time, the background removing unit 115 removes the background from the camera image in which the entire animal is included in the acquired camera image. In the case of overlapping detection of the animal by more than a set number of PIR sensors, the background of at least one camera image taken is removed, or after a preset time has elapsed from the first point in time when the appearance of the animal is detected The background of at least one or more camera images captured is removed.

In addition, the animal prediction unit 120 performs a function of predicting the type of the animal with the animal image output by removing the background. That is, the animal prediction unit 120 sets an ROI for the animal from the outputted animal image, and predicts the type of the animal with respect to the animal for which the appearance is sensed by the set ROI.

Meanwhile, since the setting of the ROI has been described with reference to FIG. 4 , further detailed description will be omitted.

The animal prediction unit 120 performs a function of predicting the detected animal type by applying the set ROI to a learning model provided to predict the animal type according to the acquired camera image.

At this time, in order to minimize the power consumption of the low-power animal monitoring apparatus 100 , the controller 130 is configured to operate the camera image acquisition unit 110 only when the appearance of the animal is detected through the animal detection unit 105 . ) to acquire a camera image of the animal that has detected the appearance, and to recognize the type of animal through the animal prediction unit 120 .

In addition, the animal prediction unit 120 applies the ROI to at least one or more learning models according to the acquired camera image and selects the highest value among the probabilities of being a specific animal output from each learning model. Predict the type of animal.

However, the control unit 130, according to the monitoring result of the battery monitoring unit 150, does not perform the prediction when the state of charge for the amount of charging power of the battery is less than or equal to the normal state, and the state of charge is good. In the case of abnormality, the prediction is performed. In addition, when the charging state is in the weak state, the controller 130 generally controls the low-power animal monitoring apparatus 100 to perform only a preset basic operation.

That is, the control unit 130 performs the prediction using one ROI or performs the prediction using a plurality of ROIs according to the state of charge for the amount of charging power of the battery, so that the battery is charged. This is to enable efficient use of the available power. Meanwhile, since predicting the type of the animal has been described with reference to FIGS. 6 and 7 , further detailed description will be omitted.

In addition, the metadata processing unit 125 records and generates the detected detection result or the predicted prediction result as metadata, and stores the generated metadata in the memory 155 . In addition, the metadata processing unit 125 performs a function of transmitting the stored metadata to the animal monitoring server 200 in a preset period or in real time.

In addition, the memory 155 is composed of a first memory 156 composed of a low-power memory such as RAM and a second memory 157 composed of a hard disk.

In addition, the control unit 130 stores the metadata generated by the metadata processing unit 125 and the camera image acquired by the camera image acquisition unit 110 in the first memory 156, but the battery monitoring According to the monitoring result of the unit 150 , if the charging state is good, the metadata and the camera image stored in the first memory 156 are stored in the second memory 157 .

Through this, instead of always accessing and storing data in the hard disk (second memory), which consumes a lot of power when storing data, the monitoring is performed in a low-power memory (first memory) with low power consumption according to the amount of charging power of the battery 145. Data is first stored as much as possible and stored in the hard disk when a sufficient amount of power is secured, so that the low-power animal detection device 100 can efficiently consume power.

In addition, when the metadata and camera image stored in the first memory 156 are stored in the second memory 157 , the metadata and camera image stored in the first memory 156 are automatically deleted, and the first memory Even when the metadata and images stored in (156) or the second memory 157 are transmitted to the animal monitoring server 200, they are automatically deleted to secure a storage space.

In addition, when there is a request for an animal image from the animal monitoring server 200 that has received the metadata, the animal image transmission unit 135 transmits the animal image stored in the memory 155 or in advance. It performs a function of transmitting the image of the animal according to a set period.

In addition, the control unit 130 controls the animal image transmission unit 135 to transmit a camera image of the animal according to the request or cycle to the animal monitoring server 200 , but monitor the battery. According to the monitoring result of the unit 150, when the amount of power charged in the battery 145 is sufficiently secured (that is, when the state of charge is good), the camera image is transmitted, thereby transmitting a large-capacity image. The operation of the low-power animal monitoring apparatus 100 is prevented from being stopped due to power consumption, thereby enabling continuous animal monitoring.

In addition, the power supply unit 140 receives power generated from the solar panel 400 , charges the battery 145 , and supplies power required for the operation of the low-power animal monitoring device 100 .

On the other hand, the solar panel 400 is preferably provided integrally with the low-power animal monitoring device 100, but depending on the installation environment of the low-power animal monitoring device 100, the low-power animal monitoring device 100 and It may be installed to be spaced apart and configured to supply power to the low-power animal monitoring device 100 .

In addition, the battery monitoring unit 150 monitors the amount of power charged in the battery 145, and provides a charging state, which is a monitoring result, to the control unit 130, and according to the monitoring result, the low-power animal monitoring device 100 ) so that it can be controlled to operate with low power.

9 is a block diagram showing the configuration of an animal monitoring server according to an embodiment of the present invention.

As shown in FIG. 9 , the animal monitoring server 200 according to an embodiment of the present invention includes a learning model generator 210 that generates a learning model for each animal for predicting the type of animal, and the generated learning. A learning model providing unit 220 for providing a model to the animal monitoring apparatus 100 , a metadata receiving unit 230 for receiving metadata from the animal monitoring apparatus 100 , and photographing an animal from the animal monitoring apparatus 100 . It is configured to include an image receiving unit 240 for receiving one camera image and a monitoring data providing unit 250 for providing monitoring data including the metadata, camera image, or a combination thereof to the user terminal 300 .

The learning model generating unit 210 is a first learning model that learns a visible light image for each animal to predict the type of animal based on the visible light image, learns a thermal image for each animal and selects the type of animal based on the thermal image It performs a function of generating a second learning model for prediction and a third learning model for predicting the type of animal based on the night vision image by learning the night vision image for each animal.

At this time, each of the generated learning models is trained to take an ROI set in an image from which the background is removed by photographing the animal, and to output a probability that the animal is a specific animal for the ROI.

In addition, the learning model providing unit 220 transmits the created learning model to the animal monitoring apparatus 100 so that the animal monitoring apparatus 100 can provide a learning model for each animal.

Also, the metadata receiving unit 230 receives metadata from at least one or more animal monitoring apparatuses 100 and stores the received metadata in the database 500 . The metadata includes a detection result of detecting the appearance of an animal in the animal monitoring apparatus 100 or a prediction result of predicting the type of the detected animal.

At this time, the metadata consisting of the detection result of detecting the appearance of the animal or the metadata consisting of the prediction result of predicting the type of the animal is the charging of the battery 145 provided in each animal detection device 100 . Generation according to the state of charge with respect to the amount of power is as described above.

In addition, the metadata is transmitted and received from the at least one or more animal monitoring apparatuses 100 according to a preset period. However, whenever the metadata is generated, it may be transmitted and received in real time.

Also, the image receiving unit 240 receives a camera image of an animal from the at least one or more animal monitoring apparatuses 100 and stores it in the database 500 .

As described above, the camera image includes a visible light image, a thermal image, a night vision image, or a combination thereof.

Also, the image receiving unit 240 performs a function of requesting and receiving the camera image from at least one or more animal monitoring apparatuses 100 according to a preset period.

In addition, when the metadata is received, the image receiving unit 240 may request the camera image from the animal monitoring apparatus 100 that has transmitted the metadata to receive the image from the corresponding animal monitoring apparatus 100 . there is.

In addition, the received metadata and the camera image are stored for each animal monitoring apparatus 100, the reception time at which the metadata and the image are received is recorded, and accumulated according to the reception time.

In addition, the monitoring data providing unit 250 provides monitoring data including the received metadata, camera image, or a combination thereof to the user terminal 300 so that the monitoring data can be checked.

In addition, the monitoring data providing unit 250 outputs the received monitoring data to a display connected to the animal monitoring server 200 so that the user can check the monitoring data.

In addition, the animal monitoring server 200 provides a function for the user to check the monitoring data stored in the database 500 by accessing the web through the user terminal 300 .

10 is a flowchart illustrating a procedure for monitoring an animal with low power according to an embodiment of the present invention.

As shown in FIG. 10 , in the procedure for monitoring an animal with low power through the low-power animal monitoring apparatus 100 according to an embodiment of the present invention, first, the low-power animal monitoring apparatus 100 detects the appearance of the animal. to perform an animal detection step (S110).

The animal detection step detects the appearance of the animal by sensing infrared rays emitted from the body of the animal through a PIR sensor module including at least one PIR sensor.

Next, the low-power animal monitoring apparatus 100 acquires a camera image by photographing the animal that detects the appearance using a camera, and stores the acquired camera image in the memory 155. A camera image acquisition step carry out (S120).

The camera is configured to include a visible light camera, a thermal imaging camera, a camera equipped with a night vision goggles, or a combination thereof, and the camera image acquisition step includes a visible light image of an animal that detects the appearance through the camera, heat Acquire images or night vision images.

Next, the low-power animal monitoring apparatus 100 monitors the amount of charging power of the battery 145 and records only the detected detection result as metadata when the charging state of the battery 145 is less than the good state. It performs a function of storing in the memory 155 .

On the other hand, when the state of charge is the weak state, the low-power animal monitoring apparatus 100 performs only a preset basic operation until the state of charge of the battery 145 becomes above the normal state as described above. It's like a bar.

In addition, when the amount of charging power of the battery 145 is greater than or equal to a good state, a background removal step of removing and outputting a background from the acquired camera image is performed (S140).

The removal of the background includes storing the camera image (first image of animal detection) obtained by shooting at the first time point when the appearance of the animal is detected through the animal detection step in the buffer of the low-power animal monitoring apparatus 100, and , is performed by calculating a difference image for each of at least one or more camera images captured after the initial image of animal detection from the first image of animal detection stored in the buffer, and removing a background that does not include animals.

Next, the low-power animal monitoring apparatus 100 sets an ROI on the outputted animal image, and performs an animal prediction step of predicting the type of animal with the set ROI (S150).

In this case, in the animal prediction step, when the charging state of the battery 145 is monitored and the charging state is good, any one of the ROIs set in at least one learning model according to the acquired camera image , and predicts the type of the animal by selecting the highest value among the probabilities of being a specific animal output from the learning model.

Also, in the animal prediction step, when the charging state is a full state, the set ROI is applied to at least one learning model according to the acquired camera image in the order in which the camera images are taken, and output from the learning model The type of the animal is predicted by calculating the median value or the average value for the probability of being a specific animal, respectively, for each learning model, and selecting the highest value as a result of the calculation.

Here, the learning model is a first learning model generated by learning a visible light image as an animal image for learning, a second learning model generated by using a thermal image as an animal image for learning, and a second learning model generated by learning an endoscopic image as an animal image for learning It consists of 3 learning models, and each learning model is generated for each animal.

Next, the low-power animal monitoring apparatus 100 records the predicted prediction result as metadata and stores it in a memory (S160).

That is, the animal monitoring apparatus 100 acquires the camera image only when the appearance of the animal is detected, and as a result of monitoring the amount of charging power of the battery 145 , the animal monitoring device 100 only obtains the animal when the charging state is greater than or equal to the good state. It is to predict the type of , so that operation with low power is possible.

In addition, the low-power animal monitoring apparatus 100 records and stores the predicted result as metadata according to the state of charge, or records and stores only the sensed detection result as the metadata, so that the battery 145 is stored. This is to enable the animal monitoring apparatus 100 to continuously operate by efficiently using the charged power.

In addition, the low-power animal monitoring apparatus 100 stores the metadata and the camera image in the first memory 156 which is a low-power memory. In this case, the metadata stored in the first memory 156 and the camera image are stored in the second memory 157 to efficiently use the power as described above.

Next, the low-power animal monitoring apparatus 100 transmits the stored metadata to the animal monitoring server 200 ( 170 ).

The stored metadata may be metadata recorded with a detection result or metadata recorded with the prediction result, and may be transmitted according to a preset period or transmitted in real time.

Next, the low-power animal monitoring apparatus 100 performs an animal image transmission step of transmitting the stored camera image to the animal monitoring server 200 (S170).

The stored camera image is provided according to the preset period, and when there is a request for the image from the animal monitoring server 200 that has received the metadata, it is transmitted according to the request.

On the other hand, the animal monitoring device 100 transmits the camera image periodically or according to the request. By transmitting the high-capacity image, it is prevented that the electric power charged in the battery 145 is all consumed, so that the animal monitoring apparatus 100 can continuously operate.

As described above, the present invention relates to a low-power animal monitoring apparatus and method, which acquires a camera image of the animal only when the appearance of the animal is detected, and controls the detected animal according to the amount of charging power of the battery. Predict the type to enable operation with low power, store the sensed or predicted result and the acquired image as much as possible in a low-power memory, and store it in a hard disk according to the amount of charging power of the battery or animal By transmitting the data to the monitoring server, the animal monitoring apparatus continuously operates even in an environment of limited power supply by efficiently using the power charged in the battery to effectively monitor the animal.

In the above, the preferred embodiment according to the present invention has been mainly described above, but the technical spirit of the present invention is not limited thereto, and each component of the present invention is changed or modified within the technical scope of the present invention to achieve the same purpose and effect. it could be

In addition, although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above, and the technical field to which the present invention belongs without departing from the gist of the present invention as claimed in the claims In addition, various modifications may be made by those of ordinary skill in the art, and these modifications should not be individually understood from the technical spirit or perspective of the present invention.

100: low-power animal monitoring device 105: animal detection unit
110: image acquisition unit 115: background removal unit
120: animal prediction unit 125: metadata processing unit
130: control unit 135: animal image transmission unit
140: power unit 145: battery
150: battery monitoring unit 155: memory
156: first memory 157: second memory
200: animal monitoring server 210: learning model generation unit
220: learning model providing unit 230: metadata receiving unit
240: image receiving unit 250: monitoring data providing unit
300: user terminal 400: solar panel
500: database

Claims (12)

an animal detection unit configured to sense the appearance of an animal by sensing infrared rays emitted from the body of the animal through at least one sensor;
a camera image acquisition unit configured to acquire a camera image by detecting the appearance of the animal;
a background removing unit for removing a background from the acquired camera image and outputting an animal image; and
Including; an animal prediction unit for predicting the type of the animal with the animal image from which the background has been removed;
The background removing unit,
Comprising generating an animal image by calculating a difference image obtained by subtracting a camera image captured later from a camera image captured at the first time point when the appearance of the animal is detected and removing a background that does not include the animal A low-power animal monitoring device. delete The method according to claim 1,
The camera image is
At least one of a visible light image, a thermal image, or a night vision image,
The animal prediction unit,
In the case of the visible light image, an ROI is set along the outermost contour for each animal, and the type of the animal is predicted with the ROI,
In the case of the thermal image, a pattern for a heat map in which heat is sensed for each animal is set as an ROI, and the type of the animal is predicted with the ROI,
In the case of the night vision image, setting an ROI according to the pattern of the fluoroscopic image for each animal, and predicting the type of the animal with the ROI. The method according to claim 1,
The animal prediction unit,
It includes predicting the type of the detected animal by applying each of the acquired camera images to a learning model provided to predict the type of animal according to the acquired camera image,
The learning model is
A first learning model generated by learning a visible light image for each animal, a second learning model generated by learning a thermal image for each animal, and a third learning generated by learning a night vision image for each animal includes a model,
Low-power animal monitoring device, characterized in that generated according to the type of the animal. 5. The method according to claim 4,
The animal prediction unit,
If the amount of charging power of the battery is in a good state, the animal image from which the background is removed is applied to the learning model to predict the type of the animal by selecting the highest probability among the output results of the learning model,
When the amount of charging power of the battery is full, the obtained images with the background removed are applied to the learning models in the order in which they were taken, and the median or average value of the output results of each of the learning models is calculated and the highest value Low-power animal monitoring device, characterized in that predicting the type of the animal by selecting the. The method according to claim 1,
The low-power animal monitoring device,
a metadata processing unit that records and stores the sensed detection result or the predicted prediction result as metadata, and transmits the stored metadata to an animal monitoring server; and
When there is a request for the image of the detected animal from the animal monitoring server that has received the metadata, an animal image transmission unit for transmitting the acquired camera image to the animal monitoring server; characterized in that it further comprises Low-power animal monitoring devices. an animal detection step of detecting the appearance of an animal by sensing infrared rays emitted from the body of the animal through at least one sensor;
a camera image acquisition step of acquiring a camera image according to detecting the appearance of the animal;
a background removal step of removing a background from the acquired camera image and outputting an animal image; and
Including; an animal prediction step of predicting the type of the animal with the animal image from which the background has been removed;
The background removal step is
Comprising generating an animal image by calculating a difference image obtained by subtracting a camera image captured later from a camera image captured at the first time point when the appearance of the animal is detected and removing a background that does not include the animal A low-power animal monitoring method. delete 8. The method of claim 7,
The camera image is
At least one of a visible light image, a thermal image, or a night vision image,
The animal prediction step is
In the case of the visible light image, an ROI is set along the outermost contour for each animal, and the type of the animal is predicted with the ROI,
In the case of the thermal image, a pattern for a heat map in which heat is sensed for each animal is set as an ROI, and the type of the animal is predicted with the ROI,
In the case of the night vision image, the low-power animal monitoring method further comprising: setting an ROI according to the pattern of the fluoroscopic image for each animal, and predicting the type of the animal with the ROI. 8. The method of claim 7,
The animal prediction step is
It includes predicting the type of the detected animal by applying each of the acquired camera images to a learning model provided to predict the type of animal according to the acquired camera image,
The learning model is
A first learning model generated by learning a visible light image for each animal, a second learning model generated by learning a thermal image for each animal, and a third learning generated by learning a night vision image for each animal includes a model,
Low-power animal monitoring method, characterized in that generated according to the type of animal. 11. The method of claim 10,
The animal prediction step is
If the amount of charging power of the battery is in a good state, the animal image from which the background is removed is applied to the learning model to predict the type of the animal by selecting the highest probability among the output results of the learning model,
When the amount of charging power of the battery is full, the obtained images with the background removed are applied to the learning models in the order in which they were taken, and the median or average value of the output results of each of the learning models is calculated and the highest value Low-power animal monitoring method, characterized in that predicting the type of the animal by selecting. 8. The method of claim 7,
The low-power animal monitoring method comprises:
a metadata processing step of recording and storing the sensed detection result or the predicted prediction result as metadata, and transmitting the stored metadata to an animal monitoring server; and
When there is a request for the image of the detected animal from the animal monitoring server that has received the metadata, an animal image transmission step of transmitting the acquired camera image to the animal monitoring server; characterized in that it further comprises Low-power animal monitoring methods.

Images