KR20230105912A - Collision risk notification system and method through warm-blooded animal detection and approach predictive learning - Google Patents

Abstract

Collision risk notification system through detection of warm-blooded animals and learning of approach prediction according to an embodiment of the present invention detects firstly with a lidar sensor when approaching an object, calculates the approach distance, and secondarily uses an image sensor to detect It is possible to detect an object with an image, and if the object detected and identified by the photographed image corresponds to a warm-blooded animal, it is possible to determine whether or not there is a risk of collision with a driver around the vehicle. It is characterized in that it is possible to output a preset warning alarm for each stage according to the type, approach direction and approach distance, or transmit and output a warning alarm to the driver's terminal through a communication network.

Description

Collision risk notification system and method through warm-blooded animal detection and approach predictive learning}

The present invention relates to a system and method for notifying the danger of collision through detection of warm-blooded animals and predictive approach learning, and more particularly, to detecting warm-blooded animals approaching a vehicle through predictive learning and providing a warning alarm when determining danger. And it relates to a collision risk notification system and method through approach predictive learning.

With the recent development of technology, vehicle drivers apply and operate ADAS (Advanced Driver Assistance Systems), a system that pursues the purpose of autonomous driving rather than safety for comfort.

However, as the system fails to prepare or cope with unexpected situations such as the unexpected appearance of warm-blooded animals while driving that newly occur due to ADAS, the number of crash accidents accumulates over time and the limit of technology causes accidents to increase. there is.

In order to solve this problem, road-kill accident measures are being prepared, but one of the problems of ADAS is that accidents that occur due to the inability to detect the characteristics of warm-blooded animals hiding and acting freely or not recognizing stopped objects have recently increased and become an issue as a social problem. The situation is.

In addition, there are many cases where it is difficult to predict in advance about warm-blooded animals even by actual drivers, so road-kill accidents occur frequently every year, and preparation for this is urgent.

Therefore, in order to solve the above-mentioned problem, predict and detect an object that suddenly appears while driving to determine whether there is a risk of collision, and detect a warm-blooded animal that provides a warning alarm to avoid collision risk and collision through approach prediction learning. Research on risk notification systems and methods has become necessary.

Republic of Korea Patent Publication No. 10-2009-0126586 (published on December 09, 2009)

An object of the present invention is to use a LiDAR sensor to detect positional changes according to the x, y, and z coordinates of an animal, and to prevent humans from recognizing an animal that is hidden secondarily in an environment where sufficient location and visibility are not secured. Detecting warm-blooded animals that detects minute radiant light in an area with an image sensor (cmos sensor) and uses deep learning to detect by overlapping an optical amplifier (light avalanche) and provides step-by-step warning alarms set according to the degree of danger And to provide a collision risk notification system and method through approach predictive learning.

In the collision risk notification system through detection of warm-blooded animals and learning of approach predictive learning according to an embodiment of the present invention, when approaching an object, firstly detect it with a lidar sensor to calculate the approaching distance, and secondly take a picture using an image sensor If the object detected and identified by the captured image corresponds to a warm-blooded animal, it is possible to determine whether or not there is a risk of collision with a driver around the vehicle, and the object identified when determined to be a risk of collision It is characterized in that it is possible to output a preset warning alarm for each stage according to the type, approach direction and approach distance, or send and output a warning alarm to the driver's terminal through a communication network.

The collision risk notification system through detection of warm-blooded animals and approach predictive learning detects warm-blooded animals primarily with the lidar sensor, and secondarily utilizes a CMOS sensor in consideration of a situation where it is difficult to detect warm-blooded animals with lidar. A sensor unit equipped with a CMOS image sensor for acquiring images to identify and detect animals, an optical amplifier for detecting and detecting radiant light in areas that are difficult to detect, and a wide-angle lens when an object detection issue occurs. The lens unit, which is controlled so that the viewing angle is selectively used according to the algorithm set for each situation, and the light from the surface are radiated by the optical amplifier for the area where the lidar sensor cannot search, and the contrast ratio is a constant ratio of contrast ratio. A color contrast unit that can search for objects based on rain and the sensor unit are used to detect warm-blooded animals primarily with lidar sensors, and CMOS secondarily in consideration of the situation in which it is difficult to detect warm-blooded animals with lidar. It is characterized in that it includes a searcher for controlling to detect warm-blooded animals by utilizing a sensor and an optical amplifier.

The collision risk notification system through the detection of warm-blooded animals and approach predictive learning includes a detection unit that identifies and detects the warm-blooded animal detected by the search unit and provides the identification result to the control unit, receives the identification result from the detection unit, and identifies According to the result, the control unit controls to output a predetermined alarm for each step in consideration of the type, approach direction and approach distance of the object, and a driver of a vehicle located within a surrounding danger radius sends a predetermined danger notification signal according to the risk determination result from the control unit It is characterized in that it further comprises a communication unit including a communication protocol for transmitting to the terminal, and a collection analysis unit serving as a database that collects and stores learning data necessary for detection and prediction of warm-blooded animals from the control unit.

The control unit learns to predict the warm-blooded animal by using a deep learning learning model to facilitate identification of the warm-blooded animal, and uses the learning result of the warm-blooded animal approach prediction algorithm as the deep learning model to facilitate identification of the warm-blooded animal. It is characterized in that it is possible to calculate a singular point, or to calculate a singular point for a direction in which an approach of a warm-blooded animal is predicted, or an approach distance.

In the collision risk notification method through warm-blooded animal detection and approach predictive learning using the collision risk notification system through warm-blooded animal detection and approach predictive learning according to an embodiment of the present invention, collision through the warm-blooded animal detection and approach predictive learning The danger notification system detects when the warm-blooded animal approaches with a lidar sensor and calculates the approach distance of the warm-blooded animal; If it is difficult to detect a warm-blooded animal with the lidar sensor, detecting using an optical amplifier for detecting a second object detection image and radiant light; When using an optical amplifier to identify an object, using a wide viewing angle of the optical amplifier and identifying the object through contrast with the color of the target; If the identified object corresponds to a warm-blooded animal, determining whether there is a risk of collision when approaching a nearby driver's vehicle; When it is determined that there is a risk of collision, outputting a preset warning alarm for each step according to the type, approaching direction, and approaching distance of the object, or transmitting and outputting the warning alarm to the driver's terminal and the manager's terminal.

The collision risk notification method through the detection of warm-blooded animals and approach predictive learning utilizes the learning result of the warm-blooded animal approach prediction algorithm to facilitate the identification of warm-blooded animals, or to identify peculiarities about the direction and approach distance in which the warm-blooded animals are predicted to approach. A step of detecting is further included.

In the collision risk notification method through the detection of warm-blooded animals and approach predictive learning, the detected singularity and identification result data can be stored in a DB and used for deep learning learning, and the accuracy of the warm-blooded animal approach prediction algorithm improves as learning is repeated. It is characterized in that it can facilitate the detection of the type of warm-blooded animal or increase the prediction accuracy for the approach direction and approach distance.

The collision risk notification method through the detection of warm-blooded animals and approach predictive learning may be performed by a computer program recorded on a computer-readable storage medium.

The collision risk notification system through the detection of warm-blooded animals and approach predictive learning of the present invention prevents vehicle collision accident situations that occur due to the elderly, bicycles, wild animals, children, and Min-sik law abusive play, and the sudden appearance of people or animals It has the advantage of minimizing social losses by preventing secondary human and material accidents that occur due to failure to cope with

Through predictive learning using the primary lidar and the secondary mineslide, it is possible to detect the hidden appearance or sudden change of the object in case of an unexpected action, and according to the prediction result, a collision prediction alarm is sent to the driver, By making it possible to cope with it, it is possible to prevent large-scale accidents that can not be done by automobile ADAS and cause detection errors and collide as they are. has the advantage of minimizing

1 is a block diagram showing the configuration of a collision risk notification system through warm-blooded animal detection and approach predictive learning according to an embodiment of the present invention.
2 is a flowchart illustrating the concept of a collision risk notification method through warm-blooded animal detection and approach predictive learning according to an embodiment of the present invention.
3 is a flowchart of a collision risk notification method through warm-blooded animal detection and approach predictive learning according to an embodiment of the present invention.
4 is a diagram for explaining the process of Real or Fake Decision Model by way of example.

Hereinafter, specific embodiments of the present invention will be described in detail with reference to the drawings. However, the spirit of the present invention is not limited to the presented embodiments, and those skilled in the art who understand the spirit of the present invention may add, change, delete, etc. other elements within the scope of the same spirit, through other degenerative inventions or the present invention. Other embodiments included within the scope of the inventive idea can be easily proposed, but it will also be said to be included within the scope of the inventive concept. In addition, components having the same function within the scope of the same idea appearing in the drawings of each embodiment are described using the same reference numerals.

1 is a block diagram showing the configuration of a collision risk notification system through warm-blooded animal detection and approach predictive learning according to an embodiment of the present invention, and FIG. 2 is warm-blooded animal detection and approach predictive learning according to an embodiment of the present invention. 3 is a flow chart for explaining the concept of a collision risk notification method through , and FIG. 3 is a flow chart of a collision risk notification method through warm-blooded animal detection and approach predictive learning according to an embodiment of the present invention.

The collision risk notification system 1000 through the detection of warm-blooded animals and approach predictive learning of the present invention detects when a warm-blooded animal approaches with lidar to calculate the approach distance, and detects a second warm-blooded animal when it is difficult to detect warm-blooded animals with lidar. As a means, it can be detected by using an image captured by an image sensor and an optical amplifier (light avalanche), and if an object identified from an image captured by an image sensor corresponds to a warm-blooded animal, whether there is a risk of collision when approaching a nearby driver can be determined, and when determined to be a risk of collision, a preset warning alarm is output for each step according to the type of approach object, approach direction, approach distance, etc., or a warning alarm is sent to the driver terminal 2000 and the manager terminal 2000 through the communication network. It can be sent and printed.

Referring to FIG. 1, a collision risk notification system 1000 through warm-blooded animal detection and approach predictive learning to specifically perform the above functions includes a sensor unit 100, a lens unit 200, and a color contrast unit 300. ), a search unit 400, a detection unit 500, a control unit 600, a communication unit 700, and a collection and analysis unit 800.

The sensor unit 100 may include a lidar sensor, a CMOS image sensor used for image acquisition, and an optical amplifier.

That is, the sensor unit 100 is an optical amplifier that primarily detects warm-blooded animals with a lidar sensor and secondarily detects an object detection image and radiant light of a CMOS sensor in consideration of a situation in which it is difficult to detect warm-blooded animals with lidar. It is provided in two forms to identify and detect warm-blooded animals using the light avalanche of In addition, here, the warm-blooded animal may include not only animals that are subject to roadkill, but also humans.

Among the sensor unit 100, the lidar sensor may detect the displacement and position change of the warm-blooded animal to calculate the approach distance.

In addition, the possibility of detection may be increased by correcting an image appearing in a CMOS sensor of the sensor unit 100 .

Specifically, the correction method for the refraction, scattering, reflection, and absorption image, which is diffuse reflection and regular reflection appearing in the captured image, is determined by looking at the average value, standard deviation, and degree of dispersion (scatter) of the designated pixel reflected in the two images displayed in CMOS. The probability is increased by correcting and readjusting the value, and the relational expression is shown in Equation 1 below.

Here, if GL > 255, GL = 255, and if GL < 0, the range is limited to GL = 0.

The Kelvin value adjustment is based on the empirical linear correction technique, and the driver's usual route is input into the system to accumulate empirical raw data. Further determinations can be made and a calibration-generated image can be produced with the regression model.

Furthermore, distortion and noise of various visual signals can be filtered. For example, if the captured image is distorted due to the water and ice on the floor or the mirage distortion caused by the temperature difference in the air, correct the Kelvin value to the minimum. This can reduce the scattering effect and adjust it to the output value.

For other noise, filtering can be applied, and for example, one of Kalman filtering and Gaussian filtering techniques can be applied.

The lens unit 200 may provide various viewing angles for use of a wide viewing angle so as to easily identify a specific object when a specific object is detected while detecting a target warm-blooded animal with a normally set viewing angle of the CMOS sensor.

Specifically, the lens unit 200 operates a zoom lens with a wide angle function of 67 °, a standard lens with a 45 ° function, and a lens with a minimalism function of 21 °. The 67° and 21° lenses can be controlled so that the viewing angles are selectively used according to an algorithm set for each situation.

The color contrast unit 300 corresponds to an area where the lidar sensor cannot search (a case where the lidar sensor is located in a straight line distance and there are no obstructions (obstacles, etc.), so that distance measurement and corresponding object identification are possible. Except for), it is possible to search for an object in a hidden area based on a constant proportional ratio of the contrast ratio of color coordinates by copying light reflected from surfaces such as the ground surface, slope surface, and wall surface. At this time, the following mathematical Equation 2 can be utilized.

Here, M is the total radiation, which rapidly increases with the fourth power of the temperature (T), Mλ is the spectral radiation, σ is the Stefan-Boltzmann constant, and T is the sunspot temperature.

The color contrast unit 300 distinguishes the distance, color, and contrast ratio of objects (objects) by the color contrast unit 300 by moving the lens left and right by a certain distance in order to search for a dark or invisible area. By measuring the color contrast and detecting the object by reflecting the measurement result, the mechanism works to increase the reliability.

The search unit 400 uses the sensor unit 100 to primarily detect warm-blooded animals with a LiDAR sensor, and secondarily detects warm-blooded animals using lidar, taking into account the situation in which it is difficult to detect warm-blooded animals with lidar. It can be controlled so that the identification of warm-blooded animals can be easily detected by using.

Furthermore, the search unit 400 may utilize a neural network learning algorithm to easily recognize a warm-blooded animal, that is, an object.

The search unit 400 may recognize an object by applying a CNN algorithm, which is an object recognition algorithm, from the corrected image.

The CNN algorithm is a neural network algorithm that recognizes an object through a convolution operation. Recognizable.

The R-CNN algorithm is a neural network algorithm that first creates a candidate region and trains CNN based on it to find the location of an object in an image. The process of converting each candidate region into the same size, extracting a feature through CNN, and classifying objects in the candidate region using a SVM (Suppor Vector Machine) using the extracted feature. Since the position of the candidate region is not accurate, the position of the object region box can be accurately corrected through regression learning.

In addition, other neural network algorithms include Fast R-CNN, R-FCN, YOLO (You only Look Once), TensorFlow, and SSD (Single Shot MultiBox Detector). It can be applied instead of CNN, and object recognition speed can be improved by simultaneously recognizing multiple objects by applying the neural network algorithm described above.

Furthermore, algorithms such as AdaBoost, Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and Principal Component Analysis (PCA) are added to increase the object recognition rate. All of these algorithm techniques identify the region to be recognized based on the appearance, detect the region around the object using a model trained by a set of video images to be used for training, and restrict various surroundings. As conditions are overcome through training, object recognition accuracy and reliability can be improved as a result.

The detection unit 500 may identify and detect the warm-blooded animal detected by the birdwatching unit 400 and provide the identification result to the control unit 600 .

The control unit 600 receives the identification result from the detection unit 500, and can learn to predict the warm-blooded animal by using a deep learning learning model to easily identify the warm-blooded animal.

Specifically, as a deep-learning model, it is possible to calculate a singular point (feature point or threshold) that facilitates identification of a warm-blooded animal by utilizing the learning result of a warm-blooded animal approach prediction algorithm, or to predict the approach of a warm-blooded animal in the direction and approach distance. singularities can be calculated.

Furthermore, the warm-blooded animal approach prediction algorithm may use a temperature sensor or an infrared camera in addition to the sensor unit 100 to distinguish warm-blooded animals such as animals from general objects, and collect temperature information or camera acquisition information collected at this time. Therefore, it can be used for learning to predict the approach of warm-blooded animals.

In addition, the controller 600 may control to output a preset alarm for each stage in consideration of the type, direction, distance, etc. of the predicted object according to the identification result.

For example, step-by-step alarms are classified into minor, medium, and major levels, and the warm-blooded animal is a small animal (or a non-warm-blooded animal-object, etc.-), or the direction is less likely to collide with the driver or the distance is far. In this case, it may be determined as slight, and if the warm-blooded animal is a large animal, or the direction is likely to collide with the driver, or the distance is close, it may be set to be determined as serious.

In the case of the form of the alarm for each stage, it is notified in the form of a visual display on the driver terminal 2000 in the case of a slight alarm, and in the case of a serious alarm, it is desirable to raise awareness by including display indication as well as other forms of notification such as speaker output and vibration. do.

Furthermore, as a deep-learning model, a risk judgment prediction model can be used. The risk judgment prediction model continuously collects danger notification signals generated when a warm-blooded animal approaches a vehicle within a danger radius, location information of a warm-blooded animal, and , Determines and predicts the collision risk situation as an input variable during learning, resets the collision risk radius as a prediction result, optimal step-by-step collision risk radius boundary value (threshold value) or step detailed setting, step-by-step alarm type, alarm intensity or alarm It is possible to reset the number of outputs, the critical time required for the vehicle control (emergency stop) command, etc., through which it is possible to detect an accurate dangerous situation and provide an optimized warning for each step of the risk of collision.

In addition, the risk judgment prediction model is a learning and prediction model algorithm for providing warnings in stages of an optimized collision risk, for example, using a support vector machine (SVM), a deep neural network (DNN), or a convolutional neural network (CNN) or Various types of artificial intelligence algorithms can be used, such as a recurrent neural network (RNN) method.

In particular, SVM is known as an optimized method for finding an optimal decision boundary capable of well distinguishing a plurality of dimensions for data having a plurality of dimensions.

In addition, when the SVM technique as described above is applied to the acquired data, as the number of data required for learning increases and the transmitted data is accumulated more and more, the number of machine learning training increases, and as a result, the number of modeling obtained through training increases. Accuracy gradually increases.

As a result, if the matrix is analyzed and predicted using the above-described SVM technique, it is possible to specifically analyze the collision risk radius approach pattern of warm-blooded animals, and also by securing long-term data on these patterns, as a result, the collision risk by stage. It can be used to set the bounds (threshold) of the radius.

The communication unit 700 may include a communication protocol for transmitting a preset danger notification signal from the control unit 600 to the vehicle driver's terminal 2000 located within a surrounding danger radius according to a collision risk determination result.

Here, the terminal 2000 includes the ADAS system of the vehicle itself or the portable terminal 2000 possessed by the driver, and may be a monitoring server or manager terminal that additionally needs to check the risk of collision.

The collection and analysis unit 800 serves as a database that collects and stores learning data necessary for predicting the detection of warm-blooded animals from the control unit 600, and configures the learning algorithm of the control unit 600 when necessary to find outliers through deep learning learning. It can also perform a role such as production.

Furthermore, referring to FIG. 4, deep learning learning may be performed through a Real or Fake Decision Model of Data.

Specifically, actions are controlled by feedback based on prediction errors, new sensory information is obtained, and predictions are changed at the same time to automatically update (optimize) perceptions according to the following model patterns.

More specifically, the captured images are unclear, shaken, noisey information due to hidden warm-blooded animals, fog, direct sunlight on the camera, information that delays or omits warm-blooded animal detection (identification) due to insufficient lighting, and partially visible objects Percept Interpretant, which infers false alarms that are misrecognized and misjudged due to objects that look like people, insects attached to camera lenses, combination with headlights during heavy rain, shadows of strong light, etc. Information can be continuously produced based on a model using Equations 3 and 4 below.

Ms: Machine Signal

Kb: Kelvin Bridge

σ : normal distribution standard deviation

Opt cnd: Optimal decision result

: 어떤 복잡한 상황이 전개되는 분포

: a distribution in which a complex situation develops

The above-described collision risk notification system 1000 describes the process of the collision risk notification method through the detection of warm-blooded animals and approach predictive learning. , and calculates the approach distance of the detected warm-blooded animal (S100).

At this time, if it is difficult to detect the warm-blooded animal with the lidar sensor, it can be detected using an optical amplifier (light avalanche) as a secondary means for detecting the warm-blooded animal (S102).

A singularity can be detected by using the learning result of the warm-blooded animal approach prediction algorithm (S104).

In addition, when an optical amplifier is used for accurate object identification, a wide viewing angle of the optical amplifier is used, and the object can be identified through contrast with the target color (S106).

Now, if the identified object corresponds to a warm-blooded animal, it is possible to determine whether there is a risk of collision when approaching a nearby driver's vehicle (S108).

In addition, when it is determined that there is a risk of collision, a preset warning alarm for each step according to the type of approach object, approach direction, approach distance, etc. may be output, or may be transmitted and output to the driver terminal 2000 and the manager terminal 2000 (S110).

Then, after receiving a warning alarm, the driver terminal 2000 changes the driving direction so that the driver manually controls the vehicle to avoid an approaching warm-blooded animal, or if necessary for the driver's safety due to a serious danger, the driver directly brakes the vehicle directly. Or it can be automatically controlled to change direction.

Furthermore, the detected singularity and identification result data can be stored in a DB and used for deep learning learning. Prediction accuracy for distance, direction, etc. may be increased (S112).

During deep learning learning, learning by correcting singularities can help improve prediction accuracy.

Furthermore, the method for notifying collision risk through detection of warm-blooded animals and learning of approach prediction of the present invention is implemented as a computer program stored in a storage medium for execution through combination with a computer, or in the form of a module loaded into computer hardware in which the computer program operates. can be implemented as

In addition, collision risk notification through warm-blooded animal detection and approach predictive learning according to an embodiment of the present invention The method may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the medium may be those specially designed and configured for the present invention or those known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, SSD (Solid State Drive), and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler.

In this specification, a 'terminal' may be a wireless communication device that guarantees portability and mobility, and may be, for example, any type of handheld-based wireless communication device such as a smart phone, a tablet PC, or a laptop computer. In addition, the 'terminal' may also be a wired communication device such as a PC capable of accessing other terminals or servers through a communication network. In addition, a communication network refers to a connection structure capable of exchanging information between nodes such as terminals and servers, such as a local area network (LAN), a wide area network (WAN), and the Internet (WWW : World Wide Web), wired and wireless data communications network, telephone network, and wired and wireless television communications network.

Examples of wireless data communication networks include 3G, 4G, 5G, 3rd Generation Partnership Project (3GPP), Long Term Evolution (LTE), World Interoperability for Microwave Access (WIMAX), Wi-Fi, Bluetooth communication, infrared communication, ultrasonic communication, visible light communication (VLC: Visible Light Communication), LiFi, and the like, but are not limited thereto.

100; sensor part
200; lens part
300; Color Contrast
400; birdwatcher
500; detection unit
600; control unit
700; Ministry of Communications
800; Collection and analysis department
1000; Collision Hazard Notification System
2000; terminal

Claims (8)

When approaching an object, it is first detected by the lidar sensor to calculate the approach distance, and secondarily, the image sensor is used to detect the object with the captured image.
If the object detected and identified by the photographed image corresponds to a warm-blooded animal, it is possible to determine whether or not there is a risk of collision with a driver around the vehicle.
Detection of warm-blooded animals, characterized in that when determining the risk of collision, a warning alarm can be output in advance according to the type, approach direction and approach distance of the object identified, or a warning alarm can be transmitted and output to the driver terminal through a communication network and collision risk notification system through approach predictive learning. The method of claim 1,
The collision risk notification system through the detection of the warm-blooded animal and approach prediction learning
A CMOS image sensor that firstly detects warm-blooded animals with the lidar sensor and acquires images to identify and detect warm-blooded animals using a CMOS sensor secondarily in consideration of a situation where it is difficult to detect warm-blooded animals with lidar A sensor unit having an optical amplifier for detecting and detecting radiation in an area that is difficult to detect;
When an object detection issue occurs, a lens unit that uses a wide-angle lens to selectively control the viewing angle according to an algorithm set for each situation;
A color contrast unit capable of searching for an object based on a constant proportional ratio of contrast ratios in which color coordinates are compared by copying light from the surface with an optical amplifier for an area where the lidar sensor cannot search;
Using the sensor unit, firstly detects warm-blooded animals with lidar sensor, and secondarily uses CMOS sensor and optical amplifier to detect warm-blooded animals in consideration of the difficult situation where detecting warm-blooded animals is controlled by lidar birdwatcher
Collision risk notification system through warm-blooded animal detection and approach predictive learning, characterized in that it comprises a. The method of claim 2,
The collision risk notification system through the detection of the warm-blooded animal and approach prediction learning
A detection unit for identifying and detecting the warm-blooded animal detected by the birdwatching unit and providing the identification result to the control unit;
The control unit receiving the identification result from the detection unit and controlling to output a preset alarm for each stage in consideration of the type, approaching direction, and approaching distance of the object according to the identification result;
A communication unit including a communication protocol for transmitting a predetermined danger notification signal according to a result of the risk determination from the control unit to a terminal of a vehicle driver located within a surrounding danger radius;
Collection analysis unit serving as a database that collects and stores learning data necessary for predicting detection of warm-blooded animals from the control unit
Collision risk notification system through warm-blooded animal detection and approach predictive learning, characterized in that it further comprises. in claim 3
The control unit
In order to easily identify warm-blooded animals, a deep learning learning model is used to learn to predict warm-blooded animals,
As the deep learning model, by using the learning result of the warm-blooded animal approach prediction algorithm, a singularity can be calculated to facilitate the identification of the type of warm-blooded animal, or a singularity can be calculated for the direction and approach distance in which the approach of the warm-blooded animal is predicted Collision risk notification system through warm-blooded animal detection and approach prediction learning. In the collision risk notification method through warm-blooded animal detection and approach predictive learning using a collision risk notification system through warm-blooded animal detection and approach predictive learning,
The collision risk notification system through the detection of the warm-blooded animal and approach predictive learning detects when the warm-blooded animal approaches with a lidar sensor, and calculates the approach distance of the detected warm-blooded animal;
If it is difficult to detect a warm-blooded animal with the lidar sensor, detecting using an optical amplifier for detecting a second object detection image and radiant light;
When using an optical amplifier to identify an object, using a wide viewing angle of the optical amplifier and identifying the object through contrast with the color of the target;
If the identified object corresponds to a warm-blooded animal, determining whether there is a risk of collision when approaching a nearby driver's vehicle;
Outputting a step-by-step preset warning alarm according to the type, approach direction, and approach distance of the object when it is determined that there is a risk of collision, or transmitting and outputting it to the driver terminal and manager terminal
Collision risk notification method through warm-blooded animal detection and approach prediction learning. The method of claim 5,
A step of facilitating the identification of warm-blooded animals by utilizing the learning results of the warm-blooded animal approach prediction algorithm, or detecting singularities in the direction and approach distance in which the warm-blooded animal is predicted to approach
Collision risk notification method through warm-blooded animal detection and approach prediction learning further comprising. The method of claim 6,
The detected singularity and identification result data can be stored in a DB and used for deep learning learning, and as learning is repeated, the accuracy of the warm-blooded animal approach prediction algorithm improves, making it easy to detect the warm-blooded animal type, approaching direction, approaching distance Collision risk notification method through warm-blooded animal detection and approach predictive learning, characterized in that it can increase the prediction accuracy for. A computer program recorded on a computer-readable storage medium for performing the collision risk notification method through the detection of warm-blooded animals and learning of approach prediction according to any one of claims 5 to 7.

Images