KR20150050134A - Method of enhancing performance for road-environment recognition device based on learning and apparatus for the same - Google Patents

Abstract

Disclosed are a method for enhancing performance of a road-environment recognition device based on learning and an apparatus therefor. According to the present invention, the method for enhancing performance of a road-environment recognition device based on learning comprises steps of: generating matching data by obtaining road data and information acquisition time; extracting features in the whole area and a local area of a road on the basis of the matching data; detecting a driving mode of a present vehicle to be transmitted to a recognition information processing unit; and determining whether there is learning in an environment recognition device on the basis of the features in the whole area and a local area and the driving mode to provide recognition information about a driving environment recognized by the learning environment recognition device to an intelligent vehicle.

Description

TECHNICAL FIELD The present invention relates to a method for improving performance of a learning-based road environment recognizer, and a device for the same. [0002]

The present invention relates to a multi-sensor-based road environment recognizer, and more particularly, to a vehicle environment recognition system that continuously learns a road environment recognizer while a driver is driving a vehicle in a semi-fully self- Technology.

The development of autonomous vehicles and their related technologies are rapidly developing beyond the capabilities of Advanced Driver Assistance Systems (ADAS). An autonomous vehicle is a vehicle that recognizes the surrounding environment without the driver's control, judges the driving situation, and controls the vehicle to drive the vehicle to a given destination. In recent years, such autonomous vehicles have been attracting attention as future personal transportation vehicles because they can reduce traffic accidents, improve transportation efficiency, and have various advantages such as fuel savings and driver convenience. Several categories of autonomous driving technology can be categorized as Driver Assistance technology, Unmanned Vehicle or Driverless Car, or Self-driving, Autonomous Driving technology. Driver assistive technology refers to a technique that warns or assists the driver in either longitudinal or transverse directions. Unmanned automobiles or autonomous navigation technologies can be used in combination. However, unmanned vehicles can be used as unmanned vehicles in areas such as defense where people are not on board, A vehicle traveling from the boarding to the destination can be distinguished as an autonomous vehicle.

Until now, autonomous vehicle technology has been using high-priced laser radar or GPS (Global Positioning System) to recognize the surrounding environment and estimate the position of driver's vehicle. In addition, distributed processing is performed using a plurality of computer clusters for recognition information processing and judgment. For steering, speed, and gear control, it can be controlled through modification such as addition of motor or sensor.

In this environment, in order for the intelligent vehicle to perform semi-autonomous / autonomous driving on the actual road, it is possible to utilize the surround sensor information and the external infrastructure information using the camera, the radar, the ultrasonic wave, and the laser radar installed in the vehicle. The laser radar is composed of a laser transmitter unit, a receiver unit, a signal processing unit and a display unit. The laser radar unit irradiates a strong laser beam having high directivity to a remote object and receives reflected light or scattered light of the object through a receiving telescope. By measuring the distance to the object, detecting and identifying the object, and measuring the velocity of the moving object by using the Doppler effect, the sensor can perform signal processing. In particular, the technology of recognizing road / non-road regions and obstacles based on three-dimensional information using surround sensors is essential for autonomous navigation. Most of the existing technologies related to the extraction of the road region mainly consisted of a method of recognizing a driving lane on the road. However, in order for an intelligent vehicle to be actually used on a public road, it is necessary to have a recognizing function capable of running on an area without a lane on the road, and a high performance recognizer capable of recognizing the environment information in detail in various road environments and situations do. In this regard, in the conventional driving environment recognition methods, there are used methods of constructing a database of driving environments in which an actual vehicle can be operated, and learning a recognizer using a built-in database using a comparison / teaching method. However, if the environment information similar to the new road environment does not exist in the database when the vehicle is actually operated, the performance of the recognizer will be significantly deteriorated.

Therefore, in order to solve these problems, the present invention can optimize the recognizer to the driving environment by repeatedly learning the recognizer while the driver is operating frequently or the driver safely operates the automobile. Based on this, the self- We introduced a technology that has improved recognition performance.

Related prior art is Korean Patent Laid-Open Publication No. 2011-0094954 on a vehicle autonomous navigation system based on driving history information.

It is an object of the present invention to provide a road environment recognizer optimized for autonomous driving of an intelligent automobile by performing a process of continuously learning a road environment recognizer during a driving environment frequently operated by a driver or a driver operating a vehicle safely .

According to another aspect of the present invention, there is provided a method for improving performance of a learning-based road environment recognizer, including: generating registration data by obtaining road data and information acquisition time; Extracting global / regional features of the road based on the matching data; Sensing the current driving mode of the vehicle and delivering it to the learning-based recognition information processor; And determining whether the environment recognizer is learning based on the global / regional characteristics and the driving mode, and providing the intelligent vehicle with recognition information on the travel environment recognized by the learned environment recognizer.

In this case, if the driving mode is the driver mode, the providing step may further include optimizing the environment identifier through the non-mapping / guidance learning method.

In this case, when the driving mode is the autonomous mode, the providing step may provide the intelligent vehicle with recognition information on the driving environment recognized by the environment recognizer with the variable.

In this case, the generating step may acquire the road data and information acquisition time using at least one of a 2D / 3D laser radar and a radar sensor configured by a camera and a multi-array.

At this time, the step of delivering to the learning-based recognition information processing unit may determine that one of the driver mode, semi-autonomous mode, or autonomous mode is the traveling mode and can deliver it.

According to another aspect of the present invention, there is provided an apparatus for enhancing the performance of a learning-based road environment recognizer, including: a surround sensor for acquiring road data and information acquisition time; A sensor processing unit for matching the road data and information acquisition time to generate matching data; A sensor data preprocessing unit for extracting global / regional features of the road based on the matching data; A driving license information processor for sensing the current driving mode of the vehicle and transmitting the driving mode information to the learning based recognition information processor; And a learning-based recognition information processor for determining whether the environment recognizer is learning based on the global / regional feature and the driving mode, and for providing recognition information on the driving environment recognized by the learned environment recognizer to the intelligent vehicle .

At this time, the learning-based recognition information processing unit can optimize the parameters of the environment recognizer through the non-mapping / guidance learning method when the driving mode is the driver mode.

At this time, the learning-based recognition information processing unit can provide the intelligent vehicle with recognition information on the driving environment recognized by the environment recognizer with the variable when the driving mode is the autonomous mode.

At this time, the surround sensor may be composed of at least one of a 2D / 3D laser radar and a radar sensor composed of a camera and a multi-array.

At this time, the driving license information processing unit can determine either the driver mode, semi-autonomous mode, or autonomous mode as the driving mode and deliver it.

According to the present invention, it is possible to provide a road environment recognizer optimized for autonomous driving of an intelligent automobile by performing a process of continuously learning a road environment recognizer during a driving environment frequently operated by a driver or a driver driving a car safely have.

1 is a flowchart illustrating a method for improving performance of a learning-based road environment recognizer according to an exemplary embodiment of the present invention.
2 is a block diagram illustrating an apparatus for improving performance of a learning-based road environment recognizer according to an embodiment of the present invention.
FIG. 3 is a flowchart illustrating a method for improving performance of a learning-based road environment recognizer according to an exemplary embodiment of the present invention.

The present invention will now be described in detail with reference to the accompanying drawings. Hereinafter, a repeated description, a known function that may obscure the gist of the present invention, and a detailed description of the configuration will be omitted. Embodiments of the present invention are provided to more fully describe the present invention to those skilled in the art. Accordingly, the shapes and sizes of the elements in the drawings and the like can be exaggerated for clarity.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a method for improving performance of a learning-based road environment recognizer according to an exemplary embodiment of the present invention.

Referring to FIG. 1, a method for improving performance of a learning-based road environment recognizer according to an exemplary embodiment of the present invention may first acquire road data and information acquisition time (S110). The road data and information acquisition time can be obtained from surround sensors such as a camera, a radar, and a laser radar mounted on a vehicle. The laser radar is composed of a laser transmitter unit, a receiver unit, a signal processing unit and a display unit. The laser radar unit irradiates a strong laser beam having high directivity to a remote object and receives reflected light or scattered light of the object through a receiving telescope. By measuring the distance to the object, detecting and identifying the object, and measuring the velocity of the moving object by using the Doppler effect, the sensor can perform signal processing. Such a surround sensor may include a camera and a 2D / 3D laser radar composed of multiple arrays, a radar sensor, and the like. It is possible to recognize the lane and feature on the road through the surround sensor, and to recognize environmental factors such as facilities around the road and roadside trees located on the lateral sides of the vehicle.

Thereafter, the matching data may be generated by matching the road data and the information acquisition time (S120). It is possible to generate the registration data by performing the temporal / spatial registration process on the road data and the information acquisition time acquired from the surround sensor.

The global / local feature points of the road can be extracted based on the matching data thus generated (S130). The global / local feature points of the extracted roads can be transmitted to the learning-based recognition information processing unit.

Further, the current driving mode of the vehicle may be sensed and transmitted to the learning-based recognition information processing unit (S140). The traveling mode of the vehicle can be determined to be one of the driver mode, the semi-autonomous mode, and the autonomous mode.

In step S150, it is determined whether or not the environment recognizer is learning based on the global / regional characteristics of the road and the driving mode of the vehicle in steps S130 and S140. If the driving mode is the driver mode, the learning process for the currently running road can be executed to store the learning result through the non-mapping / guidance learning method. The stored learning results are stored in a variable database and can be used to optimize variables of the environment recognizer. When the vehicle travels in the autonomous mode, the variables stored in the variable database are applied to the environment recognizer without executing the learning process, so that the road environment can be recognized by the environment recognizer optimized for the road during the current operation.

Through the learning, recognition information of the road environment recognized as an environment recognizer whose parameters are optimized can be provided as an intelligent vehicle (S160). Intelligent vehicles can autonomously run more accurately and safely through the provided recognition information. Most of the existing road area extraction techniques are mainly based on the recognition of driving lanes on the road. However, when the intelligent automobile using such a method actually runs autonomously on a general road, the recognition function is significantly deteriorated in an irregular environment such as a road area without a lane or an unpaved road, I could not do the autonomous inaccurate running. In order to solve such a problem, the present invention can provide a high-performance environment recognizer capable of recognizing and running the surrounding environment information in various road environments and situations through the surround sensor installed in the vehicle.

2 is a block diagram illustrating an apparatus for improving performance of a learning-based road environment recognizer according to an embodiment of the present invention.

2, an apparatus 200 for improving performance of a learning-based road environment recognizer according to an exemplary embodiment of the present invention includes a surround sensor 210 for collecting data, a sensor processing unit 220 for generating matching data, A driving data processor 240 for sensing and transmitting a driving mode of the vehicle, a sensor data preprocessing unit 230 for extracting global / regional characteristics of the environment recognizer, And a learning-based recognition information processing unit 250 provided to the vehicle 260.

The surround sensor 210 can acquire road data and information acquisition time of the road in operation. The surround sensor 210 may include at least one of a camera, a 2D / 3D laser radar and a radar sensor. The laser radar is composed of a laser transmitter unit, a receiver unit, a signal processing unit and a display unit. The laser radar unit irradiates a strong laser beam having high directivity to a remote object and receives reflected light or scattered light of the object through a receiving telescope. By measuring the distance to the object, detecting and identifying the object, and measuring the velocity of the moving object by using the Doppler effect, the sensor can perform signal processing. By using these sensors, it is possible to recognize lanes and features on the road, and environment factors such as facilities around a road and roadside trees located on the lateral sides of the vehicle can be recognized.

The sensor processing unit 220 can generate the matching data by matching the road data and the information acquisition time acquired by the surround sensor 210. [ The matching data generated by performing the temporal / spatial matching process on the road data and information acquisition time can be transmitted to the sensor data preprocessing unit 230.

The sensor data preprocessing unit 230 can extract the global / regional characteristics of the road based on the matching data received from the sensor processing unit 220.

The driving license information processor 240 may sense the current driving mode of the vehicle and may transmit information on the driving mode to the learning based recognition information processing unit 250. [ The current driving mode of the vehicle can be determined to be one of the driver mode, the semi-autonomous mode, and the autonomous mode through the interface for driving the vehicle, and the information on the current driving mode can be transmitted.

The learning-based recognition information processing unit 250 may determine whether the environment recognizer is learning, and provide the recognition information on the environment recognized by the learned environment recognizer to the intelligent vehicle 260. [ If the driving mode transmitted from the driving license information processing unit 240 is the driver mode, variables of the environment recognizer can be optimized through the non-mapping / guidance learning method. If the road environment is frequently operated by the driver or the driver is driving the vehicle safely, it is possible to realize the optimizer more optimally at the time of autonomous driving later by continuously learning the road environment recognizer during the operation. When the driving mode received from the driving license information processing unit 240 is the autonomous mode, the driving environment recognition information recognized by the environmental recognizer with the variable parameters can be provided to the intelligent vehicle 260 through the iterative learning . By improving the recognition performance of the environment recognizer in this way, more accurate and safe autonomous travel of the intelligent vehicle 260 can be achieved during autonomous operation. Most of the existing road area extraction techniques are mainly based on the recognition of driving lanes on the road. However, when the intelligent automobile using such a method actually runs autonomously on a general road, the recognition function is significantly deteriorated in an irregular environment such as a road area without a lane or an unpaved road, I could not do the autonomous inaccurate running. In order to solve such a problem, the present invention can provide a high-performance environment recognizer capable of recognizing and running the surrounding environment information in various road environments and situations through the surround sensor installed in the vehicle.

FIG. 3 is a flowchart illustrating a method for improving performance of a learning-based road environment recognizer according to an exemplary embodiment of the present invention.

Referring to FIG. 3, the detailed process of the performance improving method of the learning-based road environment recognizer according to an embodiment of the present invention can process data acquired through the surround sensor installed in the vehicle (S310). In this process, the road data and information acquisition time can be obtained by using a surround sensor composed of at least one of a 2D / 3D laser radar and a radar sensor composed of a camera and multiple arrays. The laser radar is composed of a laser transmitter unit, a receiver unit, a signal processing unit and a display unit. The laser radar unit irradiates a strong laser beam having high directivity to a remote object and receives reflected light or scattered light of the object through a receiving telescope. By measuring the distance to the object, detecting and identifying the object, and measuring the velocity of the moving object by using the Doppler effect, the sensor can perform signal processing. By analyzing the data acquired through the surround sensor, it is possible to recognize lanes and features on the road, and environmental factors such as the facilities around the roads and roadside trees located on the lateral sides of the vehicle can be recognized. The matching data can be generated by matching the obtained road data and information acquisition time. The matching data can be generated by temporally / spatially matching the road data and the information acquisition time, and the generated matching data can be transmitted to the sensor data preprocessing unit.

The data filtering and the global / local feature points of the road can be extracted based on the received matched data processed in step S310. The learning result generated after learning the environment recognizer in the environment where the minutiae of the extracted roads are applied can be stored in the variable database and can be applied to optimize the environment recognizer in the autonomous driving later.

After extracting the minutiae, it is possible to determine whether the current driving mode of the vehicle is the driver mode through the driving license information processing unit (S325).

If it is determined in step S325 that the current driving mode is the driver mode, the learning process of the environment recognizer may be executed (S330). The environment recognizer can be repeatedly learned through the map / non-biodiagnosis learning method during the road environment where the driver normally travels or while the driver is driving safely.

Thereafter, the learning result on the road to which the feature point is applied can be stored (S340).

The learning result may be stored in the variable database as one variable (S350).

If it is determined in step S325 that the current driving mode is not the driver mode, it can be determined whether the current driving mode is the autonomous mode (S335).

If it is determined in step S335 that the current driving mode is the autonomous mode, a variable for optimizing the environment recognizer may be requested and applied in the variable database of step S350 (S360).

The road environment recognition can be performed using the environment recognizer optimized through the applied variable in step S360. By recognizing the environment recognizer that is optimized by applying the variable according to the learning result, the road environment can be recognized more accurately.

The road environment recognition information recognized by the environment recognizer optimized in step S370 may be output to the vehicle controller and output device of the intelligent vehicle (S380). By outputting the recognized road environment information to the intelligent vehicle through the optimized environment recognizer, it is possible to carry out autonomous running more accurately and safely during autonomous driving. Most of the existing road area extraction techniques are mainly based on the recognition of driving lanes on the road. However, when an intelligent automobile using such a method actually runs autonomously on a general road, the recognition function is significantly deteriorated in an irregular environment such as a road area without a lane or an unpaved road, I could not do the autonomous inaccurate running. In order to solve such a problem, the present invention can provide a high-performance environment recognizer capable of recognizing and running the surrounding environment information in various road environments and situations through the surround sensor installed in the vehicle.

As described above, the method and apparatus for improving the performance of the learning-based road environment recognizer according to the present invention are not limited to the configuration and method of the embodiments described above, All or some of the embodiments may be selectively combined.

200: Performance enhancement device of learning-based road environment recognizer
210: Surround sensor
220:
230: Sensor data preprocessing section
240: running driver information processor
250: learning-based recognition information processor
260: Intelligent vehicle

Claims (1)

Acquiring road data and information acquisition time to generate registration data;
Extracting global / regional features of the road based on the matching data;
Sensing the current driving mode of the vehicle and delivering it to the learning-based recognition information processor; And
Determining whether the environment recognizer is learning based on the global / regional characteristics and the driving mode, and providing recognition information on the travel environment recognized by the learned environment recognizer to the intelligent vehicle A method for improving the performance of a road environment recognizer.

Images