KR102163455B1 - Remote delivery system and method using unmanned drone pilotless aircraft - Google Patents

Abstract

The present invention relates to a remote delivery system using an unmanned aerial vehicle and a delivery method thereof, and more particularly, to a remote delivery system using an unmanned aerial vehicle for solving problems of the unmanned aerial vehicle to provide a delivery service safely through control of take-off and landing points (algorithm for maintaining the altitude of the unmanned aerial vehicle), recipient recognition/authentication, a warning method in case of approaching a person (animal) other than a correct recipient, and rainfall prediction, and a delivery method thereof.

Description

Remote delivery system and method using unmanned drone pilotless aircraft

The present invention relates to a remote delivery system and delivery method using an unmanned aerial vehicle, and in more detail, control of take-off and landing points (altitude maintenance algorithm of unmanned aerial vehicle), recipient recognition/authentication, warning method when approaching a person (animal) other than the correct recipient , It relates to a remote delivery system and delivery method using an unmanned aerial vehicle configured to solve the problem of the unmanned aerial vehicle through rainfall prediction, etc. and to provide a safe delivery service.

In general, a drone (DRON) refers to an unmanned airplane flying through remote control from the ground without a person on board. At this time, drones are also called unmanned aerial vehicles (UAVs) because they fly through remote control.

Early drones were developed for military purposes. For example, drones have been developed for targeting, reconnaissance and surveillance, anti-submarine attacks, reconnaissance and information collection for practice shooting of air force aircraft or AA guns.

Drones are equipped with state-of-the-art equipment such as remote detection devices and satellite control devices, and are used in places that are difficult to access or dangerous areas to collect information, and are used as a function of attack aircraft to attack enemies instead of ground forces by installing attack weapons.

On the other hand, drones are being used in a variety of private sectors besides their military nature. For example, it is used for environments in which it is difficult for people to take pictures directly (for example, volcanic craters, fire sites, etc.), expensive aerial photography, and unmanned courier (delivery) services.

Unmanned delivery drones deliver goods to buyers through GPS (satellite navigation system). In other words, drones for unmanned parcel delivery deliver goods to the delivery location by flying based on the current location and delivery location measured using GPS information.

In addition, in the case of a CCTV drone that acts as a CCTV monitoring the target using the installed camera, it has the advantage of securing precise and diverse images through free movement, but there is a problem that the administrator must control the movement in the vicinity. .

In particular, research on the technology of using drones for delivery is being actively conducted, and drones have the advantage of being able to deliver while solving problems of time and place unattended, so various efforts to maximize these advantages have been attempted. Has become.

However, in general, when delivery is performed using a drone, the problem is how to verify that the delivery has been delivered to the correct recipient, compared to the delivery by a courier. Therefore, there is a demand for the development of authentication methods and devices when delivering goods.

In addition, if the altitude of the delivery drone is lower than the altitude to be maintained, the throttle value is raised to raise the drone, and if the altitude is higher than the altitude to be maintained, the throttle value is lowered to lower the drone.

However, while maintaining altitude, there are cases when the drone moves away from the altitude it needs to maintain due to several reasons. The causes include an imbalance in thrust due to a difference in motor speed, wind, and errors in the throttle calculation value for the rise or fall of the drone.

If the current altitude of the delivery drone is lower than the altitude that must be maintained, the throttle value is raised to raise the drone, and if it is higher than the altitude that must be maintained, the throttle value is lowered to lower the drone.

However, while maintaining altitude, there are cases when the drone moves away from the altitude it needs to maintain due to several reasons. The causes include an imbalance in thrust due to a difference in motor speed, wind, and errors in the throttle calculation value for the rise or fall of the drone.

Since wind is an environmental factor and motor speed problems are mainly caused by hardware defects, the main cause of the drone's ability to maintain altitude is an error in the throttle calculation value.

This error is caused by the fact that the current altitude is not accurately measured when trying to move the drone to the altitude it should maintain at the current altitude.

Therefore, in order to perform stable altitude maintenance, it is necessary to be able to measure the accurate altitude of the current drone.

In general, to measure the altitude of a drone, an acceleration sensor built into the inertial navigation system is used. In this case, since the acceleration is integrated to calculate the altitude, an integration error is accumulated. In addition, the fluctuation value of the acceleration measurement value due to the vibration of the drone is also used in the altitude calculation, causing an error in the measurement altitude. So, to prevent this problem with the accelerometer, a barometer is used together to measure altitude.

However, barometers measure changes in barometric pressure to calculate altitude, and react sensitively to external factors such as wind or temperature.

For example, if the air pressure around the drone is lowered due to the wind generated by the propeller when operating the drone, the altitude is measured as a negative value rather than zero. As a result, even if an acceleration sensor and a barometer are used together, it is difficult to expect an accurate altitude measurement, such as the measured value changes irregularly.

Due to the problems of the general drone altitude measurement method described above, in existing studies, a separate sensor is added and used for altitude measurement together with other sensors. As a result, there is a problem that the gas becomes heavier, the manufacturing cost increases, and the calculation processing is increased.

In addition, since there is a limit on the distance that can be measured depending on the sensor, the accuracy varies depending on the altitude.

Therefore, even if the delivery drone identifies the delivery destination with the correct address, in order to prevent a collision with the building due to failure to maintain the altitude, there is a need for a method to increase the accuracy of altitude measurement without adding other sensors.

In addition, accurate measures must be in place to prevent accidents delivered to persons other than the correct recipient.

Korean Patent Registration No. 1873021 Korean Patent Publication No. 2016-0142017

[1] Z. Zhang, "A Flexible new Technique for Camera Calibration," IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. 22, pp. 1330-1334, 2000. [2] D.C. Brown, "Close-range Camera Calibration," Photogrammetric Engineering, Vol. 37, pp.855-866, 1971.

The present invention has been made in order to solve the above problems, and provides an unmanned aerial vehicle using a recipient recognition/authentication method for estimating a parameter of a facial recognition model using a face recognition algorithm and a take-off and landing point control by an altitude maintenance algorithm of the unmanned aerial vehicle. It is an object to provide a remote delivery system used.

In addition, an object of the present invention is to provide a remote delivery system using an unmanned aerial vehicle capable of warning when approaching a person (animal) who is not a correct recipient, and protecting delivered goods by predicting rainfall.

In order to solve the above problems, the present invention includes the steps of: inputting a delivery address into a delivery server and arriving at the delivery address; After arriving at the delivery address, take-off and landing points by the altitude maintenance algorithm of the unmanned aerial vehicle based on the corrected throttle value calculated by the PID control method from the error between the altitude to be maintained and the current altitude of the drone, and the altitude to be maintained and the current measured altitude. Maintenance phase; A recipient recognition/authentication step of extracting a set of intersection points from the image photographed with the facial pattern in the take-off and landing point maintenance step, and estimating a parameter of the facial recognition model from the set of the extracted intersection points; And completing delivery after delivery of the delivery item after completion of the recipient recognition/authentication.

The final throttle value is Equation 1. The facial recognition model is calculated through Equation 2.

The unmanned aerial vehicle requests the recipient's voice recognition information from the mobile terminal, and obtains the recipient's voice recognition information through a microphone of the mobile terminal.

The unmanned aerial vehicle requests information on the wearable device from the mobile terminal, and the information on the wearable device is a unique identification number of the wearable device worn by the recipient.

For each correction pattern image captured by the facial recognition model, four outer reference points are designated, the initial intersection point is estimated based on the number and size information of the selected grid shape, and the shape of the extracted intersection points is originally If it is located inside the position, the radiation distortion coefficient is adjusted to a negative value, and if it is outside, the radiation distortion coefficient is adjusted to a positive value to correct the position of each intersection.

A short-range sensor that generates a sensor detection signal when an outsider approaches before completion of the recipient recognition/authentication and transmits the sensor detection signal to the control unit, and a speaker notifying of danger by varying decibels or hertz according to the control signal of the control unit.

The volume control signal is linearly controlled as the approach distance becomes closer through a short-range sensor including a Doppler sensor that detects the approach of an outsider.

The present invention made as described above can increase the accuracy of altitude measurement without adding other sensors in order to prevent a collision with a building due to failure to maintain altitude even if the delivery unmanned aerial vehicle identifies the delivery destination at the correct address. .

In addition, the present invention can prevent errors caused by distorted images by using a distortion conversion correction formula that is the largest cause of facial recognition errors in order to check whether delivery has been made to the correct recipient.

In addition, the present invention provides a warning when a person other than the correct recipient approaches, or has a function of maintaining the altitude until the correct recipient approaches, thereby increasing the accuracy of delivery.

In addition, the present invention can avoid as soon as possible so as not to get wet by using a deep neural network model for rainfall recognition.

1 is a view showing a conceptual diagram of a remote delivery system using an unmanned aerial vehicle according to an embodiment of the present invention.
FIG. 2 is a diagram showing a step of identifying a recipient by using a remote delivery system using an unmanned aerial vehicle according to an embodiment of the present invention.
3 is a diagram showing an operation process of a remote delivery system using an unmanned aerial vehicle according to an embodiment of the present invention.
4 is a diagram showing the configuration of a remote delivery system using an unmanned aerial vehicle according to another embodiment of the present invention.

In order to fully understand the present invention, preferred embodiments of the present invention will be described with reference to the accompanying drawings. The embodiments of the present invention may be modified in various forms, and the scope of the present invention should not be construed as being limited to the embodiments described in detail below. This embodiment is provided to more completely explain the present invention to those of ordinary skill in the art. Accordingly, the shape of the element in the drawings may be exaggerated to emphasize a clearer description. It should be noted that in each drawing, the same member may be indicated by the same reference numeral. In addition, detailed descriptions of known functions and configurations that are determined to unnecessarily obscure the subject matter of the present invention will be omitted.

The present invention solves the problems of the unmanned aerial vehicle through take-off and landing point control (altitude maintenance algorithm of the unmanned aerial vehicle), recipient recognition/authentication, a warning method when approaching a person (animal) other than the correct recipient, rainfall prediction, etc., and provides a safe delivery service. The unmanned aerial vehicle 200 is configured to include a sensing unit (near-range sensor) 101, a control unit 250, a speaker 102, and the like.

Control of take-off and landing points (algorithm for maintaining the altitude of the drone)

As shown in FIGS. 1 to 4, the key information necessary for maintaining the altitude of the unmanned aerial vehicle 200 is the altitude to be maintained, the current altitude of the unmanned aerial vehicle, and a corrected throttle value.

The corrected throttle value is calculated from the error between the altitude to be maintained and the current measured altitude, and generally uses a PID control (Proportional-Integral-Differential controller) method.

In addition, the algorithm for maintaining the altitude of an unmanned aerial vehicle according to the present invention largely goes through three processes. The first is the process of measuring the current altitude of the drone using an accelerometer or other sensor, and the second is the process of calculating the corrected throttle value to achieve the target maintenance altitude.

The third and final step is to add the calculated corrected throttle value to the current throttle value.

The sensing unit 101 may include one or more sensors for sensing at least one of information in the unmanned aerial vehicle, information on surrounding environments surrounding the unmanned aerial vehicle, and user information. For example, the sensing unit is a GPS sensor, a gyroscope sensor, an illumination sensor, a touch sensor, an acceleration sensor, a magnetic sensor, and gravity. Sensor (G-sensor), motion sensor (motion sensor), RGB sensor, infrared sensor (IR sensor: infrared sensor), fingerprint recognition sensor (finger scan sensor), ultrasonic sensor (ultrasonic sensor), optical sensor (ex. For example, cameras), microphones, battery gauges, environmental sensors (e.g., barometers, hygrometers, thermometers, radiation sensors, heat sensors, gas sensors, etc.), chemical sensors (e.g. For example, it may include at least one of an electronic nose, a healthcare sensor, a biometric sensor, etc.).

The corrected throttle value calculated to maintain the altitude of the actual drone is added to the current throttle value, or the maintenance altitude information is changed when the user wants to change the maintenance altitude, or the final used for calculating the motor value when performing altitude maintenance. The throttle value is calculated as in Equation 1.

은 고도 유지를 시작할 때의 스로틀 값인

에 보정 스로틀 값인

을 더해준 것이다.The final throttle value

Is the throttle value when you start maintaining altitude

Is the corrected throttle value at

Was added.

First, as shown in Fig. 3, it is checked whether the altitude maintenance mode is operated (S201).

Then, the altitude maintenance mode is executed (S202).

Continue setting the maintenance altitude according to the current altitude and the start throttle value according to the current throttle value (S203).

Then, the current altitude is measured and a throttle correction value is calculated using the PID control unit (S204).

Finally, it is checked whether the user changes the maintenance altitude (S205), the altitude is recalculated, and after confirming whether the altitude has already been changed (S207), the recalculation is performed (S206, S208).

Recipient recognition/authentication

As shown in FIG. 2, the unmanned aerial vehicle 200 requests authentication to determine that the recipient is the right person to receive the goods.

As an embodiment, the unmanned aerial vehicle 200 may request information on the recipient's face from the mobile terminal 100. Since the recipient's facial recognition information can be obtained through the camera of the mobile terminal 100, the unmanned aerial vehicle 200 may request the facial recognition information from the mobile terminal 100 (S101, S102).

After that, the authentication request and the confirmation request are sent to complete delivery to request a return (S103 to S111).

As an embodiment, the unmanned aerial vehicle 200 may request the recipient's voice recognition information from the mobile terminal 100.

Since the recipient's voice recognition information can be obtained through the microphone of the mobile terminal 100, the unmanned aerial vehicle 200 can request the voice recognition information from the mobile terminal 100. To this end, the recipient's voice is stored in the delivery server 300 in advance, for example, when the goods are loaded on the unmanned aerial vehicle 200, the recipient's voice information may be provided to the unmanned aerial vehicle 200 if necessary.

As an embodiment, the unmanned aerial vehicle 200 may request electronic tag (RFID) authentication information stored in the mobile terminal 100 from an NFC (Near Field Communication) recognition device.

The mobile terminal 100 includes electronic tag information, and the unmanned aerial vehicle 200 can obtain electronic tag recognition information by connecting the mobile terminal 100 to the NFC recognition device including the NFC recognition device. May request electronic tag recognition information from the mobile terminal 100.

As an embodiment, the unmanned aerial vehicle 200 may request password information from the mobile terminal 100. Since password information can be obtained through the user input unit of the mobile terminal 100, the unmanned aerial vehicle 200 can request the password information from the mobile terminal 100.

As an embodiment, the unmanned aerial vehicle 200 may request pattern recognition information from the mobile terminal 100.

Since pattern recognition information can be obtained through the user input unit of the mobile terminal 100, the unmanned aerial vehicle 200 can request the pattern recognition information from the mobile terminal 100.

The unmanned aerial vehicle 200 may request fingerprint recognition information from the mobile terminal 100. As shown in FIG. 5F, since fingerprint recognition information can be obtained through the fingerprint recognition unit of the mobile terminal 100, the unmanned aerial vehicle 200 can request the fingerprint recognition information from the mobile terminal 100.

As an embodiment, the unmanned aerial vehicle 200 may request information on a wearable device from the mobile terminal 100. The information on the wearable device may be a unique identification number of the wearable device worn by the recipient. Since the mobile terminal 100 can obtain information on the wearable device through the recognition unit, the unmanned aerial vehicle 200 can request information on the wearable device from the mobile terminal 100.

However, according to another embodiment, the unmanned aerial vehicle 200 may request at least one of facial recognition information, voice recognition information, electronic tag recognition information, password information, pattern recognition information, and fingerprint recognition information from the mobile terminal.

Crossing points are extracted through four steps for each correction pattern image captured using facial recognition information, etc., recognized by the sensing unit 101 (camera).

(1) Designate four outer reference points, (2) estimate the initial intersection point based on the number and size information of the grid shape included in the selected interior, and (3) the shape of the extracted intersection points are located inside the original position. If it is present, the radiation distortion coefficient is adjusted to a negative value, and if it is outside, the radiation distortion coefficient is adjusted to a positive value to correct the positions of each intersection point.

For the image photographed with the facial pattern, the above process is repeated to extract a set of intersection points, and parameters of the facial recognition model are estimated from the extracted set of intersection points using Zhang's method and Brown's iterative algorithm.

At this time, by grasping the positional relationship between the facial patterns, it is possible to estimate rotation and movement parameters of each pattern in coordinates based on the camera.

That is, it is defined as a problem of minimizing the sum of the distances between the actual coordinates of the face and the coordinates calculated by the estimated distortion model, and an optimal solution can be found.

는 영상의 좌표이고,Where N is the number of facial recognition pattern images used for learning, M is the number of intersections found in each image,

Is the coordinates of the image,

는 왜곡 모델로 추정한 좌표로써, A는 카메라의 초점 거리, 주점, 비틀림 계수, 접선 왜곡을 포함하는 벡터이고,

는 방사 왜곡 계수이며,

는 각각 i번째 영상의 외부 파라미터로써 회전과 이동을 나타내는 벡터이고,

는 j번째 좌표의 실제 3차원 좌표이다.

Is the coordinates estimated by the distortion model, where A is the vector including the focal length of the camera, the main point, the distortion factor, and the tangential distortion

Is the radiation distortion coefficient,

Is a vector representing rotation and movement as external parameters of the i-th image,

Is the actual three-dimensional coordinate of the j-th coordinate.

In addition, as an additional distortion conversion correction formula, a clearer image (rectangular image) can be obtained by using a rectangular coordinate transformation method or the like.

Here, the DOV (Direction of View vector) is obtained by the horizontal angle (δ), the vertical angle (β), and the radius (R) of the image view vector according to Equation 2, and a point on the circle 1/2 transforms the perspective distortion. It becomes the origin for

Here, the image rotation degree is calculated by the angle θ and the image magnification factor is applied to m.

When the image of an arbitrary circle according to Equation 2 is projected onto the equation, the polar coordinates (r, δ) on the X and Y two-dimensional plane are expressed by the relational expression as in Equation 4.

How to warn when approaching a person (animal) other than the correct recipient

As shown in FIG. 4, the present invention generates a sensor detection signal through the sensing unit 101 when an outsider approaches, so that the controller 250 controls the speaker 102 by varying decibels (dB) or hertz (Hz). You can communicate the danger through it.

When the microwave signal generated by the near-field sensor 101 is reflected by a moving object, the frequency of the signal changes in proportion to the speed of the object due to the Doppler effect.

The frequency of the signal reflected from a moving object with a constant period remains the same, but the phase changes with time.

f ₀ is the frequency of the input signal, v is the speed of the moving object, c is the speed of light, f is the amount of change in frequency, λ is the wavelength of the input signal, and x(t) is the displacement of the object.

According to the Doppler theory, for an object moving with a displacement that changes periodically, the frequency is constant, but the phase ㆈ(t) changes with time. This can be expressed in Equation 6.

Therefore, the present invention focuses attention by first notifying a siren or a warning sound so that the worker can concentrate on the guidance message according to the phase ㆈ(t).

Here, λ is the wavelength of the signal. The signal reflected from the object is phase-modulated. If the magnitude of the displacement change of this object is less than the magnitude of the signal wavelength, the phase change will be small.

Therefore, as the approach distance of a person gets closer through a short-range sensor including a Doppler sensor (phase change measurement), the first volume control signal <the second volume control signal can be used, and various applications are possible as described later.

In addition, sound can be transmitted through a speaker by synchronizing the volume of sound according to a sensor signal of a certain frequency.

Therefore, when approaching quickly, the sound may increase depending on the change in high frequency or phase φ(t), and the amount of change in displacement of the object.

Rainfall forecast

로 나누고, 각 영상 부분에서 무작위로 연속 프레임 표본을 선택해

로 나타낸다. K video portions of the same length as input video V through the sensing unit 101

Divided by and randomly select consecutive frame samples from each image part

Represented by

를 N 개의 조각

으로 나누면, TSSN은 아래 수학식 7과 같이 조각 집합

에 대한 함수의 형태로 표현된다.bracket

N pieces

Divided by, TSSN is a set of pieces as shown in Equation 7 below.

Is expressed in the form of a function for

에대해

는 입력 조각

과 매개변수

를 대상으로 하는 합성곱 신경망을 나타낸다.At this time,

About

The input piece

And parameters

Represents a convolutional neural network targeting.

는 조각

를 입력으로 받아 각 레이블에 대한 점수를 산출한다.

The piece

It takes as input and calculates the score for each label.

은 여러 조각들의 결과로부터 표본 수준의 합의를 산출하는 조각 합의(patch consensus) 함수를 나타낸다.

Represents a patch consensus function that yields a sample-level consensus from the results of several pieces.

는 표본

의 조각 합의 결과이다.

Is the sample

Is the result of a piece agreement.

는 여러 표본들의 결과로부터 영상 수준의 합의를 산출하는 부분 합의 (segmental consensus) 함수를 나타낸다.

Denotes a segmental consensus function that calculates the image-level consensus from the results of several samples.

는 부분 합의

가 주어졌을 때, 입력 영상에 대한 레이블 확률 분포를 예측한다.Prediction function

Partial agreement

Given is, predict the label probability distribution for the input image.

은 각 표본에서 생성된 조각의 개수를 나타낸다.

의 설정에 기초해, 본 발명에서는

를 3으로,

를 Softmax 함수로 설정한다. 부분 합의

와 조각 합의

은 각각 단순한 형태인

와

로 정의한다. 이 때, 레이블 점수

는 모든 조각의 동일한 레이블에 대한 점수의 함수 형태이다.

Represents the number of pieces created in each sample.

Based on the setting of, in the present invention

To 3,

Is set with the Softmax function. Partial agreement

Piece agreement with

Are each simple form

Wow

Is defined as At this time, the label score

Is a function form of the score for the same label of all pieces.

Aggregation functions g and r can be set in various forms such as average, maximum value, and weighted average.

In the present invention, an average is used to report the final recognition result.

Using a general categorical cross-entropy loss function, the final loss function for the partial sum G is expressed as Equation 8.

는 레이블

의 정답 레이블을 의미한다. 해당 손실 함수는

와

에 따라 미분 가능하거나, 적어도 기울기 계산이 가능하다. In this case, C is the number of labels,

The label

Means the correct answer label. The corresponding loss function is

Wow

Differentiable according to, or at least the slope calculation is possible.

Therefore, it is possible to optimize the parameter W of the deep neural network model from multiple samples and fragments by applying a back-propagation algorithm.

의 기울기는 수학식 9와 같이 유도된다.In the backpropagation process, the loss function for the parameter W of the model

The slope of is derived as in Equation 9.

가 부분 합의 G와 모든 조각 수준 예측으로부터 계산된 조각 합의

에 의해 최적화 가능함을 보여준다.In order to learn the parameters of a general deep neural network model, the present invention uses a gradient-based optimization algorithm. The above equation is a loss function

Is the partial sum G and the slice sum calculated from all slice level predictions

Shows that optimization is possible by

또는 손실 함수

등을 사용하여 배달 물품이 젖지 않도록 최대한 빨리 무인 항공기를 대피시킬 수 있다.Therefore, the present invention is a prediction function through a deep neural network model for rainfall recognition.

Or loss function

You can use a light to evacuate the drone as soon as possible to avoid getting the delivered goods wet.

In particular, the finally learned TSSN (Temporal and Spatial Segment Networks) model showed excellent rainfall recognition performance.

Meanwhile, in the remote delivery method using an unmanned aerial vehicle, address recognition capability is important, and the control unit 250 can review whether or not the display matches the display of the information board erected on the side of the road or crosswalk. This can be done through the -to-Text module, that is, the content representing the current phenomenon of image information is converted into text through the Image-to-Text module.)

For example, if an address (doc1) such as Mannyeon-ro or Mannyeon-dong is input, an error occurs in GPS information, and when a sign (doc2) such as Mannyeon-ro intersection or Mannyeon-ro direction is found, an algorithm that recognizes characters in the image is operated. It is necessary to quickly determine whether they match each other and deliver them to the correct address.

In this case, the image may be characterized using a convolutional neural network (CNN) or a recurrent neural network (RNN) method, or a character may be extracted from the image.

As in the test program below, the jacquard similarity J(doc1,doc2) between two data doc1 and doc2 is defined as the value obtained by dividing the size of the intersection of two sets by the size of the union of the two sets.

In conclusion, if the degree of similarity is 0.0 as follows, the controller 250 determines that the two data are quite similar.

Here, split() is a split module, set().union(set()) is a union module, and set().intersection(()) is a similarity module.

100: mobile terminal
200: drone

Claims (8)

A delivery address is input to a delivery server, and the unmanned aerial vehicle arrives at the delivery address;
After the drone arrives at the delivery address, takeoff and landing using an altitude maintenance algorithm based on a corrected throttle value calculated by the PID control method from the error between the altitude to be maintained and the current altitude of the drone, and the altitude to be maintained and the current measured altitude. Maintaining a point;
A recipient recognition/authentication step of estimating a parameter of a facial recognition model with respect to an image photographed of the recipient's facial pattern in the step of maintaining the take-off and landing point;
Including; the step of completing delivery after delivery of the delivery item after completion of the recipient recognition/authentication;
The face recognition model is a remote delivery method using an unmanned aerial vehicle, characterized in that calculated through Equation 2 below.
[Equation 2]

(Where N is the number of facial recognition pattern images used for learning, M is the number of intersections found in each image,

Is the coordinates of the image,

Is the coordinates estimated by the distortion model, where A is the vector including the focal length of the camera, the main point, the distortion factor, and the tangential distortion

Is the radiation distortion coefficient,

Is a vector representing rotation and movement as external parameters of the i-th image,

Is the actual three-dimensional coordinate of the j-th coordinate. ) The method of claim 1,
The corrected throttle value is a remote delivery method using an unmanned aerial vehicle, characterized in that Equation 1 below.
[Equation 1]

delete The method of claim 1,
The unmanned aerial vehicle 200 requests the recipient's voice recognition information from the mobile terminal 100, and obtains the recipient's voice recognition information through a microphone of the mobile terminal 100. A remote delivery method using an unmanned aerial vehicle. The method of claim 1,
The unmanned aerial vehicle 200 requests information on a wearable device from the mobile terminal 100, and the information on the wearable device is a unique identification number of the wearable device worn by the recipient. Remote delivery method. The method of claim 1,
For each correction pattern image captured by the facial recognition model, four outer reference points are designated, the initial intersection point is estimated based on the number and size information of the selected grid shape, and the shape of the extracted intersection points is originally A remote delivery method using an unmanned aerial vehicle, characterized in that the position of each intersection is corrected by adjusting the radiation distortion coefficient to a negative value if it is located inside the position, and by adjusting the radiation distortion coefficient to a positive value if it is outside. The method of claim 1,
The decibel (dB) or hertz (Hz) is different according to the short-range sensor 101 that generates a sensor detection signal and transmits it to the control unit 250 when an outsider approaches before completion of the recipient recognition/authentication. Remote delivery method using an unmanned aerial vehicle, characterized in that it further comprises a speaker 102 to notify the danger. The method of claim 7,
A remote delivery method using an unmanned aerial vehicle, characterized in that as the approach distance gets closer through a short-range sensor including a Doppler sensor that detects the approach of an outsider, the volume control signal is linearly controlled.

Images