KR20230074884A - Apparatus for classifying lane occupancy situation and method thereof - Google Patents

Abstract

The present invention relates to an apparatus for classifying a lane occupancy situation and a method thereof. Provided is an apparatus for classifying a lane occupancy situation comprising: a radar module that generates radar point data for surrounding targets, generates a track based on the radar point data, and provides characteristic information of the radar point data within an area of interest including tracks located in at least one of a driver's own lane in front of a driver's own vehicle and the lanes on both sides; a pre-processing module that generates learning data by pre-processing characteristic information of the radar point data and driving information of the driver's own vehicle; and a deep neural network that constructs a lane occupancy situation classification model by performing learning using the learning data to classify the occupancy situation of the driver's own lane in front of the driver's own vehicle and the lanes on both sides according to a plurality of lane occupancy situation classes and classifies a lane occupancy situation by using the lane occupancy situation classification model. More accurate information for the vehicles in front may be provided.

Description

Apparatus and method for classifying lane occupancy situations {APPARATUS FOR CLASSIFYING LANE OCCUPANCY SITUATION AND METHOD THEREOF}

The present embodiments relate to an apparatus and method for classifying a lane occupancy situation.

Recently, as interest in safety and driver convenience increases, various vehicle safety and convenience technologies are being developed. For example, various technologies such as smart cruise technology, automatic driving technology, and automatic emergency stop technology that detects a vehicle in front and automatically follows and drives the vehicle in front are being developed.

To realize these technologies, vehicles are equipped with various sensors such as cameras, LiDAR, and LADAR. Among them, the radar is less affected by the weather environment, unlike lidar and cameras, and has the advantage of being able to stably measure targets around the vehicle not only at short distances but also at long distances.

The present embodiments may provide an apparatus and method for classifying a lane occupancy situation.

In one aspect, the present embodiments generate radar point data for surrounding targets, generate a track based on the radar point data, and generate a region of interest including a track located in at least one of an own lane and both side lanes in front of the own vehicle. a radar module providing characteristic information of radar point data within; a pre-processing module generating learning data by pre-processing characteristic information of the radar point data and own vehicle driving information; and learning to classify the occupancy situation of the own lane and both side lanes in front of the own vehicle into a plurality of lane occupancy situation classes using the learning data to construct a lane occupancy situation classification model, and using the lane occupancy situation classification model An apparatus for classifying lane occupancy situations may be provided, comprising a deep neural network for classifying lane occupancy situations.

In another aspect, the present embodiments provide the steps of acquiring characteristic information of radar point data in an area of interest including a track located in at least one of the own lane and both side lanes in front of the own vehicle, and driving information of the own vehicle; generating learning data by preprocessing characteristic information of the radar point data and driving information of the host vehicle; constructing a lane occupancy situation classification model by inputting the learning data into a deep neural network and learning to classify occupancy situations of the own lane and both side lanes in front by a plurality of lane occupancy situation classes; and classifying lane occupancy situations using the lane occupancy situation classification model.

According to the present embodiments, the deep neural network is trained to classify the lane occupancy situation in front using the distribution of the driving information of the own vehicle and the characteristic information of the radar point data, and the lane occupancy situation is classified using the learned deep neural network. More accurate information about the vehicle in front can be provided.

1 is a block diagram of an apparatus for classifying a lane occupancy situation according to the present disclosure.
2 is a diagram illustrating an embodiment of generating radar point data and tracks according to the present disclosure.
3 is a diagram illustrating an embodiment of generating learning data according to the present disclosure.
4 is a diagram illustrating seven lane occupancy situation classes according to the present disclosure.
5 is a diagram illustrating an example of a lane occupancy situation classification model according to the present disclosure.
6 is a flowchart illustrating a method for classifying lane occupancy situations according to the present disclosure.
7 is a flowchart illustrating a process of obtaining characteristic information of radar point data.
8 is a diagram schematically illustrating a process of classifying a lane occupancy situation using a deep neural network according to the present disclosure.

DETAILED DESCRIPTION Some embodiments of the present disclosure are described in detail below with reference to exemplary drawings. In adding reference numerals to components of each drawing, the same components may have the same numerals as much as possible even if they are displayed on different drawings. In addition, in describing the present embodiments, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present technical idea, the detailed description may be omitted. When "comprises", "has", "consists of", etc. mentioned in this specification is used, other parts may be added unless "only" is used. In the case where a component is expressed in the singular, it may include the case of including the plural unless otherwise explicitly stated.

Also, terms such as first, second, A, B, (a), and (b) may be used in describing the components of the present disclosure. These terms are only used to distinguish the component from other components, and the nature, sequence, order, or number of the corresponding component is not limited by the term.

In the description of the positional relationship of components, when it is described that two or more components are "connected", "coupled" or "connected", the two or more components are directly "connected", "coupled" or "connected". ", but it will be understood that two or more components and other components may be further "interposed" and "connected", "coupled" or "connected". Here, other components may be included in one or more of two or more components that are “connected”, “coupled” or “connected” to each other.

In the description of the temporal flow relationship related to components, operation methods, production methods, etc., for example, "after", "continued to", "after", "before", etc. Alternatively, when a flow sequence relationship is described, it may also include non-continuous cases unless “immediately” or “directly” is used.

On the other hand, when a numerical value or corresponding information (eg, level, etc.) for a component is mentioned, even if there is no separate explicit description, the numerical value or its corresponding information is not indicated by various factors (eg, process factors, internal or external shocks, noise, etc.) may be interpreted as including an error range that may occur.

In the present disclosure, a vehicle may be a concept including a car, a motorcycle, and the like. In addition, the vehicle may be a concept including all of an internal combustion engine vehicle having an engine as a power source, a hybrid vehicle having an engine and an electric motor as a power source, and an electric vehicle having an electric motor as a power source. Hereinafter, vehicles will be mainly described with respect to vehicles.

In the following description, front means a forward driving direction of the vehicle, and rear means a backward driving direction of the vehicle. In addition, the own lane means the lane where the own vehicle is located, and both side lanes mean the left and right lanes of the own lane.

In the present disclosure, a target refers to a moving object such as another vehicle or a person around the vehicle or an object including infrastructure facilities around the vehicle. A target may be used interchangeably with an object, and may be used as a meaning of an object that can be used to control a vehicle among objects.

Hereinafter, an apparatus and method for classifying a lane occupancy situation according to embodiments of the present disclosure will be described with reference to the accompanying drawings.

1 is a block diagram of an apparatus for classifying a lane occupancy situation according to the present disclosure.

Referring to FIG. 1 , an apparatus for classifying a lane occupancy situation according to the present disclosure generates radar point data for a surrounding target, generates a track based on the radar point data, and generates an own lane and both side lanes in front of the own vehicle. A radar module 10 providing characteristic information of radar point data in an area of interest including a track located on at least one of the , and a preprocessing module generating learning data by preprocessing characteristic information of the radar point data and vehicle driving information. (20), using the learning data, learn to classify the occupancy situation of the own lane and both side lanes in front of the own vehicle into a plurality of lane occupancy situation classes, construct a lane occupancy situation classification model, and use the lane occupancy situation classification model and a deep neural network 30 that classifies the lane occupancy situation.

In detail, the radar module 10 may transmit radar signals at predetermined intervals and receive signals reflected from targets to scan surrounding targets. The radar module 10 may include at least one transmit antenna for transmitting a radar signal to the outside of the vehicle and at least one receive antenna for receiving a signal reflected from a target around the vehicle. The radar module 10 may be a frequency modulated continuous wave (FMCW) radar.

The radar module 10 may include at least one of a front radar module mounted on the front of the vehicle, a rear radar module mounted on the rear of the vehicle, and a lateral or side-rear radar module mounted on each side of the vehicle. The present disclosure is preferably applied to a forward radar module.

The radar module 10 receives a signal from which the transmitted radar signal is reflected from the target, generates radar point data, which is a point where the signal is reflected from the target, and analyzes the radar point data to detect characteristic information of the radar point data. and may include an electronic control unit (ECU) or processor for this purpose. Data transfer or signal communication from the radar sensor of the radar module 10 to the ECU may utilize a communication link, such as a suitable vehicle network bus or the like.

The radar module 10 may generate a track based on radar point data and detect speed information of the track and location information of the track. In addition, the radar module 10 may determine whether the track is in a stationary state based on the speed information of the track, and based on the location information of the track, the track is located in at least one of the own lane and the left and right side lanes in front of the own vehicle. You can judge whether or not to do it. In one embodiment, a track located in at least one of the own lane and both side lanes in front of the own vehicle may be defined as a track requiring lateral resolution. In another embodiment, a track in a stationary state and located in at least one of the own lane and both side lanes in front of the own vehicle may be defined as a track requiring lateral resolution.

The radar module 10 may detect characteristic information of radar point data within a region of interest including a track requiring lateral resolution and provide the detected information to the preprocessing module 20 . Here, the characteristic information of the radar point data may include distance information and azimuth information of the radar point data.

The lane occupancy situation classification apparatus according to an embodiment may further include a driving information sensing module 40 . The driving information sensing module 40 may sense driving information of the host vehicle.

The driving information may include, for example, at least one of steering torque, steering angle, provisional motor information, vehicle speed, vehicle motion information, and vehicle attitude information, and the driving information sensing module 40 is configured to sense the steering torque. It may include at least one of a torque sensor, a steering angle sensor for sensing a steering angle, a motor position sensor for sensing information about a steering motor, a vehicle speed sensor, a vehicle motion detection sensor for sensing the movement of a vehicle, and a vehicle posture detection sensor. there is. In addition to this, the driving information sensing module 40 may mean a sensor for sensing various data inside the vehicle.

The preprocessing module 20 may generate learning data by preprocessing characteristic information of radar point data provided from the radar module 10 and driving information of the host vehicle provided from the driving information sensing module 40 .

The deep neural network 30 learns to classify the occupancy situations of the own lane and the left and right lanes in front of the own vehicle into 7 lane occupancy situation classes using the learning data, builds a lane occupancy situation classification model, and classifies the lane occupancy situation Lane occupancy situations can be classified using the model. Seven lane occupancy situation classes will be described later with reference to FIG. 4 .

2 is a diagram illustrating an embodiment of generating radar point data and tracks according to the present disclosure.

Referring to FIG. 2 , the radar module 10 according to the present disclosure detects the surroundings of the own vehicle 100 through a certain detection area SR, and targets 210 and 220 present around the own vehicle 100, For example, it can detect other vehicles. Here, the radar module 10 may be mounted in front of the vehicle 100 as shown in FIG. 2, but is not limited thereto.

When the radar signal transmitted from the radar module 10 reaches the targets 210 and 220 within the detection area SR and the signals reflected from the targets 210 and 220 are received by the radar module 10 again, the targets 210 and 220 Radar point data (Pa, Pb), which is a point where a signal is reflected at , are generated.

The radar module 10 may group radar point data included in a preset gate size and generate the grouped radar point data as one track. For example, the radar module 10 groups radar point data (Pa) corresponding to the target 210 located in the left lane of the front to generate Track # 1, and the target located in the own lane of the front ( 220), Track#2 may be generated by grouping the radar point data Pb.

The radar module 10 may detect track speed information based on the speed information of the radar point data Pa and Pb, and determine whether the track is a stopped track based on the track speed information. Illustratively, when the host vehicle 100 is in a stopped state and the speed of the track is 0, it may be determined that the track is stopped. As another example, if the own vehicle 100 is moving and the speed of the track is the same as the moving speed of the own vehicle 100 and only the sign is opposite, it can be determined as a stopped track.

The radar module 10 may determine whether the track is located in at least one of the own lane and left and right side lanes in front of the own vehicle based on the location information of the track. A track located in at least one of the own lane and both side lanes in front of the own vehicle may be defined as a track requiring lateral resolution. Alternatively, a track in a stationary state and located in at least one of the own lane and both side lanes in front of the own vehicle may be defined as a track requiring lateral resolution.

3 is a diagram illustrating an embodiment of generating learning data according to the present disclosure.

Referring to FIG. 3, the radar module 10 scans the surroundings at a predetermined period, and extracts characteristic information (Ladar Point Feature, LPF) of radar point data within a region of interest (ROI) including a track requiring lateral resolution for each scan and can be provided to the preprocessing module 20.

The characteristic information (LPF) of the radar point data may include distance information and azimuth information of each of the radar point data, and the distance information and azimuth information of the radar point data within the region of interest (ROI) in each scan may be obtained at a corresponding scan time point. It can be viewed as a distribution of radar point data within the region of interest (ROI). The characteristic information (LPF) of radar point data may be defined as a radar point data distribution.

The driving information sensing module (not shown) may be synchronized with the scanning operation of the radar module 10 to sense the driving information of the own vehicle at every scan and provide the same to the preprocessing module 20 .

The pre-processing module 20 receives the characteristic information (LPF) of the radar point data from the radar module 10 and receives the driving information (Vehicle Info, VI) of the host vehicle from the driving information sensing module, and receives the preset number of times (n When the track required for lateral resolution is maintained in the number of scans), the characteristic information (LPF) of n radar point data acquired in n number of scans and the n vehicle driving information (VI) obtained at the time point of n number of scans Time-series learning data (Pcd ₁ , Pcd ₂ ,..., Pcd _n ) may be generated by pre-processing.

Pcd _i is learning data corresponding to the i-th scan, and includes the characteristic information (LPF) of the radar point data obtained in the i-th scan and the own vehicle driving information (VI) obtained from the driving information sensing module at the time of the i-th scan. can do.

The pre-processing may include a process of vectorizing characteristic information (LPF) of radar point data and own vehicle driving information (VI) for each scan. Pcd _i may be expressed as a vector including characteristic information (LPF) of radar point data obtained in the i-th scan and host vehicle driving information (VI) obtained from the driving information sensing module at the time of the i-th scan as components.

In addition, the preprocessing may further include a normalization process. If the scale of the data is transformed according to a certain rule through a normalization process, the deep neural network (30 in FIG. 1) can help to make the data easier to use.

4 is a diagram illustrating seven lane occupancy situations according to the present disclosure.

Referring to FIG. 4 , in class #1 to class #7, when the own vehicle 100 is located in the middle lane among three lanes, the front own lane and left and right side lanes are occupied by other vehicles 200 Indicates the situation on a case-by-case basis.

In the present disclosure, lane occupancy situations are classified into seven types of class #1 to class #7. Hereinafter, class #1 to class #7 will be defined as lane occupancy situation classes.

Class #1 represents a situation in which both the own lane and both side lanes in front are occupied by another vehicle 200, and in class #2, the own lane and left lane in front are occupied by other vehicles 200 and the right lane indicates an empty situation. Class #3 represents a situation where the own lane and the right lane in front are occupied by other vehicles 200 and the left lane is empty, and class #4 indicates that the left and right lanes in front are occupied by other vehicles 200 and the own lane represents an empty situation.

And, in class #5, class #6, and class #7, only the front left lane is occupied by another vehicle 200, only the front own lane is occupied by another vehicle 200, and the front right lane is occupied by another vehicle 200. This indicates a situation in which only the lane is occupied by another vehicle 200 .

5 is a diagram illustrating an example of a lane occupancy situation classification model according to the present disclosure.

Referring to FIG. 5 , the lane occupancy situation classification model uses time-series learning data (Pcd ₁ , Pcd ₂ , ..., Pcd _n ) provided from the preprocessing module 20 to provide characteristic information of radar point data and driving information of the own vehicle. A bidirectional LSTM (Long Short-Term Memory) module 310 that learns characteristics between the lane occupancy situation classes and the characteristic values output from the bidirectional LSTM module 310 are fused to derive characteristic values for each lane occupancy situation class and a softmax layer 340 that generates a probability value for each lane occupancy situation class by using a characteristic value for each lane occupancy situation class. In addition, the deep neural network 30 may further include a concatenate & flatten layer 320 between the bidirectional LSTM module 310 and the fully connected layer 330 .

Bidirectional LSTM module 310 may include n forward LSTM cells (F1 through Fn) and n backward LSTM cells (B1 through Bn).

The n forward LSTM cells (F1 to Fn) each receive as inputs n pieces of training data (Pcd ₁ , Pcd ₂ ,..., Pcd _n ) corresponding to n scans, and are serially connected to each other in chronological order (forward direction). The n backward LSTM cells (B1 to Bn) receive as inputs n learning data (Pcd ₁ , Pcd ₂ ,..., Pcd _n ) corresponding to n scans, and are serially connected to each other in the reverse order (reverse direction) of time. . A forward LSTM cell and a backward LSTM cell that take the learning data (Pcd _i ) corresponding to the i-th scan as an input value are paired, and a feature value of the learning data (Pcd _i ) corresponding to the i-th scan is extracted.

The first forward LSTM cell (F1) among n forward LSTM cells (F1 to Fn) receives Pcd ₁ , which is the first data among time series learning data (Pcd ₁ , Pcd ₂ ,…, Pcd _n ) as an input value, and inputs ( Passing or discarding data is performed by input, forget, and output gates, and based on this, a hidden value is finally created and stored in memory. The data input afterwards is the same as passing through the three types of gates, but the next hidden value is determined by combining with the hidden value generated by the previous input value.

Among the n backward LSTM cells (B1 to Bn), the last backward LSTM cell (Bn) receives Pcd ₁ , which is the first data among time series learning data (Pcd ₁ , Pcd ₂ ,…, Pcd _n ), as an input value, and performs input, forget, Passing or discarding data is performed by the output gate, and based on this, a hidden value is finally created and stored in memory. The data input afterwards is the same as passing through the three types of gates, but the next hidden value is determined by combining with the hidden value generated by the previous input value.

Each of the forward LSTM cells (F1 to Fn) and the backward LSTM cells (B1 to Bn) use Pcd _t at time step t and information h _t-1 from the previous time to obtain a later hidden value h _t as shown in Equation 1 below. yields

Here, σ means a sigmoid function, and W and b indicate weight and bias values corresponding to each gate in the LSTM cell.

For reference, the existing simple recurrent neural network model may have a problem of not remembering past data if the sequence length of the input value is long, such as a gradient loss problem, so a long input value corresponding to n scans Not suitable for predicting existing lane occupancy situations.

In this disclosure, an LSTM cell is utilized. As discussed above, an LSTM cell is composed of input, forget, and output gates, through which the given input data is passed or discarded. The gradient vanishing problem of the recurrent neural network model can be solved by performing operations such as discarding more recent data and memorizing more past data, if necessary, through the forgetting gate.

Looking at the learning sequence of the bidirectional LSTM module 310, first, n learning data (Pcd ₁ , Pcd ₂ , ..., Pcd _n ) are sequentially input to n forward LSTM cells (F1 to Fn) in chronological order, Forward LSTM cells (F1 to Fn) are learned. Then, n pieces of learning data (Pcd ₁ , Pcd ₂ , ..., Pcd _n ) are sequentially input to n backward LSTM cells (B1 to Bn) in reverse order of time, and the backward LSTM cells (B1 to Bn) are learned. .

n feature values output from n forward LSTM cells (F1 to Fn) and n feature values output from n backward LSTM cells (B1 to Bn) pass through the concatenation and flattening layer 320 to obtain 2n one-dimensional vectors It is converted into a form and enters the input of the fully connected layer 330.

The fully connected layer 330 derives characteristic values for each of the 7 lane occupancy situation classes (class #1 to class #7) by performing an operation of fusing 2n characteristic values output from the concatenation and flattening layer 320. The fully connected layer 330 includes 2n input nodes respectively corresponding to 2n characteristic values and 7 output nodes respectively corresponding to 7 lane occupancy situation classes (class #1 to class #7).

The softmax layer 340 calculates probability values for each of the seven lane occupancy situation classes (class #1 to class #7) using the characteristic values output from the fully connected layer 330, and among the seven lane occupancy situation classes The class with the highest probability value is classified as a lane occupancy situation. The softmax layer 340 may calculate an error value for each class based on the probability value. In the learning process, weights and bias values of the bidirectional LSTM module 310 and the fully connected layer 330 are adjusted so that the error value is small.

6 is a flowchart illustrating a method for classifying lane occupancy situations according to the present disclosure, and FIG. 7 is a flowchart illustrating a process of obtaining characteristic information of radar point data.

Referring to FIG. 6 , the lane occupancy situation classification method according to the present disclosure includes characteristic information of radar point data in an area of interest including a stop track located in at least one of the own lane in front of the own vehicle and both side lanes, and the own vehicle Obtaining driving information of (S601); generating learning data by pre-processing the characteristic information of the radar point data and the driving information of the own vehicle (S602); Classifying lane occupancy situations using the learned deep neural network (S603); may be included.

Referring to FIG. 7 , a process of obtaining characteristic information of radar point data in more detail is as follows.

First, when a radar signal is transmitted from the radar module and the radar signal is reflected from the target and received by the radar module, the radar module generates radar point data corresponding to the reflection point of the signal and collects the radar point data included in the preset gate size. A track is created by grouping (S701).

Then, based on the speed information of the track, it is determined whether the track is in a stopped state (S702).

As a result of the judgment in step S702, if it is determined that the track is not in a stopped state, the process returns to step S701, and if it is determined that the track is in a stopped state, the process proceeds to step S703.

In step S703, it is determined whether the track is located in at least one of the own lane and both side lanes in front of the host vehicle based on the track location information.

As a result of the determination in step S703, if it is determined that the track is not located in at least one of the front own lane and both side lanes, the process returns to step S701, and the track is located in at least one of the own lane and both side lanes. If it is determined to be true, characteristic information of radar point data within the region of interest including the corresponding track is extracted.

The embodiment of FIG. 7 corresponds to a case of classifying a lane occupation situation by a target (vehicle) in a stationary state, and includes a step of determining whether a track is in a stationary state. However, this is just one example, and the present disclosure is also applicable to a moving target. In this case, the step of determining whether the track is in a stopped state can be omitted.

Referring to FIG. 6 again, after acquiring the characteristic information of the radar point data and the driving information of the own vehicle, the characteristic information of the radar point data and the driving information of the own vehicle are preprocessed to generate learning data (S602).

Pre-processing is performed when a track is maintained in a preset number of scans (n times). The preprocessing includes a process of vectorizing characteristic information of radar point data and vehicle driving information for each scan. In addition, the preprocessing may further include a normalization process. Through preprocessing, n time-series learning data (Pcd ₁ , Pcd ₂ , ..., Pcd _n in FIG. 5 ) corresponding to n scans are generated.

A lane-occupying situation classification model is built by training a deep neural network to classify lane-occupying situations ahead using the learning data.

Then, when training data is input to the trained deep neural network, the calculated feature value is obtained by passing through the bidirectional LSTM module (310 in FIG. Probability values corresponding to each of the seven lane occupancy situation classes (class #1 to class #7) are calculated through (330 in FIG. 5) and the softmax layer (340 in FIG. 5), and among the seven lane occupancy situation classes The class with the highest probability value is classified as a lane occupancy situation (S603).

8 is a diagram schematically illustrating a lane occupancy situation classification process using a deep neural network according to the present disclosure.

Referring to FIG. 8 , input values of the deep neural network 30 according to the present disclosure include characteristic information (distance information, location information) of radar point data and driving information of the own vehicle, and the deep neural network 30 includes the radar point data. The lane occupancy situation is classified as one of seven lane occupancy situation classes based on the distribution of data characteristic information and driving information of the host vehicle, and the result is output. FIG. 8 illustratively illustrates a case in which the lane occupancy situation classification result is output as that both the own lane and both side lanes in front are occupied by other vehicles.

Hereinafter, the reason for classifying lane occupancy situations based on the distribution of characteristic information of radar point data and driving information of the own vehicle in the present disclosure will be described.

In order to increase the reliability of smart cruise technology, automatic driving technology, and automatic emergency stop technology, accurate information about targets (vehicles) in front is required. As one of the known target identification methods, a method of estimating track characteristic information (moving direction and speed) using a radar module and identifying a forward target by clustering tracks estimated to have the same characteristic information is known. . However, when there is no difference in speed between the vehicles, such as when the vehicles in front are stopped side by side or are driving side by side, it is difficult to distinguish between vehicles and thus accurate information cannot be provided. In addition, even when the estimated track characteristic information is inaccurate or BEV (Bird Eye View) does not appear due to hardware limitations of the radar module, limitations of radar signal characteristics, etc., accurate information about the vehicle in front cannot be provided.

As described above, according to the present disclosure, a deep neural network is trained to classify a lane occupancy situation based on the distribution of driving information of the own vehicle and characteristic information of radar point data, and the lane occupancy situation is classified using the learned deep neural network. Therefore, it is possible to provide accurate information about vehicles located side by side. In addition, since the track characteristic information is not used for vehicle recognition, it is possible to solve the recognition error problem due to the hardware limitation of the radar module and the limitation of the radar signal characteristic.

The above description is merely illustrative of the technical idea of the present disclosure, and various modifications and variations can be made to those skilled in the art without departing from the essential characteristics of the technical idea. In addition, since the present embodiments are not intended to limit the technical idea of the present disclosure, but to explain, the scope of the present technical idea is not limited by these embodiments. The scope of protection of the present disclosure should be construed by the following claims, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of rights of the present disclosure.

10: radar module
20: preprocessing module
30: deep neural network

Claims (9)

Generates radar point data for surrounding targets, generates a track based on the radar point data, and generates characteristic information of the radar point data in an area of interest including a track located in at least one of the own lane and both side lanes in front of the own vehicle. Radar module provided;
a pre-processing module generating learning data by pre-processing characteristic information of the radar point data and own vehicle driving information; and
A lane occupancy situation classification model is built by learning to classify the occupancy situation of the own lane and both side lanes in front of the own vehicle into a plurality of lane occupancy situation classes using the learning data, and the lane occupancy situation classification model is used deep neural networks that classify occupancy situations;
Lane occupancy situation classifying device comprising a. According to claim 1,
Characteristic information of the radar point data includes distance information and azimuth information. According to claim 1,
The driving information of the own vehicle includes at least one of steering torque, steering angle, steering motor information, vehicle speed, vehicle motion information, and vehicle attitude information. According to claim 1,
The radar module scans surrounding targets at a predetermined period to generate radar point data;
When the track is maintained while the radar module performs preset n scans, the preprocessing module converts characteristic information of n radar point data obtained in n scans and vehicle driving information obtained at n scan time points. An apparatus for classifying a lane occupancy situation, characterized in that it is configured to preprocess. According to claim 1,
The lane occupancy situation classification model is
a bidirectional Long Short-Term Memory (LSTM) module for learning characteristics between the characteristic information of the radar point data and the own vehicle driving information and a plurality of lane occupancy situation classes;
A fully connected layer for deriving a characteristic value for each lane occupancy situation class by fusing the characteristic values output from the bidirectional LSTM module; and
a softmax layer for generating a probability value for each of the plurality of lane occupancy situation classes by using the characteristic value for each lane occupancy situation class and classifying a lane occupancy situation class having the highest probability value as a lane occupancy situation;
Lane occupancy situation classifying device comprising a. obtaining characteristic information of radar point data in an area of interest including a track located in at least one of an own lane and both side lanes in front of the own vehicle, and driving information of the own vehicle;
generating learning data by preprocessing characteristic information of the radar point data and driving information of the host vehicle;
Constructing a lane occupancy situation classification model by inputting the learning data into a deep neural network and learning to classify occupancy situations of the own lane and both side lanes in front by a plurality of lane occupancy situation classes; and
and classifying lane occupancy situations using the lane occupancy situation classification model. According to claim 6,
The lane occupancy situation classification method, characterized in that the characteristic information of the radar point data includes distance information and azimuth information. According to claim 6,
In the preprocessing step,
When the track is maintained while a radar module generating radar point data by scanning surrounding targets performs preset n scans, characteristic information of n radar point data obtained in n scans and n scan time points An apparatus for classifying lane occupancy situations, characterized in that for pre-processing vehicle driving information obtained from According to claim 6,
The process of obtaining characteristic information of the radar point data,
generating radar point data for nearby targets using a radar module and generating a track based on the radar point data;
determining whether the track is located in the front own lane or/and both side lanes; and
obtaining distance information and azimuth information of radar point data in an area of interest including tracks located in front own lanes and/or both side lanes;
Lane occupancy situation classification method comprising a.

Images