KR102046518B1 - Surrounded environment estimation system of a pluraoty of automobiles, server for estimating surrounded environment, and surrounded environment estimation method of a pluraoty of automobiles - Google Patents

Abstract

According to an embodiment of the present invention, a system for estimating a surrounding environment having a server communicating with a moving object through a network includes a plurality of moving objects and a server, and each of the plurality of moving objects outputs one or more sensing data. A sensing data output unit and a first communication unit transmitting the output sensing data to the server, wherein the server is configured to input a second communication unit to receive the sensing data from the plurality of moving objects and the received sensing data; Machine learning is performed based on a learning model using data, and the surrounding environment of the plurality of moving objects is estimated based on an ambient environment estimate value which is output by applying the output sensing data to the learning model that has undergone the machine learning. Ambient environment estimator may be included.

Description

SURROUNDED ENVIRONMENT ESTIMATION SYSTEM OF A PLURAOTY OF AUTOMOBILES, SERVER FOR ESTIMATING SURROUNDED ENVIRONMENT, AND SURROUNDED ENVIRONMENT ESTIMATION METHOD OF A PLURAOTY OF AUTOMOBILES}

The present invention relates to a system for estimating the surrounding environment of a plurality of moving objects, a server for estimating the surrounding environment, and a method for estimating the surrounding environment for a plurality of moving objects of the server. More particularly, the present invention relates to machine learning of a learning model using sensing data as input. The present invention relates to a surrounding environment estimation system for a plurality of moving objects, a server for surrounding environment estimation, and a surrounding environment estimation method for a plurality of moving objects of the server.

The driving assistance system detects dangers and warns that there is a risk of an accident through visual, audible and tactile factors, as well as actively decelerating or braking the vehicle to avoid forward collisions. It is a safety device.

In addition, the driver assistance system may perform lane departure warning, blind spot monitoring, improved rear monitoring, and the like.

The driving assistance system is classified into various types according to its function. Among them, the Forward Collision Warning System (FCW) detects vehicles driving in the same direction from the front of the driving lane and gives drivers a visual, audible and tactile warning to avoid collisions with the vehicles ahead. System.

Next, the Advanced Emergency Braking System (AEBS) detects the possibility of a collision with a car located in front of the driving lane and warns the driver, and if there is no driver response or a collision is inevitable, It is a system for automatically decelerating a car for the purpose of mitigation and avoidance.

However, such a driving assistance system judges and handles the risks of each individual vehicle by itself. Therefore, it is difficult to collect the data generated from the multiple vehicles in one place and determine the risk factors that are equally repeated in the multiple vehicles. There is.

Therefore, when the surrounding environment (for example, the road environment) of the plurality of moving objects is unsuitable for driving, it is necessary to apply braking to the current driving such as informing a vehicle driver of a dangerous condition.

SUMMARY OF THE INVENTION The present invention is derived from the above-mentioned necessity, and an object of the present invention is to apply a brake to current driving such as informing a vehicle driver of a dangerous state when the surrounding environment of a plurality of moving objects is not suitable for driving.

In addition, by collecting data generated from a plurality of moving objects in one place and separately managed, it is possible to enable the collective processing related to the notification of the dangerous state.

In particular, an object of the present invention is to collect more accurate information by utilizing a large amount of big data collected from sensors in a plurality of moving objects to obtain surrounding environment information necessary for estimating the surrounding environment of the plurality of moving objects.

In addition, by performing machine learning with the collected big data, it aims to collect more automated and real-time fast information.

According to an embodiment of the present invention, a system for estimating a surrounding environment having a server communicating with a plurality of moving objects through a network includes a plurality of moving objects and a server, and each of the plurality of moving objects outputs one or more sensing data outputting sensing data. An output unit and a first communication unit for transmitting the output sensing data to the server, wherein the server includes a second communication unit for receiving the sensing data from the plurality of moving objects and the received sensing data as input data. A surrounding environment estimator configured to perform machine learning based on a model, and estimate the surrounding environments of the plurality of moving objects based on the surrounding environment estimation value output by applying the output sensing data to the learning model that has undergone the machine learning. It may include.

According to another embodiment of the present invention, a server communicating with a plurality of moving objects through a network may include a second communication unit for receiving sensing data from a sensing data output unit of each of the plurality of moving objects and the received sensing data as input data. A machine environment based on a learning model to perform a machine learning, and an environment environment for estimating the environment surrounding the plurality of moving objects based on an environment value estimated by applying the received sensing data to the machine learning model that has been subjected to the machine learning. It may include an estimator.

According to another aspect of the present invention, there is provided a method for estimating the surrounding environment for a plurality of moving objects of a server communicating with a plurality of moving objects through a network, the method comprising: receiving sensing data from a sensing data output unit of each of the plurality of moving objects; Performing a machine learning based on a learning model using the received sensing data as input data; and applying the received sensing data to the learning model subjected to the machine learning to output the plurality of environment information based on an ambient environment estimated value. Estimating the surrounding environment of the mobile body.

According to another embodiment of the present invention, a system for estimating a surrounding environment having a server for communicating with a plurality of moving objects through a network includes a plurality of moving objects and a server, and each of the plurality of moving objects outputs one or more sensing data outputting sensing data. A data output unit and a first communication unit for transmitting the output sensing data to the server, wherein the server is based on a second communication unit receiving the sensing data from the plurality of moving objects and the learning model as the input of the sensing data. A learning unit for performing machine learning and generating update data for updating predetermined variables for estimating the surrounding environment of the plurality of moving objects through the machine learning, and updating the predetermined variable based on the update data. Ambient environment of the plurality of moving objects based on the part and the updated predetermined variable It may include estimating portion that estimates the environment.

According to another embodiment of the present invention, a server communicating with a plurality of moving objects through a network includes a second communication unit for receiving sensing data from a sensing data output unit of the plurality of moving objects, and a learning model having the sensing data as input. A learning unit for performing machine learning based on the machine learning and generating update data for updating predetermined variables for estimating the surrounding environment of the plurality of moving objects through the machine learning, and updating the predetermined variable based on the update data. And an surrounding environment estimating unit for estimating the surrounding environments of the plurality of moving objects based on the updating unit and the updated predetermined variable.

According to another aspect of the present invention, there is provided a method for estimating the surrounding environment for a plurality of moving objects of a server communicating with a plurality of moving objects through a network, the method comprising: receiving sensing data from a sensing data output unit of the plurality of moving objects; Performing machine learning based on a learning model that takes sensing data as input; generating update data for updating a predetermined variable for estimating a surrounding environment of the plurality of moving objects through the machine learning; And updating the predetermined variable based on the estimated variable, and estimating the surrounding environment of the plurality of moving objects based on the updated predetermined variable.

According to the present invention, when the surrounding environment of the plurality of moving objects is not suitable for driving, braking may be applied to the current driving such as informing a vehicle driver of a dangerous state.

In addition, by collecting data generated from a plurality of moving objects in one place and separately managing, it is possible to collectively process the notification of the dangerous state.

In particular, by using a large amount of big data collected from sensors in the plurality of moving objects to obtain the surrounding environment information necessary for estimating the surrounding environment of the plurality of moving objects, more accurate information can be collected.

In addition, by performing machine learning with the collected big data, more automated and real-time fast information can be collected.

1 and 2 are block diagrams of an environment estimation system according to an embodiment of the present invention.
3 shows the basic structure of a neural network.
4 to 7 are block diagrams of an environment estimation system, respectively, according to another embodiment of the present invention.

DETAILED DESCRIPTION OF THE INVENTION The following detailed description of the invention refers to the accompanying drawings that show, by way of illustration, specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It is to be understood that the various embodiments of the invention are different but need not be mutually exclusive. For example, certain shapes, structures, and characteristics described herein may be embodied in other embodiments without departing from the spirit and scope of the invention with respect to one embodiment.

In addition, it is to be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present invention, if properly described, is defined only by the appended claims, along with the full range of equivalents to which such claims are entitled. Like reference numerals in the drawings refer to the same or similar functions throughout the several aspects.

1 and 2 are block diagrams of a system for estimating the surrounding environment of a plurality of moving bodies according to an embodiment of the present invention.

1 and 2, the surrounding environment estimation system of the plurality of moving objects of the present invention may include a plurality of moving objects 100a, 100b.. 100n and a server 200.

Each of the plurality of moving objects 100a, 100b .. 100n outputs sensing data. The server 200 receives sensing data output from each of the plurality of moving objects 100a, 100b .. 100n, performs machine learning on the sensing data based on a predetermined learning model, and utilizes machine learning performance results. The surrounding environment of the plurality of moving objects 100a, 100b .. 100n can be estimated.

For example, if the sensing data output from each of the plurality of moving objects 100a, 100b..100n shows the same / similar predetermined pattern at the same road point, the road may be damaged or slippery. This can be estimated. The surrounding environment estimation information is collected by the server 200 and comprehensively analyzed, and the result is transmitted to each of the plurality of moving objects 100a, 100b .. 100n, and each of the plurality of moving objects 100a, 100b .. 100n. This result can be referred to later movement of the moving object.

FIG. 2 is a detailed block diagram of the plurality of moving objects 100a, 100b .. 100n (hereinafter referred to as the moving object 100) and the server 200 of FIG.

Specifically, referring to FIG. 2, the moving object 100 may include a sensing data output unit 110 and a first communication unit 120.

The sensing data output unit 110 outputs one or more sensing data. The sensing data output unit 110 may transmit one or more sensing data to the first communication unit 120.

The sensing data output unit 110 may be a moving sensor provided or mounted in the moving object. The moving object sensor is a position sensor (GPS), a pressure sensor, an acceleration speed sensor, an angle vehicle speed sensor, a flow sensor, a temperature sensor, an intake pressure sensor, a steering sensor, an anti-collision sensor, an airbag sensor, a throttle position sensor (TPS), an air temp sensor (ATS), accelerator pedal sensor (APS), vehicle speed sensor, G sensor, gyro sensor, collision detection sensor, garage sensor, suspension control sensor, air conditioner sensor, radar sensor, rider sensor, inertial measurement sensor, satellite navigation sensor, camera sensor , Pedal sensors, oxygen saturation sensors, infrared sensors, and the like. Here, the moving object sensor is not limited to those listed above, but should be understood as meaning all sensors provided or mounted on the moving object.

In addition, the sensing data of the sensing data output unit 110 may be directly transmitted to the first communication unit 120, or may be transmitted to the first communication unit 120 after a predetermined processing process according to another exemplary embodiment. In the latter case, the sensing data can be processed, sorted or processed and output.

The sensing data output unit 110 may include a plurality of sensing data output units 110a, 110b,..., 110n. The plurality of sensing data output units 110a, 110b,..., 110n may output different sensing data. For example, the first sensing data output unit 110a may output data sensed by the G sensor as the G sensor, and the second sensing data output unit 110b may be the APS sensor as the APS sensor and the APS sensor. It is possible to output the data detected by. Meanwhile, the plurality of sensing data output units 110a, 110b,..., 110n may output the same type of sensing data.

The sensing data output unit 110 may output one or more sensing data while the driver drives the moving object 100. The sensing data output unit 110 may output the sensing data in real time or may automatically output the predetermined time.

The first communication unit 120 receives one or more sensing data output from the sensing data output unit 110 and transmits the received sensing data to the server 200 which communicates through a wired / wireless network. The first communication unit 120 may transmit the processed sensing data to the server 200 after processing, modulating or encrypting the received sensing data.

The server 200 includes a second communication unit 220 and the ambient environment estimation 210.

The second communication unit 220 communicates with the first communication unit 120 of the mobile unit 100 through a wired / wireless network to receive predetermined data from the first communication unit 120 of the mobile unit 100. The received data includes one or more sensing data output from the sensing data output unit 110 of the moving object 100. The second communication unit 220 processes one or more sensing data output from the sensing data output unit 110 by processing, demodulating, or decoding predetermined data received from the first communication unit 120 of the moving object 100. You can extract data. The second communication unit 220 provides the extracted sensing data to the surrounding environment estimating unit 210.

The surrounding environment estimation unit 210 generates surrounding environment estimation information of the plurality of moving objects 100 based on one or more sensing data provided from the sensing data output unit 110. In addition, a predetermined learning model 215 is included to generate ambient environment estimation information.

In detail, the environment estimation unit 210 performs machine learning based on the learning model 215 using the sensing data provided from the sensing data output unit 110 as input data. The learning model 215 undergoes machine learning, and the surrounding environment estimator 210 applies the sensing data output from the sensing data output unit 110 to the learning model to output the surrounding environment estimate. In addition, the surrounding environment estimating unit 210 may estimate the surrounding environment of the plurality of moving objects based on the output of the surrounding environment estimation value.

The training model 215 is an artificial neural network (hereinafter, referred to as a neural network) that receives one or more sensing data provided from the sensing data output unit 110 and outputs output data associated with the surrounding environment of the plurality of moving objects. May be.) The basic structure of the neural network will be described with reference to FIG. 3.

The second communication unit 220 may receive surrounding environment estimation information of the plurality of moving objects output from the surrounding environment estimation unit 200, and transmit the received surrounding environment estimation information to the first communication unit 120.

In addition, the first communication unit 120 may receive surrounding environment estimation information of the plurality of moving objects from the second communication unit 220 and transmit the information to the controller (ECU, not shown). Then, the controller (ECU, not shown) may control the operation of the moving body later based on the surrounding environment estimation information. For example, when information such as a road in which the moving object 100 is located is lost or slippery is obtained, the controller causes the moving object 100 to avoid the corresponding part of the road.

As shown in FIG. 2, since a predetermined process for generating the surrounding environment estimation information is performed in the server 200, the advantage of enabling the batch processing of the surrounding environment estimation information related to the plurality of moving objects 100a, 100b .. 100n is possible. have.

3 shows the basic structure of a neural network.

Referring to FIG. 3, the basic structure of the neural network includes an input layer and an output layer. The input layer is a layer for receiving predetermined input data from the outside, and the output layer is a layer for outputting predetermined output data to the outside.

Each input layer and output layer includes one or more nodes. The node of the input layer is referred to as the input node, and the node of the output layer is referred to as the output node. Each input node of the input layer may be fully connected to each output node of the output layer as shown in FIG. 3, or may be incompletely connected to some output nodes of the output layer.

One output node of the output nodes of the output layer receives the input data from the input node (s) of the input layer connected thereto, and multiplies by a predetermined variable (or weight, w). Then, one output node sums up all the received ones and weights them, and then outputs predetermined output data through a preset activation function.

One output node has an activation function. The activation function is a step function, a sign function, a linear function, a logistic sigmoid function, a hyper tangent function, a ReLU function, and a softmax ( softmax) function. The activation function may be appropriately determined by a person skilled in the art according to the learning method of the neural network.

The neural network machine learns by repeatedly updating (or modifying) the variable w to an appropriate value. The neural network machine learning methods include supervised learning and unsupervised learning.

Supervised learning sets a variable w so that the output data obtained by inserting the input data into the neural network can be similar to the target data while the target output data desired by any neural network is calculated for the input data. It is a learning method to update. Representative neural networks that perform supervised learning include monolayer perceptron and multilayer perceptron as shown in FIG. 3. Here, the multilayer perceptron further has one or more hidden layers between the input and output layers shown in FIG. The hidden layer has one or more hidden nodes.

Unsupervised learning is a learning method that produces consistent output data for similar input data without setting target data that an arbitrary neural network wants to calculate for input data. Representative neural networks that perform unsupervised learning include self-organizing feature maps (SOMs) and the Boltzmann machine (Boltzmann machine).

Referring back to FIG. 2, the ambient environment estimation 210 includes a learning model 215, and the learning model 215 includes a neural network.

When the training model 215 is the neural network illustrated in FIG. 3, the sensing data provided by the sensing data output unit 110 is input to an input layer of the neural network. The number / type of sensing data provided may correspond one-to-one with the number of input nodes of the input layer. Each of the output nodes of the output layer may be associated with different ambient environment estimation information.

For example, as illustrated in FIG. 3, the first to fourth sensing data are input for each of the four input nodes of the input layer, and the output data of the first output node of the output layer may include data related to the first surrounding environment. The output data of the second output node may be data related to the second surrounding environment.

The training model 215 performs machine learning using one or more sensing data provided from the sensing data output unit 110. Implementing machine learning is the process of updating the variables of a neural network.

Specifically, when one or more sensing data is input to the input nodes of the input layer of the learning model 215, the learning model 215 outputs corresponding to the surrounding environment of the moving object 100 using the currently set variable. The data is output, and the learning model 215 updates the currently set variable by using the output data. The process of updating the set variable is a machine learning process. This process is repeated repeatedly. The sensing data is applied to the learning model 215 composed of updated variables to estimate the surrounding environment of the moving object 100. Here, the learning model 215 according to the supervised learning method, the learning model 215 may update the currently set variable so that the output data is similar to the target data. Meanwhile, according to the non-supervised learning method of the learning model 215, the learning model 215 may update a currently set variable such that similar sensing data become consistent output data.

The training model 215 uses the neural network shown in FIG. 2, but is not limited thereto. The neural network of the training model 215 may be implemented by any one of a multilayer perceptron, a self-organizing feature map, and a Boltzmann machine. Other neural networks may be implemented.

As described above, according to the surrounding environment estimation system illustrated in FIGS. 1 and 2, one or more sensing data output from the sensing data output unit 110 are provided to the learning model 215 of the surrounding environment estimation unit 210. do. The learning model 215 outputs predetermined output data using the provided sensing data as an input, and the output data corresponds to a predetermined surrounding environment. In addition, the learning model 215 updates the variables in the learning model 215 based on the output data to perform machine learning.

Although not illustrated in FIGS. 1 and 2, a preprocessor (not shown) may be disposed between the sensing data output unit 110 and the surrounding environment estimation unit 210. The preprocessor (not shown) may be disposed in front of the learning model 215 in the surrounding environment estimator 210.

The preprocessor (not shown) processes one or more sensing data provided by the sensing data output unit 110. One or more sensing data processed by the preprocessor (not shown) is provided to the surrounding environment estimator 210. Here, processing the one or more sensing data by the preprocessor (not shown) may be normalizing the sensing data. Normalization may reduce cell saturation of the learning model 215 of the surrounding environment estimator 210.

In addition, although not shown in FIGS. 1 and 2, the movable body 100 may further include a memory unit (not shown). The memory unit (not shown) may separately store one or more sensing data provided from the sensing data output unit 110.

4 is a block diagram of an ambient environment estimation system according to another embodiment of the present invention.

The surrounding environment estimation system according to another embodiment of the present invention shown in FIG. 4 is compared with the surrounding environment estimation system according to the embodiment of the present invention shown in FIGS. 1 and 2. There is a difference. Therefore, the surrounding environment estimator 210 will be described in detail below.

The surrounding environment estimator 210 includes a plurality of learning models 215a and 215b. The plurality of learning models 215a and 215b may include at least a first learning model 215a and a second learning model 215b. Each of the plurality of learning models 215a and 215b may be a neural network shown in FIG. 3. The plurality of learning models 215a and 215b estimate the surrounding environment of the moving body based on one or more sensing data (s) provided from the sensing data output unit 110, and each of the plurality of learning models 215a and 215b. Updates the currently set variables with the estimates of the environment.

In detail, one or more sensing data (s) provided from the sensing data output unit 110 are input to the first learning model 215a, and the first learning model 215a outputs predetermined output data as a result of performing the machine learning. Output (s) The predetermined output data (s) output from the first learning model 215a is input to the second learning model 215b, and the second learning model 215b performs machine learning again to perform predetermined output data (s). Outputs Here, the output data (s) output from the first learning model 215a may be referred to as intermediate data. In the first learning model 215a and the second learning model 215b, the machine learning is repeatedly performed to update a variable set therein, and the sensing data output from the sensing data output unit 100 is output to the first learning model. It applies to 215a and the 2nd learning model 215b, and outputs the surrounding environment estimation value of a mobile body.

The output data (s) output from the second learning model 215b is associated with the surrounding environment of the moving object. That is, the output data (s) output from the second learning model 215b may be used as a value indicating the surrounding environment of the moving object.

Meanwhile, all the sensing data provided by the sensing data output unit 110 may not be input to the first learning model 215a. In other words, some of the sensing data provided by the sensing data output unit 110 may not be input to the first learning model 215a but may be directly input to the second learning model 215b. The second learning model 215b may use output data output from the first learning model 215a and sensing data provided from the sensing data output unit 110 as input data.

In FIG. 4, two learning models in the surrounding environment estimator 210 are used, but the learning models may be three or more. Here, the learning data of the three or more learning models may be used as output data of the previous learning model as input data of the next learning model, or sensing data of some of the sensing data provided from the sensing data output unit 110 is input into one learning model. Some other sensing data may be input to another learning model, and output data output from one learning model and another learning model may be input to another learning model.

Meanwhile, output data output from the second learning model 215b is transmitted to the moving object 100 via the second communication unit 220. Then, the controller ECU (not shown) can control the performance of the moving object based on the corresponding output data.

The surrounding environment estimation system according to another embodiment of the present invention shown in FIG. 4 is compared to the surrounding environment estimation system according to the embodiment of the present invention shown in FIGS. 1 and 2 by using a plurality of learning models. Estimate the surrounding environment. Therefore, the surrounding environment estimation result may be faster and faster, and the optimization of the variables in each learning model may be faster.

5 is a block diagram of an ambient environment estimation system according to still another embodiment of the present invention.

The surrounding environment estimation system according to still another embodiment of the present invention shown in FIG. 5 is compared with the surrounding environment estimation system according to the embodiment of the present invention shown in FIGS. 1 and 2. There is a difference. Therefore, the surrounding environment estimator 210 will be described in detail below.

The surrounding environment estimator 210 includes a learning model 215 and a feedback unit 216. The learning model 215 is the same as the learning model 215 shown in FIG. 1, and a detailed description thereof will be omitted.

The feedback unit 216 determines whether the variable of the learning model 215 is updated, and the variable updating of the learning model 215 may be controlled according to the determination.

The feedback unit 216 receives the output data of the learning model 215, and determines whether to update the variable of the learning model 215 based on the received output data. For example, when the output data output from the learning model 215 indicates a predetermined ambient environment estimate value, the feedback unit 216 determines whether the ambient environment estimate value output from the learning model 215 is abnormal output data. . In other words, it is determined whether the ambient environmental estimate is out of the normal range. If the output data output from the training model 215 is not abnormal data, the feedback unit 216 allows for subsequent variable updating of the training model 215, that is, repetitive machine learning, and if the data is abnormal, the feedback is feedback. The unit 216 does not allow subsequent variable updating of the learning model 215, that is, repetitive machine learning.

If it is determined that the abnormal surrounding environment through the feedback unit 216, further repetitive machine learning is disallowed, thereby preventing an overload on performing the machine learning of the surrounding environment estimation unit.

Meanwhile, the learning model 215 illustrated in FIG. 5 may be replaced with a plurality of learning models 215a and 215b illustrated in FIG. 4. In this case, the feedback unit 216 may determine whether to update the variables of each of the plurality of learning models 215a and 215b and control the updating of the variables of each of the plurality of learning models 215a and 215b.

6 is a block diagram of an ambient environment estimation system according to still another embodiment of the present invention.

A surrounding environment estimation system according to another embodiment of the present invention shown in FIG. 6 is different from the surrounding environment estimation system 210 in comparison with the surrounding environment estimation system according to the embodiment of the present invention shown in FIG. 1. have. Therefore, below, the surrounding environment estimation unit 210 will be described in detail.

The surrounding environment estimator 210 includes a learning model 215 and a surrounding environment refiner 217. The learning model 215 is the same as the learning model 215 shown in FIG. 1, and a detailed description thereof will be omitted.

The surrounding environment refiner 217 subdivides the surrounding environment estimate value output from the learning model 215, and the surrounding environment estimator 210 based on the subdivided surrounding environment estimate value output from the surrounding environment refiner 217. It is possible to estimate the surrounding environment.

For example, the ambient environment estimate output from the learning model 215 needs to be subdivided into quadratic. When the road dig information is output, it is necessary to distinguish whether the road dig is 'strong', 'medium' or 'weak'. The predetermined threshold for determining whether the road is dug is '30', and the case where the ambient environment estimate output from the learning model 215 is '40', '50', and '60' are the same. In addition to indicating the indentation of the road, and in each case can be determined as' weak, 'middle', 'river', more accurate information can be delivered. Thus, according to the embodiment of FIG. 6, the surrounding environment segmentation 217 subdivides the surrounding environment estimation value output from the learning model 215, and the surrounding environment estimation unit 210 is output from the surrounding environment segmentation unit 217. It is possible to estimate the surrounding environment of the moving body based on the refined surrounding environment estimate.

The surrounding environment refiner 217 may be implemented by a predetermined algorithm other than the neural network used in the learning model 215.

Meanwhile, the learning model 215 illustrated in FIG. 6 may be replaced with a plurality of learning models 215a and 215b illustrated in FIG. 4. In this case, the surrounding environment refiner 217 may receive predetermined output data from one or more learning models, and output the divided output data.

One or more sensing data provided to the learning model 215 according to FIGS. 2, 4, 5, and 6 and output data output from the learning model 215 may be defined as various examples. It will be described in detail below.

<Ambient Environment Estimation-Road Breaking Information>

For example, the one or more sensing data provided to the input node (s) of the training model 151 may include the position data sensed by the position sensor GPS, the acceleration data sensed by the G sensor, and the steering sensor. The steering wheel rotational speed data sensed by the steering sensor, the steering wheel rotation angle data sensed by the steering sensor, and the steering wheel rotation direction data sensed by the steering sensor. Output

In detail, when a change in the acceleration data value sensed by the G sensor is sharp at the position sensed by the position sensor GPS, it may be determined that a predetermined shock is generated in the moving object due to the sudden change. In addition, even when a change in at least one data value among the steering wheel rotation angle, the rotation speed, and the rotation direction sensed by the steering sensor is sudden, it is determined that the moving object 100 tries to avoid a specific object, which is, May be judged to be dug.

Whether the change amount of the acceleration data value, the handle rotation angle change amount, the rotation speed change amount, and the rotation direction change amount is large is determined based on a predetermined threshold value, and when the threshold value is exceeded, the road may be determined to be dug. The threshold information may be preset and stored in a memory (not shown).

The information output by the G sensor and each information output by the steering sensor may be used to determine that the road is dug separately or in combination of at least one.

<Ambient Environment Estimation-Road Slip Information>

For example, the one or more sensing data provided to the input node (s) of the training model 151 may include position data sensed by a position sensor (GPS), valve rotation angle data sensed by a TPS sensor, and airbags. The impact force sensing data sensed by the sensor is included, and the output node (s) of the training model 151 includes slip information of the road.

Specifically, in the position sensed by the position sensor GPS, when the amount of change in the valve rotation angle data sensed by the TPS sensor is large or the magnitude of the impact force sensing data sensed by the airbag sensor is large, the road at the position is It is determined that the road is icy, and slip information can be output.

The amount of change in the valve rotation angle data sensed by the TPS sensor or the size of the impact force sensing data sensed by the airbag sensor is determined based on a predetermined threshold, and the threshold information is preset and stored in a memory (not shown). Can be.

The data output by the TPS sensor and the data output by the airbag sensor may be used to output slip information, respectively, or in combination with at least one.

<Ambient Environment Estimation-Curvature Information of Road>

For example, the one or more sensing data provided to the input node (s) of the training model 151 may include position data sensed by the position sensor (GPS), acceleration data sensed by the G sensor, and gyro sensor. It includes the rotation speed data sensed by the output node, the output node (s) of the learning model 151 includes the degree of curvature of the road.

Specifically, in a position sensed by the position sensor (GPS), when the change in the acceleration data value sensed by the G sensor is sudden or the change in the rotation speed data value sensed by the gyro sensor is sudden, the road of the corresponding position The curvature of can be judged to be relatively larger than other roads.

Whether the change in the acceleration data value sensed by the G sensor is abrupt or the change in the rotation speed data value sensed by the gyro sensor is determined based on a predetermined threshold value, and the threshold information is determined by a memory (not shown). May be preset and stored.

The data output by the G sensor and the data output by the gyro sensor may be used to output curvature information of the road, respectively or in combination.

7 is a block diagram of an ambient environment estimation system according to yet another embodiment of the present invention.

Referring to FIG. 7, the surrounding environment estimation system according to another embodiment of the present invention includes a moving object 100 and a server 200.

The moving object 100 is implemented with one or more moving objects, and includes a sensing data output unit 110 and a first communication unit 120.

The sensing data output unit 110 is the same as the sensing data output unit 110 shown in FIG. 2.

The sensing data output unit 110 may output one or more sensing data, and the output sensing data may be provided to the learner 230 and the surrounding environment estimator 210.

The first communication unit 120 transmits one or more sensing data received from the sensing data output unit 110 to the server 200 which communicates through a network. The first communication unit 120 may transmit the processed sensing data to the server 200 after processing, modulating or encrypting the received sensing data.

The server 200 includes a learner 230, an updater 240, an ambient environment estimator 210, and a second communication unit 220. However, the learning unit 230 receives one or more sensing data from the second communication unit 220.

The second communication unit 220 receives predetermined data from the first communication unit 120 of the moving object 100. The received data includes one or more sensing data output from the sensing data output unit 210 of the moving object 100. The second communication unit 220 processes one or more sensing data output from the sensing data output unit 110 by processing, demodulating, or decoding predetermined data received from the first communication unit 120 of the moving object 100. You can extract data. The second communication unit 220 provides the extracted sensing data to the learner 230.

The learner 230 may have a learning model that performs machine learning based on the sensing data received from the second communication unit 220, and may perform machine learning based on the learning model. Here, the learning model may be a neural network shown in FIG. 3.

The learner 230 may generate update data for updating a predetermined variable for estimating the surrounding environment of the surrounding environment estimator 210 based on the result of the machine learning. The update data includes information about variables (or weights) of the machine-learned learning model of the learner 230. For example, assuming that the learning model machine-trained by the sensing data is currently set to a predetermined variable, the learner 230 may generate information about the current predetermined variable as update data.

The learner 230 provides the generated update data to the updater 240.

The updater 240 controls the surrounding environment estimator 210 based on the update data provided from the learner 230. Specifically, the updater 240 controls to update a predetermined variable of the surrounding environment estimator 210 according to the provided update data.

The surrounding environment estimator 210 estimates the surrounding environment of the moving object based on one or more sensing data provided from the sensing data output unit 210 and predetermined updated variables provided from the updater 240. To this end, the surrounding environment estimator 210 may have a predetermined learning model or algorithm. Here, the learning model of the surrounding environment estimator 210 may be the same as the learning model of the learner 230 or may be different from the learning model of the learner 230, but may be another neural network as illustrated in FIG. 3. Can be. In this case, the learning model of the surrounding environment estimator 210 has a predetermined variable. The predetermined variable is updated by the updater 240. Meanwhile, the surrounding environment estimator 210 may be a predetermined algorithm. In this case, the predetermined algorithm may be controlled based on the update data.

For example, when the learning model of the learning unit 230 and the learning model of the surrounding environment estimating unit 210 are the same, one or more sensing data output from the sensing data output unit 110 may be learned by the learning unit 230. It is input to the input node of the model, through which the learning model of the learning unit 230 performs machine learning. The predetermined variable of the learning model is updated by machine learning of the learning model of the learning unit 230. The learner 230 outputs updated information about the predetermined variable as update data. The output update data is provided to the updater 240, and the updater 240 updates predetermined variables of the learning model of the surrounding environment estimator 210 based on the update data.

The surrounding environment estimation unit 210 may generate the surrounding environment estimation information of the plurality of moving objects 100 based on the updated predetermined variable. The surrounding environment estimation information is transmitted to the second communication unit 220 and the first communication unit 120, so that each controller (not shown) of the plurality of moving objects 100 may move the moving object 100 based on the corresponding surrounding environment estimation information. You can control the operation of.

In the surrounding environment estimation system according to the embodiment of the present invention illustrated in FIG. 7, the learning unit 230 and the surrounding environment estimation unit 210 are distinguished from the surrounding environment estimation system shown in FIG. 2. By distinguishing the learning unit 230 and the surrounding environment estimating unit 210, an effect that cannot be obtained through the surrounding environment estimation system shown in FIG. 2 may be exhibited.

For example, in the surroundings estimation system according to the embodiment of the present invention shown in FIG. 7, even when the updated variable information of the learning unit 230 is deleted due to a system error, the surroundings estimation unit 210 is updated. Since the variable information is stored separately, recovery of the variable information of the learner 230 may be facilitated.

Meanwhile, the surrounding environment estimator 210 illustrated in FIG. 7 may be replaced with the surrounding environment estimator 210 illustrated in FIGS. 4 to 6. In this case, the functions of the surrounding environment estimator shown in FIGS. 4 to 6 may be applied in the same or similar manner.

On the other hand, also in the case of Figure 7, the above-mentioned <environmental environment estimation-road depression information>, <ambient environment estimation-road slip information> <ambient environment estimation-road curvature information> are the same / similar to each other Can be applied.

The features, structures, effects, and the like described in the above embodiments are included in one embodiment of the present invention, and are not necessarily limited to only one embodiment. Furthermore, the features, structures, effects, and the like illustrated in the embodiments may be combined or modified with respect to other embodiments by those skilled in the art to which the embodiments belong. Therefore, contents related to such combinations and modifications should be construed as being included in the scope of the present invention.

Claims (15)

An ambient environment estimation system having a server that communicates with a plurality of moving objects through a network,
Each of the plurality of moving bodies,
One or more sensing data output units configured to output sensing data; And
A first communication unit which transmits the output sensing data to the server;
Including,
The server,
A second communication unit configured to receive the sensing data from the plurality of moving objects; And
Machine learning is performed based on a learning model using sensing data as input data, and the surroundings of the plurality of moving objects are based on an ambient environment estimated value that is output by applying the received sensing data to the learning model that has undergone the machine learning. Includes; surrounding environment estimation unit for estimating the environment,
The learning model is an artificial neural network using the sensing data as the input data and an ambient environment estimate corresponding to the sensing data as output data.
The surrounding environment estimator may input the sensing data as the input data to the input node through the artificial neural network, and output the surrounding environment estimate value as the output data from the output node to perform the machine learning.
Estimating the surrounding environment of the plurality of moving objects by inputting the received sensing data to the input node and outputting the surrounding environment estimate from the output node,
A surrounding environment estimation system of a plurality of moving objects.
The method of claim 1,
The learning model includes a first learning model and a second learning model,
The surrounding environment estimator performs machine learning based on the first learning model using the sensing data as input data, and based on the second learning model using output data output from the first learning model as input data. To do machine learning,
A surrounding environment estimation system of a plurality of moving objects.
The method of claim 1,
The surrounding environment estimator may further include a feedback unit configured to determine whether to perform the later machine learning of the learning model based on the surrounding environment estimate value output from the learning model.
A surrounding environment estimation system of a plurality of moving objects.
The method of claim 1,
The surrounding environment estimator may further include an surrounding environment refiner configured to subdivide the surrounding environment estimate value output from the learning model.
The surrounding environment estimating unit estimates the surrounding environment of the plurality of moving objects based on the subdivided surrounding environment estimation values output from the surrounding environment subdividing unit.
A surrounding environment estimation system of a plurality of moving objects.
In a server that communicates with a plurality of mobile objects through a network,
A second communication unit configured to receive sensing data from a sensing data output unit of each of the plurality of moving objects; And
Machine learning is performed based on a learning model using sensing data as input data, and the surroundings of the plurality of moving objects are based on an ambient environment estimated value that is output by applying the received sensing data to the learning model that has undergone the machine learning. Includes; surrounding environment estimation unit for estimating the environment,
The learning model is an artificial neural network using the sensing data as the input data and an ambient environment estimate corresponding to the sensing data as output data.
The surrounding environment estimator may input the sensing data as the input data to the input node through the artificial neural network, and output the surrounding environment estimate value as the output data from the output node to perform the machine learning.
Estimating the surrounding environment of the plurality of moving objects by inputting the received sensing data to the input node and outputting the surrounding environment estimate from the output node,
server.
In the surrounding environment estimation method for the plurality of moving objects of the server that communicates with the plurality of moving objects through the network,
Receiving sensing data from a sensing data output unit of each of the plurality of moving objects;
Performing machine learning based on a learning model using the sensing data as input data; And
Estimating the surrounding environment of the plurality of moving objects based on the surrounding environment estimate output by applying the received sensing data to the learning model that has undergone the machine learning.
The learning model is an artificial neural network using the sensing data as the input data and an ambient environment estimate corresponding to the sensing data as output data.
In the surrounding environment estimation step, the sensing data is input to the input node as the input data through the artificial neural network, and the ambient environment estimate is output from the output node as the output data to perform the machine learning. Estimating the surrounding environment of the plurality of moving objects by inputting the received sensing data to and outputting the surrounding environment estimate from the output node,
A method for estimating the surrounding environment for a plurality of moving objects of the server.
The method of claim 1,
The sensing data output unit may include at least one of a position sensor, a G sensor, and a steering sensor.
The sensing data may include position data sensed by the position sensor, acceleration data sensed by the G sensor, handle rotation speed data sensed by the steering sensor, handle rotation angle data sensed by the steering sensor, and At least one of the handle rotation direction data sensed by the steering sensor,
The surrounding environment estimated by the surrounding environment estimating unit includes depression information of a road.
A surrounding environment estimation system of a plurality of moving objects.
The method of claim 1,
The sensing data output unit includes a position sensor, a TPS sensor, and an airbag sensor,
The sensing data includes at least one of position data sensed by the position sensor, valve rotation angle data sensed by the TPS sensor, and impact force sensing data sensed by the airbag sensor,
The surrounding environment estimated by the surrounding environment estimating unit includes slippage information of a road.
A surrounding environment estimation system of a plurality of moving objects.
The method of claim 1,
The sensing data output unit includes a position sensor, a G sensor, and a gyro sensor,
The sensing data includes at least one of position data sensed by the position sensor, acceleration data sensed by the G sensor, and rotation speed data sensed by the gyro sensor.
The surrounding environment estimated by the surrounding environment estimating unit includes curvature degree information of a road.
A surrounding environment estimation system of a plurality of moving objects.
An ambient environment estimation system having a server that communicates with a plurality of moving objects through a network,
Each of the plurality of moving bodies,
One or more sensing data output units configured to output sensing data; And
A first communication unit which transmits the output sensing data to the server;
Including,
The server,
A second communication unit configured to receive the sensing data from the plurality of moving objects; And
A learning unit that performs machine learning based on a learning model that takes sensing data as inputs, and generates update data for updating predetermined variables for estimating the surrounding environment of the plurality of moving objects through the machine learning;
An update unit for updating the predetermined variable based on the update data; And
A surrounding environment estimating unit for estimating the surrounding environment of the plurality of moving objects based on the updated predetermined variable;
The learning model is an artificial neural network using the sensing data as input data and an ambient environment estimation value for outputting data for estimating the surrounding environment corresponding to the sensing data.
The surrounding environment estimator may input the sensing data as the input data to the input node through the artificial neural network, and output the surrounding environment estimate value as the output data from the output node to perform the machine learning.
Estimating the surrounding environment of the plurality of moving objects by inputting the received sensing data to the input node and outputting the surrounding environment estimate from the output node,
A surrounding environment estimation system of a plurality of moving objects.
In a server that communicates with a plurality of mobile objects through a network,
A second communication unit configured to receive sensing data from sensing data output units of the plurality of moving objects;
A learning unit that performs machine learning based on a learning model that takes sensing data as input, and generates update data for updating predetermined variables for estimating the surrounding environment of the plurality of moving objects through machine learning of the received sensing data. ;
An update unit for updating the predetermined variable based on the update data; And
A surrounding environment estimating unit for estimating the surrounding environment of the plurality of moving objects based on the updated predetermined variable;
The learning model is an artificial neural network using the sensing data as input data and an ambient environment estimation value for outputting data for estimating the surrounding environment corresponding to the sensing data.
The surrounding environment estimator may input the sensing data as the input data to the input node through the artificial neural network, and output the surrounding environment estimate value as the output data from the output node to perform the machine learning.
Estimating the surrounding environment of the plurality of moving objects by inputting the received sensing data to the input node and outputting the surrounding environment estimation value from the output node,
server.
In the surrounding environment estimation method for the plurality of moving objects of the server that communicates with the plurality of moving objects through the network,
Receiving sensing data from sensing data output units of the plurality of moving objects;
Performing machine learning based on a learning model as an input of the sensing data;
Generating update data for updating a predetermined variable for estimating the surrounding environment of the plurality of moving objects through the machine learning;
Updating the predetermined variable based on the update data; And
Estimating a surrounding environment of the plurality of moving objects based on the updated predetermined variable;
The learning model is an artificial neural network using the sensing data as input data and an ambient environment estimation value for outputting the ambient environment as output data.
In the surrounding environment estimation step, the sensing data is input to the input node as the input data through the artificial neural network, and the surrounding environment estimate is output from the output node as the output data to estimate the surrounding environment of the plurality of moving objects. ,
A method for estimating the surrounding environment for a plurality of moving objects of the server.
The method of claim 10,
The sensing data output unit includes a G sensor and a steering sensor,
The sensing data may include acceleration data sensed by the G sensor, handle rotation speed data sensed by the steering sensor, handle rotation angle data sensed by the steering sensor, and handle rotation direction data sensed by the steering sensor. Including,
The surrounding environment estimated by the surrounding environment estimating unit includes depression information of a road.
A surrounding environment estimation system of a plurality of moving objects.
The method of claim 10,
The sensing data output unit includes a TPS sensor and an airbag sensor,
The sensing data may include valve rotation angle data sensed by the TPS sensor and impact force sensing data sensed by the airbag sensor,
The surrounding environment estimated by the surrounding environment estimating unit includes slip information of a road.
A surrounding environment estimation system of a plurality of moving objects.
The method of claim 10,
The sensing data output unit includes a G sensor and a gyro sensor.
The sensing data may include acceleration data sensed by the G sensor and rotation speed data sensed by the gyro sensor.
The surrounding environment estimated by the surrounding environment estimating unit includes curvature degree information of a road.
A surrounding environment estimation system of a plurality of moving objects.

Images