KR101191151B1 - Apparatus and method for terrain-type classification - Google Patents

Abstract

The present invention provides a terrain classification apparatus and method for selecting any one of a plurality of terrain classification parameters by using image information and environment information, and classifying terrain for the corresponding image information based on the selected terrain classification parameter. It is about. To this end, the terrain classification apparatus according to the present invention includes an environment sensor unit for sensing environment information; And a plurality of pre-stored based on the type of the image information, the sensed environment information, the method setting method of the preprocessing of the preset image information, the method setting method of the preset feature extraction process, and the information about the predetermined classifier. And an image processor for selecting any one of the terrain classification parameters.

Description

Terrain classification device and its method {APPARATUS AND METHOD FOR TERRAIN-TYPE CLASSIFICATION}

The present invention relates to a topographic classification apparatus and a method thereof, and more particularly, to a topographic classification apparatus and a method that is robust to environmental changes.

Research and development for autonomous driving of unmanned vehicles is actively progressing. Here, the autonomous driving means that the driverless vehicle recognizes and determines the surrounding situation by itself using a computer (or controller) that acts as a human brain and sensors that act as eyes without human intervention. In order to efficiently perform such autonomous driving, it is necessary to properly classify the terrain of the moving route and control a vehicle suitable for the classified terrain. In particular, more accurate terrain classification is required in complex natural environments, such as unpaved roads and fields, rather than limited pavements. Accordingly, there is a method for classifying the field terrain based on the input image. However, such methods cannot perform terrain classification only in a limited environment because it does not consider the effects of the seasonal change of the terrain and the change of illumination in the field. Or relearn to the class for terrain classification immediately before the driving test.

An object of the present invention is to obtain terrain image information and environment information in front of an unmanned vehicle using an image sensor (including a camera, etc.) and an environmental sensor, and based on the terrain classification parameter according to the obtained environmental information in real time based on the image. A terrain classification device for performing terrain classification and a method thereof are provided.

It is another object of the present invention to provide a terrain classification apparatus and method for discretizing and quantifying environmental information obtained by an environmental sensor.

It is still another object of the present invention to provide a terrain classification apparatus and method for providing a parameter management method used for terrain classification using environmental information obtained by an environmental sensor.

It is still another object of the present invention to select a most similar parameter based on a preset priority when selecting a terrain classification parameter according to environmental information acquired by an environmental sensor, when there is no parameter suitable for a previously learned environment. The present invention provides a terrain classification apparatus and a method thereof.

According to an aspect of the present invention, there is provided a terrain classification apparatus, comprising: an environment sensor unit configured to sense environment information; And a plurality of pre-stored based on the type of the image information, the sensed environment information, the method setting method of the preprocessing of the preset image information, the method setting method of the preset feature extraction process, and the information about the predetermined classifier. It may include an image processor for selecting any one of the terrain classification parameters.

As an example related to the present invention, the image processing unit classifies a terrain based on the image based on the selected terrain classification parameter, removes a misclassification in a result image in which the terrain is classified, and the misclassification is performed. The removed result information can be transmitted to the integrated control device of the unmanned vehicle.

As an example related to the present invention, the environment information may include wall information and time information detected through a GPS sensor, illuminance information detected through an illuminance sensor, and humidity information detected through a humidity sensor.

As an example related to the present invention, the method setting information for performing the preprocessing of the image information may include setting information about a color space method to be used for color space conversion of the image information and information about the color space converted image information. It may include information on whether or not to perform the normalization process.

As an example related to the present invention, the method setting method setting information of the feature extraction process may include information on a feature vector extraction method to be used for feature vector extraction for the preprocessed image information and information on whether to perform a linear discriminant analysis method. It may include.

As an example related to the present disclosure, the image processing unit may include a type of the image information, environment information, method of performing preprocessing of the predetermined image information, and the predetermined feature extraction process among the plurality of terrain classification parameters. When there is no setting method information and the terrain classification parameter corresponding to the information on the predetermined classifier, based on the predetermined priority of the month information, time information, roughness information, and humidity information included in the environment information Selects first environmental information corresponding to next month information, time information, illuminance information, and humidity information; and pre-processes the type of the image information, the selected first environment information, and the image information among the plurality of terrain classification parameters. Process method setting information of the process, method method setting information of the feature extraction process, and information on the classifier You can select the corresponding terrain classification parameters.

The terrain classification method according to the present invention for achieving the above object, the type of the image information, the environmental information detected by the environmental sensor, the method of performing the pre-processing process of the predetermined image information setting information, the method of performing a predetermined feature extraction process Selecting one of the terrain classification parameters from among the plurality of previously stored terrain classification parameters based on the setting information and the information on the preset classifier; Classifying terrain for the image information based on the selected terrain classification parameter; And eliminating misclassification in the result image of which the terrain is classified.

As an example related to the present invention, the method may further include transmitting the result information from which the misclassification is removed to the integrated control apparatus of the unmanned vehicle.

As an example related to the present invention, the environment information may include wall information and time information detected through a GPS sensor, illuminance information detected through an illuminance sensor, and humidity information detected through a humidity sensor.

As an example related to the present invention, the method setting information for performing the preprocessing of the image information may include setting information about a color space method to be used for color space conversion of the image information and information about the color space converted image information. It may include information on whether or not to perform the normalization process.

As an example related to the present invention, the method setting method setting information of the feature extraction process may include information on a feature vector extraction method to be used for feature vector extraction for the preprocessed image information and information on whether to perform a linear discriminant analysis method. It may include.

As an example related to the present invention, the type of the image information, the environment information, the method of performing the preprocessing of the preset image information, the setting method of the predetermined feature extraction process, and the method of performing the predetermined feature extraction among the plurality of terrain classification parameters. And when there is no terrain classification parameter corresponding to the information on the preset classifier, month information of next order based on a preset priority of month information, time information, illuminance information, and humidity information included in the environment information, Selecting first environment information corresponding to time information, illuminance information, and humidity information; And a type of the image information, the selected first environment information, method setting information for performing the preprocessing of the image information, method setting information for performing the feature extraction, and information on the classifier among the plurality of terrain classification parameters. The method may further include selecting a corresponding terrain classification parameter.

The terrain classification apparatus and method, which are robust against environmental changes, according to an embodiment of the present invention, may acquire image information and environmental information in front of an unmanned vehicle by using an image sensor (including a camera) and an environmental sensor. By performing image-based terrain classification in real time based on the terrain classification parameter according to the information, it provides a terrain classification scheme that is robust to environmental changes and provides real-time terrain classification results for a certain area in front of the autonomous vehicle. Can improve.

In addition, the terrain classification apparatus and method that are robust to environmental changes according to an embodiment of the present invention, by discretizing and quantifying the environmental information obtained by the environmental sensor to reduce the design cost for the terrain classification, Accurate parameter setting may be possible.

In addition, the terrain classification apparatus and the method according to an embodiment of the present invention, by providing an efficient parameter management method using the environmental information obtained by the environmental sensor, it is possible to efficiently utilize a large amount of parameters.

In addition, the terrain classification apparatus and method that are robust to environmental changes according to an embodiment of the present invention, when selecting the terrain classification parameters according to the environmental information obtained by the environmental sensor, when there is no parameter suitable for the previously learned environment, By providing a method of selecting the most similar parameter based on the set priority, it is possible to increase the accuracy of the terrain classification result.

1 is a block diagram showing the configuration of a terrain classification apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating a terrain classification method according to an embodiment of the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings, and the same or corresponding components will be denoted by the same reference numerals regardless of the reference numerals and redundant description thereof will be omitted.

1 is a block diagram showing a configuration of a terrain classification apparatus according to an embodiment of the present invention. As shown in the drawing, the terrain classification apparatus 100 includes an image input unit 110, an environmental sensor unit 120, and a storage unit. 130 and an image processor 140. Not all components of the terrain classification apparatus 100 illustrated in FIG. 1 are essential components, and the terrain classification apparatus 100 may be implemented by more components than those illustrated in FIG. The terrain classification apparatus 100 may be implemented by fewer components.

The image input unit 110 may include an image sensor module (image sensor or camera). In this case, the image sensor module may include a charge-coupled device (CCD) module, an infrared (IR) module, and the like.

The image sensor module transmits image information such as a still image or a moving image obtained by the image sensor module to the image processor 140.

The environmental sensor unit 120 includes an environmental sensor such as a GPS sensor, an illumination sensor, and a humidity sensor.

The GPS sensor detects month information, time information, and the like.

The illuminance sensor detects illuminance information around the terrain classification apparatus 100.

The humidity sensor detects humidity information around the terrain classification apparatus 100.

The storage unit 130 stores data and programs required for the terrain classification apparatus 100 to operate.

In addition, the storage unit 130 stores a parameter database (or classifier parameter database) to be used for the control of the image processor 140. The parameter database may include accumulated image information, environment information, method of performing preprocessing of image information (or information related to preprocessing), method of extracting feature extraction method (or related to feature extraction). Information) and parameters (or terrain classification parameters) generated through any learning method based on the classifier information used and the like.

The image processor 140 controls the overall operation of the terrain classification apparatus 100.

In addition, the image processor 140 processes image information received through the image sensor module included in the image input unit 110 and controls to display the processed image information on a display unit (not shown).

In addition, the image processor 140 performs a preprocessing process on image information received through the image sensor module included in the image input unit 110. That is, the image processor 140 performs color space transformation on the image information or normalizes the average and variance values of a preset image.

For example, the image processor 140 may select one of an RGB color space, a Hue-Saturation-Intensity (HSI) color space, a LAB color space, and a gray color space with respect to the image information. Color space processing is performed.

In addition, the image processor 140 extracts a wavelet feature or a color feature for each block by blocking the image based on the image on which the preprocessing process is performed.

For example, the image processor 140 may perform a wavelet mean (WM), a wavelet energy (WE), a wavelet mean (WM), and a wavelet standard deviation (Wavelet) on the image on which the preprocessing process is performed. Wavelet feature or color feature is extracted using any one of Standard Deviation (WS)), the mean of color and the standard deviation of color, and linear discriminant analysis (LDA).

In addition, the image processor 140 may include the GPS sensor, the illuminance sensor, and the month information, time information, illuminance information, humidity information, etc. respectively detected by the GPS sensor, the illuminance sensor, and the humidity sensor included in the environmental sensor 120. Based on this, as shown in Table 1 below, the respective values are discretized and quantified according to the characteristics of the sensed information.

In addition, the image processing unit 140, as described in [Table 1], for the month information and time information detected by the GPS sensor, March to May is "spring", June to 8 Months are classified as "summer", September through November as "autumn," and December through February as "winter."

In addition, the image processor 140 may set the time zones of morning, afternoon, and night for each season. That is, the image processor 140 sets the morning and morning to "06 to 12 o'clock", the afternoon to "12 to 19 o'clock, and the night to" 19 to 06 o'clock, " The summer sets the morning to "05 to 12", the afternoon to "12 to 20", the night to "20 to 05", and the winter to the "07 to 12" Afternoon can be set to "12 to 18:00" and night to "18:00 to 07:00", respectively.

In addition, the image processor 140 may calculate the illumination information (or illuminance value) detected by the illuminance sensor from 0 to 30,000 Lux as “Low” based on “Lux”, which is a unit of illuminance, and 30,000. Lux to 70,000 Lux can be set to "Middle" and 70,000 Lux or more can be set to "High".

In addition, the image processor 140 is 60% or less of the humidity information (or humidity value) detected by the humidity sensor based on "%", which is a unit of humidity, 60%. To 75% can be set to "Middle", 75% to 85% can be set to "High", and 85% or more can be set to "Very High."

In addition, the storage unit 130 may include image types, environment information, preprocessing characteristics (including color space processing methods, normalization use, etc.), characteristics of feature extraction (including feature vector extraction methods, linear discriminant analysis methods, etc.). And, based on the classifier method used and the like, as described in [Table 2], database information on the parameter to which the parameter naming rule method representing each value is applied according to the characteristics of each information (parameter database). Or, it is stored by classifier parameter database) and provides the corresponding parameter when requested by the image processor.

That is, the image processor 140 names the first letter of the parameter according to the type of the image. For example, when the type of the image is a CCD, the image processor 140 names (or defines) the first letter of the parameter as “C”. When the type of the image is IR, the image processor 140 may name the first letter of the parameter. Name it "I".

In addition, the image processor 140 names the second to fourth letters of the parameter according to the environment information. For example, the image processor 140 may name the morning of spring "0", the afternoon of "1", the night of "2" according to the GPS information of the environment information, and the summer morning of " 3 ", afternoon" 4 ", nighttime" 5 ", autumn morning" 6 ", afternoon" 7 ", nighttime" 8 ", winter morning" 9 ", afternoon is" A ", night is named" B ". In addition, the image processor 140 may designate Low as "L", Middle as "M", and High as "H" according to the illumination value of the environment information. In addition, the image processor 140 may have a low value of "L", a middle value of "M", a high value of "H", and a very high value according to the humidity value of the environmental information. (Very High) is named "U".

In addition, the image processor 140 names the fifth to sixth letters of the parameter according to the color space processing method and normalization of the preprocessing process. For example, the image processor 140 may be “R” for an RGB color space, “H” for an HSI color space, or “L” for an LAB color space based on the color space processing process of the preprocessing process. "," It is named "G" for the gray color space. In addition, the image processor 140 may be named “0” when normalization is not used during the preprocessing process, or “1” when normalization is used.

In addition, the image processor 140 names the seventh to eighth letters of the parameter according to the type of the feature vector used for feature extraction and the use of LDA for the image on which the preprocessing process is performed. For example, when the wavelet mean WM and wavelet energy WE are used as the feature vector, the image processor 140 is "0", and when the wavelet mean WM and wavelet standard deviation WS are used. 1 ", where" 2 "is used when the mean of the color and the standard deviation of the color are used. In addition, the image processing unit 140 names "0" when the linear discriminant analysis method (LDA) is not used and "1" when it is used.

In addition, the image processing unit 140, for the extension name of the parameter is ".mlp" when using a neural network classifier, ".svm" when using a support vector machine (SVM), Gaussian mixture model If used, name it ".gmm".

As described above, the image processor 140 may include an 8-character file name and 3 based on the type of image, environment information, information on the preprocessing process, information on the feature extraction process, and the classifier method used. A terrain classification parameter consisting of a file name of a letter may be generated, and the generated terrain classification parameter may be loaded from the storage unit 130.

In addition to the parameter naming rule method described in the above embodiment, the parameter is named by various combinations of image types, environmental information, information on whether or not to perform a preprocessing process, information on whether or not to perform a feature extraction process, and extension names. can do.

In addition, the image processing unit 140 sets priorities for each state based on seasons and times, as shown in Table 3 above.

For example, the image processor 140 divides the parameters into "0" to "B" based on the time input from the GPS sensor.

In addition, when there is no parameter (spring and morning) corresponding to the state "0" (spring and morning), the image processing unit 140 may have a parameter corresponding to "1" corresponding to the next priority (spring). , Afternoon). In addition, when there is no parameter corresponding to “1”, the image processor 140 searches for a parameter corresponding to “6” corresponding to the next priority (autumn and morning). In addition, when there is no parameter corresponding to the “6”, the image processor 140 may select “7” (autumn or afternoon), “3” (summer or morning), or “4” corresponding to the next priority. (Summer, afternoon), "9" (winter, morning), and "A" (winter, afternoon) in order to search for the corresponding parameters, and use the retrieved parameters.

In addition, the image processing unit 140 also includes parameters corresponding to the states for the states "1", "3", "4", "6", "7", "9", and "A". Set each priority.

Also, in the case of the states "0", "1", "3", "4", "6", "7", "9", and "A", the image processing unit 140 may be morning or afternoon. Since it corresponds to the image information captured in the setting, the parameter used for priority search is set considering only the states (including 0, 1, 3, 4, 6, 7, 9, A, etc.) corresponding to the morning or afternoon. do.

In addition, in the case of the states "2", "5", "8", and "B", the image processing unit 140 corresponds to image information photographed at night, and thus, in the case of a parameter used for priority search Set by considering only the states (including 2, 5, 8, B, etc.) corresponding to the night.

As described above, the image processing unit 140 sets, for each state "0" to "B", if there is no parameter corresponding to each state, respectively, priorities of the parameters to be searched in the next order, According to the priority of the set parameter, even when there is no parameter corresponding to any state, the parameter most similar to the season and time detected by the GPS sensor can be used.

In addition, the image processing unit 140 sets priority for each state based on the illuminance value as described in Table 4 above.

For example, the image processor 140 divides the parameters into "L", "M", and "H" states based on the illuminance value detected by the illuminance sensor.

In addition, when there is no parameter corresponding to the state "L", the image processor 140 searches for a parameter corresponding to "M" corresponding to the next priority. In addition, when there is no parameter corresponding to the "M", the image processor 140 searches for a parameter corresponding to "L" corresponding to the next priority.

As described above, the image processing unit 140, for each state "L", "M", and "H", when there is no parameter corresponding to each state, the priority of the parameter to be searched in the next order Are respectively set, and the parameter most similar to the illuminance value detected by the illuminance sensor may be used even when there is no parameter corresponding to any state by the priority of the set parameters.

In addition, the image processing unit 140 sets the priority for each state based on the humidity value, as shown in Table 5 above.

For example, the image processor 140 divides the parameters into "L", "M", "H", and "U" states based on the humidity value detected by the humidity sensor.

In addition, when there is no parameter corresponding to the state "L", the image processor 140 searches for a parameter corresponding to "M" corresponding to the next priority. In addition, when there is no parameter corresponding to the "M", the image processor 140 searches for a parameter corresponding to "H" corresponding to the next priority. In addition, when there is no parameter corresponding to the "H", the image processor 140 searches for a parameter corresponding to "U" corresponding to the next priority.

As such, the image processor 140 may search for each state "L", "M", "H", and "U" in the following order when there is no parameter corresponding to each state. Priority of each parameter is set, and even if there is no parameter corresponding to an arbitrary state according to the priority of the set parameter, a parameter most similar to the humidity value detected by the humidity sensor can be used.

In addition, the image processor 140 may set priorities for seasons, times, illuminance values, and humidity values. That is, when it is necessary to determine the priority of each of at least two or more of the season, time, illuminance value, and humidity value, the image processor 140 may determine the season, time, illuminance value, and humidity value. After determining the priority of one state on the basis of the predetermined priority, the priority of the other state may be determined.

In addition, the image processor 140 may include a type of an image received by the image input unit 110 and environmental information (month and time information, illuminance information, humidity information, etc.) detected by the environmental sensor unit 120. ), Any one of a plurality of terrain classification parameters previously stored in the storage unit 130 based on information of a predetermined preprocessing process, information of a predetermined feature extraction process, and information about a predetermined (or predetermined) classifier. Select the terrain classification parameter. In this case, when there is no parameter corresponding to the detected environment information, the image processor 140 may select a parameter by using environment information corresponding to a preset priority. Here, the preselected classifier may be a neural network classifier, a support vector machine classifier, a Gaussian mixed model based classifier, or the like.

In addition, the image processor 140 performs a terrain classification process on the image information based on the selected parameter.

In addition, the image processing unit 140 removes misclassification in the result image of which the terrain is classified, and outputs the result information from which the misclassification is removed through the display unit.

In addition, the image processor 140 transmits the result information (or the terrain classification result information) from which the misclassification has been removed to an integrated control device (not shown) of the unmanned vehicle provided with the terrain classification device 100. do. In addition, the integrated control apparatus of the unmanned vehicle generates terrain shape data based on the result information (or the terrain classification result information) from which misclassification transmitted from the image processor 140 is removed, and generates the generated terrain data. The operation of the driverless vehicle can be controlled based on the terrain shape data.

In addition, the terrain classification apparatus 100 may further include a display unit (not shown) for displaying image information processed by a predetermined signal under the control of the image processor 140.

In addition, the display unit displays a terrain classification result (or result information from which misclassification has been removed) according to the image information received through the image input unit 110 under the control of the image processor 140.

As such, the terrain classification apparatus 100 obtains image information and environment information of an unmanned vehicle by using an image sensor (including a camera) and an environment sensor, and based on the terrain classification parameter according to the obtained environment information in real time. Terrain classification can be performed with

In addition, the terrain classification apparatus 100 may discretize and quantify the environmental information acquired by the environmental sensor as a numerical value.

In addition, as described above, the terrain classification apparatus 100 may provide a parameter management method used for terrain classification by using environment information obtained by an environment sensor.

In addition, when the terrain classification device 100 selects the terrain classification parameter according to the environment information acquired by the environment sensor, when there is no parameter suitable for the previously learned environment, the terrain classification apparatus 100 may be based on the preset priority. The terrain classification process can be performed by selecting similar parameters.

Hereinafter, the terrain classification method according to the present invention will be described in detail with reference to FIGS. 1 to 2.

2 is a flowchart illustrating a terrain classification method according to an embodiment of the present invention.

First, the image processor 140 performs an initialization (memory initialization, variable initialization, classifier initial parameter setting, etc.) process of the terrain classification apparatus 100 (S110).

Thereafter, the image input unit 110, through the image sensor module, image information (or image frame) photographing the front of the driverless vehicle equipped with the image input unit 110 (for example, the driving direction of the driverless vehicle). Receive Thereafter, the image processor 140 may process the received image information by a predetermined signal and display it on the display unit (S120).

Thereafter, the image processing unit 140 includes an environment such as a type of an image received through the image input unit 110, wall information and time information detected by the environmental sensor unit 120, illuminance information, and humidity information. Information, information of a predetermined preprocessing process (including information on the method used for the color space conversion and information on whether or not to perform the normalization process), information of a predetermined feature extraction process (when the feature vector is extracted Any one of a plurality of terrain classification parameters previously stored in the storage unit 130 based on information on the method used, information on whether the linear discriminant analysis method is performed), and information on a predetermined classifier. Select the terrain classification parameter. The predetermined classifier may be any one of a neural network classifier, a support vector machine classifier, and a Gaussian mixture model based classifier.

For example, with reference to Table 2, the image processor 140 may be a CCD ("C") type of image input through the image input unit 110, and the image sensor unit 120 may be connected to the environmental sensor unit 120. The visual information detected by the GPS sensor among the environmental information detected is summer-am ("3"), low illuminance value ("L"), humidity value is medium ("M"), and the pretreatment process The method used for color space conversion that is the result of performing the operation is an RGB color space method ("R"), and the normalization process is performed ("1"), and the feature vector that is the result information of the feature extraction process is extracted. The method used was the wavelet mean (WM) and wavelet standard deviation (WS) method ("1"), no linear discriminant analysis was used ("0"), and the neural network classifier (".mlp") of the plurality of classifiers. Is used, corresponding to "C3LMR110.mlp" among a plurality of terrain classification parameters previously stored in the storage unit 130 based on the information. Retrieve la meter, and selects a parameter corresponding to the retrieved "C3LMR110.mlp".

In this case, when there is no parameter corresponding to the environment information, the image processor 140 may determine the environment information based on the environment information ([Table 3] to [Table 5]) corresponding to a predetermined priority. You can select any parameter that is most similar.

For example, in the image processing 140, the image type input through the image input unit 110 is a CCD (“C”) with reference to [Table 2], and the environmental sensor unit 120 is connected to the image sensor 140. The visual information detected by the GPS sensor among the environmental information detected is summer-am ("3"), low illuminance value ("L"), humidity value is medium ("M"), and the pretreatment process The method used for color space conversion that is the result of performing the operation is an RGB color space method ("R"), and the normalization process is performed ("1"), and the feature vector that is the result information of the feature extraction process is extracted. The method used was the wavelet mean (WM) and wavelet standard deviation (WS) method ("1"), no linear discriminant analysis was used ("0"), and the neural network classifier (".mlp") of the plurality of classifiers. In the case of using, search for a parameter corresponding to "C3LMR110.mlp" among a plurality of terrain classification parameters previously stored in the storage unit 130, If there is no parameter corresponding to "C3LMR110.mlp", summer-afternoon ("4") which is the next sequence of summer-am ("3") which is time information detected by the GPS sensor based on the preset priority information. Search for and select the parameter corresponding to "C4LMR110.mlp" corresponding to "). Further, when there is no parameter corresponding to the summer-afternoon ("4") (the parameter corresponding to "C4LMR110.mlp"), the next priority of the summer-afternoon ("4") is based on the above [Table 3]. The parameters corresponding to the ranks may be searched in order to select the corresponding parameters.

In addition, the image processor 140, based on the priority set for the GPS value, illuminance value, and humidity value included in the environment information when selecting a similar parameter according to the priority because there is no parameter corresponding to the initial parameter. It may be configured to select similar parameters in the order of priority set in advance (for example, GPS values, illuminance values, and humidity values).

Thereafter, the image processor 140 performs a preprocessing process on the received image information. In addition, the image processor 140 performs a preprocessing process according to information of a preprocessing process (including information on a method used for color space conversion and information on whether to perform a normalization process). That is, the image processor 140 performs color space conversion on the image information by using any one of an RGB color space, an HSI color space, a LAB color space, and a gray color space. For example, the image processor 140 may perform color space conversion on the image information by using an RGB color space method among a plurality of color space methods.

In addition, the image processor 140 performs a normalization process on the color space-converted image information with a mean and variance value of a preset image according to whether a preset normalization process is performed on the color space-converted image information. Or do not. In this case, the determination on whether to perform the normalization process may include preset information on whether to perform the normalization process, or display information on whether to perform the normalization process on the display unit and normalize the process according to a user input. You can also set whether to perform. For example, the image processor 140 performs a normalization process on the color space-converted image information based on a predetermined value (S140).

Thereafter, the image processor 140 includes information on performance of a predetermined feature extraction process (including information on a method used when the feature vector is extracted and information on whether to perform a linear discriminant analysis (LDA) process). The preprocessed image information is blocked into a plurality of blocks, and a wavelet feature or color feature is extracted for each block of the blocked image information.

That is, the image processor 140 blocks the preprocessed image information into a plurality of blocks, and wavelet mean WM and wavelet energy WE, wavelet mean WM, wavelet standard deviation WS, and color. A feature vector of the blocked image information is extracted for each block by using any one of a mean and a standard deviation of colors. For example, the image processor 140 extracts a feature vector on the blocked image information using a wavelet mean and a wavelet standard deviation method.

Also, the image processor 140 may or may not perform a linear discriminant analysis process on the extracted feature vectors depending on whether the linear discriminant analysis method (LDA) on the extracted feature vectors is performed. In this case, the determination on whether to perform the linear discriminant analysis process may include information on whether or not to perform the linear discriminant analysis process, or the display unit may display information on whether or not to perform the discriminant analysis process. Depending on the input, it may be set whether or not to perform a linear discriminant analysis process (S150).

Thereafter, the image processor 140 classifies the terrain for the input image information based on the selected terrain classification parameter (S160).

Thereafter, the image processor 140 removes misclassification in the resultant image of which the terrain is classified. In addition, the image processor 140 may transmit the result information (or the terrain classification result information) from which the misclassification is removed to the integrated control device of the unmanned vehicle (S170).

It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the scope of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

100: terrain classification apparatus 110: image input unit
120: environmental sensor unit 130: storage unit
140: the image processing unit

Claims (12)

An environmental sensor unit for sensing environmental information; And
A plurality of terrains stored in advance based on the type of the image information, the detected environment information, the method setting information for performing the preprocessing of the preset image information, the method setting information for performing the preset feature extraction, and the information on the predetermined classifier Select one terrain classification parameter from the classification parameters,
Among the plurality of topographic classification parameters, the type of the image information, the environment information, the method setting information for performing the preprocessing of the preset image information, the method setting information for performing the preset feature extraction, and the preset classifier When there is no terrain classification parameter corresponding to the information, the month information, the time information, the illumination information, and the humidity of the next order based on the preset priority of the month information, the time information, the illumination information, and the humidity information included in the environment information Select the first environment information corresponding to the information,
Corresponding to a type of the image information, the selected first environment information, execution method setting information of a preprocessing process of the image information, execution method setting information of the feature extraction process, and information on the classifier among the plurality of terrain classification parameters. And a video processor for selecting a terrain classification parameter. The image processing apparatus of claim 1, wherein the image processor comprises:
Classify the terrain based on the image based on the selected terrain classification parameter, remove the misclassification in the result image where the terrain is classified, and transmit the result information from which the misclassification is removed to the integrated control device of the unmanned vehicle. Terrain classification device, characterized in that. The method of claim 1, wherein the environmental information,
And terrain information detected by the GPS sensor, illumination information detected by the illumination sensor, and humidity information detected by the humidity sensor. The method of claim 1, wherein the setting method setting information of the preprocessing process of the image information comprises:
And setting information on a color space method to be used for color space conversion on the image information, and information on whether or not to perform a normalization process on the color space converted image information. The method of claim 1, wherein the setting method setting information of the feature extraction process is performed.
And a feature vector extraction method to be used for feature vector extraction of the preprocessed image information, and information on whether to perform a linear discriminant analysis method. delete Based on the type of the image information, the environment information detected by the environmental sensor unit, the setting method setting information of the preprocessing process of the predetermined image information, the setting method setting method of the predetermined feature extraction process, and the preset classifier information Selecting one of the terrain classification parameters from the plurality of terrain classification parameters;
Among the plurality of topographic classification parameters, the type of the image information, the environment information, the method setting information for performing the preprocessing of the preset image information, the method setting information for performing the preset feature extraction, and the preset classifier When there is no terrain classification parameter corresponding to the information, the month information, the time information, the illumination information, and the humidity of the next order based on the preset priority of the month information, the time information, the illumination information, and the humidity information included in the environment information Selecting first environment information corresponding to the information;
Corresponding to a type of the image information, the selected first environment information, execution method setting information of a preprocessing process of the image information, execution method setting information of the feature extraction process, and information on the classifier among the plurality of terrain classification parameters. Selecting a terrain classification parameter;
Classifying terrain for the image information based on the selected terrain classification parameter; And
And removing misclassification in the resultant image of which the terrain is classified. The method of claim 7, wherein
And transmitting the result information from which the misclassification has been removed to the integrated control device of the unmanned vehicle. The method of claim 7, wherein the environmental information,
And terrain information detected by the GPS sensor, illumination information detected by the illumination sensor, and humidity information detected by the humidity sensor. The method of claim 7, wherein the setting method setting information of the preprocessing process of the image information comprises:
And setting information on a color space method to be used for color space conversion on the image information, and information on whether or not to perform a normalization process on the color space converted image information. The method of claim 7, wherein the setting method setting information of the feature extraction process,
And information on a feature vector extraction method to be used for feature vector extraction of the preprocessed image information, and information on whether to perform a linear discriminant analysis method. delete

Images