JP4652101B2 - Object recognition device and object recognition method - Google Patents

Description

  The present invention relates to an object recognition apparatus and an object recognition method for recognizing whether an object shown in a photographed image is a desired object.

Non-Patent Document 1 below discloses a technique for recognizing whether or not a conventionally photographed object is a desired object (hereinafter also referred to as a conventional object recognition device). In the conventional object recognition device, a plurality of image processes with different accuracy are performed on the captured image, and an object reflected in the captured image is based on all the images that have been subjected to the respective image processing. It was recognized whether it was a desired object.
Y. LeCun, T. Bengio “Convolutional Networks for Images, Speech, and Time-Series” In MAArbib, editor, The Handbook of Brain Theory and Neural Networks, MIT Press, Cambridge, Massachusetts, 1995, pp. 255-258

  However, in the conventional object recognition apparatus, a plurality of image processing is performed on a captured image and recognition is performed based on all the images that have been subjected to the plurality of image processing. There was a problem that.

  The present invention is intended to solve the above-described problem, and an object recognition apparatus and object recognition capable of recognizing whether an object shown in a photographed image with a reduced calculation load is a desired object. It aims to provide a method.

In order to achieve the above object, according to the present invention, an image data acquisition unit that acquires image data indicating an image, and a level of a level for recognizing a desired object with respect to the acquired image data. Based on information of the image processing means that performs a plurality of different image processing procedures and information of the image data that has been subjected to the plurality of image processing procedures at different levels, a plurality of recognizers each having a characteristic property output And a recognition unit that recognizes whether or not an object reflected in the image is a desired object, according to an output result, wherein the image processing unit recognizes the desired object. In order to make it possible, for the acquired image data, a first filter that sets a predetermined filter in a first filter set of filters each having a characteristic property Setting means; first feature information generating means for generating first feature information indicating characteristics of the image data in which a predetermined filter is set in the first filter set; and recognition of the desired object The second filter set of filters each having a characteristic property with respect to the generated first feature information and / or the first feature information generated in the past. Of the second filter information, the second filter setting means sets a predetermined filter, and the second feature information indicates a feature of the first feature information in which the predetermined filter is set in the second filter set. and a second feature information generating means for generating the image data procedure is performed, the recognition means, part of the image data of one of the image data pre-outs image processing procedure has been performed Based on the selected image data, it is recognized whether or not what is reflected in the image is the desired object, and if it cannot be recognized, the selected image is recognized until it can be recognized. Selecting part of the image data that has been subjected to the image processing procedure other than the data, and continuing recognition based on the previously selected image data and the newly selected image data An object recognition apparatus characterized by the above is provided. With this configuration, it is possible to recognize whether or not an object shown in an image taken with a reduced calculation load is a desired object.

According to the present invention, the image data acquisition means for acquiring image data representing an image, and a plurality of image processing procedures having different levels for recognizing a desired object with respect to the acquired image data. The image processing means for performing the image processing, and the output result output by the plurality of recognizers each having a characteristic property based on the information of the image data subjected to the plurality of image processing procedures having different levels, And a recognition means for recognizing whether or not the object reflected in the image is the desired object, the image processing means obtains the recognition to enable the recognition of the desired object. First filter setting means for setting a filter of a first filter set of filters each having a characteristic property with respect to the image data, and the first filter A first feature information generating means for generating first feature information indicating the characteristics of the image data in which a set filter is set; and the generated feature information to further enable recognition of the desired object. Second filter setting means for setting a filter of a second filter set of filters each having a characteristic property with respect to the first feature information, and the filter of the second filter set being set Second feature information generating means for generating the image data subjected to the image processing procedure, which is second feature information indicating the feature of the first feature information, and the recognition means includes the plurality of recognitions Depending on the output result of some of the recognizers, it is recognized whether or not what is reflected in the image is the desired object. An object recognition device is provided, wherein recognition is performed based on an output result of a part of the plurality of recognizers other than the recognizer used and an output result used in the past. The With this configuration, it is possible to recognize whether or not an object shown in an image taken with a reduced calculation load is a desired object.

According to the present invention, the image data acquisition means for acquiring image data representing an image, and a plurality of image processing procedures having different levels for recognizing a desired object with respect to the acquired image data. The image processing means for performing the image processing, and the output result output by the plurality of recognizers each having a characteristic property based on the information of the image data subjected to the plurality of image processing procedures having different levels, And a recognition means for recognizing whether or not the object reflected in the image is the desired object , the image processing means obtains the recognition to enable the recognition of the desired object. First filter setting means for setting a filter of a first filter set of filters each having a characteristic property with respect to a part of the image data, First feature information generating means for generating first feature information indicating the characteristics of the image data in which a filter of the first filter set is set, and generation to further enable recognition of the desired object Second filter setting means for setting a filter of a second filter set of filters each having a characteristic property with respect to a part of the first characteristic information, and the second filter set A second feature information generating unit configured to generate the image data subjected to the image processing procedure, which is second feature information indicating a feature of the first feature information in which a filter is set, and the recognition unit is pre-outs image processing procedure on the basis of the information of the image data subjected to some, the output result, each output from the plurality of recognizers having characteristic properties, appearing in the image A part of the image data subjected to the plurality of image processing procedures and the plurality of image processing procedures until the recognition is possible. There is provided an object recognition apparatus that performs recognition based on information on a part that has been newly subjected to the plurality of image processing procedures in addition to a part of the applied image data. With this configuration, it is possible to recognize whether or not an object shown in an image taken with a reduced calculation load is a desired object.

According to the present invention, the image data acquisition means for acquiring image data representing an image, and a plurality of image processing procedures having different levels for recognizing a desired object with respect to the acquired image data. The image processing means for performing the image processing, and the output result output by the plurality of recognizers each having a characteristic property based on the information of the image data subjected to the plurality of image processing procedures having different levels, And a recognition means for recognizing whether or not the object reflected in the image is the desired object , the image processing means obtains the recognition to enable the recognition of the desired object. First filter setting means for setting filters of a first filter set of filters each having a characteristic property with respect to the image data obtained, and the first A first feature information generating unit that reduces the image in which a filter of a filter set is set and generates first feature information indicating the feature of the reduced image, and further enables recognition of the desired object Therefore, a second filter setting means for setting a filter of a second filter set of filters each having a characteristic property with respect to the generated first feature information, and the second filter Second feature information generating means for generating the image data subjected to the image processing procedure, which is second feature information indicating the feature of the first feature information in which a set filter is set, recognition means, on the basis of the image data information which the image processing procedure has been performed, each with an output result of a plurality of recognizers are output having the characteristic properties, the image It is recognized whether or not the object is the desired object, and if it cannot be recognized, the reduced image is converted into image data information of images sequentially enlarged by the image processing means until it can be recognized. There is provided an object recognition device characterized by recognition based on the above. With this configuration, it is possible to recognize whether or not an object shown in an image taken with a reduced calculation load is a desired object.

Further, according to the present invention, a step of acquiring image data indicating an image, and a step of performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data And based on the information of the image data that has been subjected to a plurality of image processing procedures at different levels, each of which is reflected in the image by an output result output by a plurality of recognizers each having a characteristic property A step of recognizing whether the object is a desired object or not, wherein the step of performing the image processing procedure is performed in order to enable recognition of the desired object. Setting a predetermined filter among first filter sets of filters each having a characteristic property for image data; and the first filter Generating a first feature information indicating a feature of the image data in which a predetermined filter is set, and the generated first feature to further enable recognition of the desired object. Setting a predetermined filter in a second filter set of filters each having a characteristic property with respect to the information and / or the first feature information generated in the past, and the second filter Generating the image data subjected to the image processing procedure that is the second feature information indicating the feature of the first feature information in which a predetermined filter is set in the set, and the step of recognizing in selects a part of the image data among the image data pre-outs image processing procedure has been performed, copy the image based on the image data selected A part of the image data subjected to the image processing procedure other than the selected image data until it can be recognized if it is not recognized. An object recognition method is provided, in which recognition is continued based on the previously selected image data and the newly selected image data. With this configuration, it is possible to recognize whether or not an object shown in an image taken with a reduced calculation load is a desired object.

Further, according to the present invention, a step of acquiring image data indicating an image, and a step of performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data And based on the information of the image data that has been subjected to a plurality of image processing procedures at different levels, each of which is reflected in the image by an output result output by a plurality of recognizers each having a characteristic property A step of recognizing whether the object is a desired object or not, wherein the step of performing the image processing procedure is performed in order to enable recognition of the desired object. Setting a filter of a first filter set of filters each having a characteristic property for image data; and a filter of the first filter set. Generating the first feature information indicating the feature of the image data set with the data, and in order to further enable recognition of the desired object, for the generated first feature information, A step of setting a filter of a second filter set of filters each having a characteristic property; and a second feature information indicating a feature of the first feature information in which a filter of the second filter set is set Generating the image data that has been subjected to the image processing procedure. In the step of recognizing, the image is reflected in the image according to an output result of some of the plurality of recognizers. Whether or not the object is a desired object. If the object cannot be recognized, the plurality of recognition devices other than the recognizer in which the output result is used are recognized. Object recognition method and recognizes based on the output result of the output results of some of the recognizer and has been used in the past within the vessel is provided. With this configuration, it is possible to recognize whether or not an object shown in an image taken with a reduced calculation load is a desired object.

Further, according to the present invention, a step of acquiring image data indicating an image, and a step of performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data And based on the information of the image data that has been subjected to a plurality of image processing procedures at different levels, each of which is reflected in the image by an output result output by a plurality of recognizers each having a characteristic property in but the object recognition method and a step 121 determines whether the desired the object, steps of applying the image processing procedure, in order to enable the recognition of the desired the object was acquired Setting a filter of a first filter set of filters each having a characteristic property for a part of the image data, and the first filter A step of generating first feature information indicating the characteristics of the image data in which a set filter is set, and in order to further enable recognition of the desired object, the generated first feature information For a part, a step of setting a filter of a second filter set of filters each having a characteristic property, and a feature of the first feature information in which the filter of the second filter set is set the image data consists of a step of generating a second said image data by the image processing procedure has been performed is characteristic information of, in said recognizing step, the pre-outs image processing procedure is applied to a portion shown Whether or not the object shown in the image is the desired object based on the output results output from the plurality of recognizers each having a characteristic property based on the information of Recognizing and recognizing until it can be recognized, in addition to a part of the image data subjected to the plurality of image processing procedures and a part of the image data subjected to the plurality of image processing procedures There is provided an object recognition method characterized in that recognition is performed based on information of a part subjected to a plurality of image processing procedures. With this configuration, it is possible to recognize whether or not an object shown in an image taken with a reduced calculation load is a desired object.

Further, according to the present invention, a step of acquiring image data indicating an image, and a step of performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data And based on the information of the image data that has been subjected to a plurality of image processing procedures at different levels, each of which is reflected in the image by an output result output by a plurality of recognizers each having a characteristic property in but the object recognition device and a step 121 determines whether the desired the object, steps of applying the image processing procedure, in order to enable the recognition of the desired the object was acquired Setting each filter of a first filter set of filters each having a characteristic property for the image data; and Generating the first feature information indicating the characteristics of the reduced image, and further enabling the recognition of the desired object. A step of setting a filter of a second filter set of filters each having a characteristic property with respect to the first feature information, and the first feature information in which the filter of the second filter set is set Generating the image data that has been subjected to the image processing procedure, which is second feature information indicating the characteristics of the image, and in the step of recognizing, the information of the image data that has been subjected to the image processing procedure Whether the object shown in the image is the desired object based on the output results output by a plurality of recognizers each having a characteristic property. An object recognition method characterized by recognizing the image and recognizing the reduced image based on image data information of images sequentially enlarged by the image processing means until the image can be recognized. Provided. With this configuration, it is possible to recognize whether or not an object shown in an image taken with a reduced calculation load is a desired object.

  The object recognition apparatus and object recognition method of the present invention have the above-described configuration, and can recognize whether or not an object shown in a photographed image with a reduced calculation load is a desired object.

<First Embodiment>
Hereinafter, the object recognition apparatus and the object recognition method according to the first embodiment of the present invention will be described with reference to FIGS. 1 to 7B. FIG. 1 is a configuration diagram showing the configuration of the object recognition apparatus according to the first embodiment of the present invention. FIG. 2 is a diagram for explaining the flow of processing in the object recognition apparatus according to the first embodiment of the present invention. FIG. 3 is a diagram for explaining image cutting processing in the object recognition apparatus according to the first embodiment of the present invention. FIG. 4 is a diagram for explaining extraction of a candidate image in the object recognition apparatus according to the first embodiment of the present invention. FIG. 5 is a diagram for explaining generation of first feature information in the object recognition apparatus according to the first embodiment of the present invention. FIG. 6 is a diagram for explaining generation of second feature information in the object recognition apparatus according to the first embodiment of the present invention. FIG. 7A is a flowchart for explaining a part of the object recognition flow in the object recognition apparatus according to the first embodiment of the present invention. FIG. 7B is a flowchart for explaining a part of the object recognition flow in the object recognition apparatus according to the first embodiment of the present invention.

  First, the configuration of the object recognition apparatus according to the first embodiment of the present invention will be described using FIG. 1 and FIG. As illustrated in FIG. 1, the object recognition apparatus 100 includes an image data acquisition unit 101, a cutout image extraction unit 102, a first filter setting unit 103, a first feature information generation unit 104, and a second filter setting unit 105. , Second feature information generation means 106 and recognition means 107. The image data acquisition unit 101 is a unit that acquires an image shot by an external shooting unit 109 (such as a camera). In the first embodiment of the present invention, the photographing unit 109 is not a component of the object recognition device 100, but may be a component of the object recognition device 100.

  The cut-out image extraction unit 102 is a unit that cuts out the image acquired by the image data acquisition unit 101 into a predetermined size. Hereinafter, the image cutting process will be described with reference to FIGS. 3 and 4. As shown in FIG. 3, in this process, the first acquired image is rescaled (image size change). This process is necessary for finding objects of different sizes. The acquired image is resized to a predetermined maximum image size scale_max, and a scaled image is acquired. Then, the image size is changed from the maximum size at a predetermined magnification scale_step, and a scaled image is acquired repeatedly up to a predetermined minimum size of the image. Then, the cutout image extraction unit 102 cuts out candidate images from the scaled image. This cutting method will be described with reference to FIG. As shown in FIG. 4, a candidate image is obtained by scanning a window in both the x-axis and y-axis directions of the image. Here, when the units of the x-axis and y-axis of the region that is one unit of the scanning region are set to step_x and step_y, step_x> = 1 and step_y> = 1. Even if they are the same, they are not the same. May be. The extraction position of the candidate image may be based on information of other means such as a radar other than scanning the image as described above.

  In the following description, the object is described as a person, but is not limited to a person. As shown in FIG. 2, the first filter setting means 103 is a means for setting a predetermined filter in a first filter set 200 of filters (also called convolution kernels) each having a characteristic property. . Here, the first filter set 200 is a set of filters generated in advance based on information (data) given in order to detect a person. The number is 12 here, but this number It is not limited. In addition, each filter having a characteristic property is, for example, a filter for extracting a human feature, and a filter that specializes in extracting a head feature or a feature of a leg. Says a filter that is good at it. In FIG. 2, for example, the filter 201 corresponds to a filter that is good at extracting features of the head, and the filter 202 is equivalent to a filter that is good at extracting features of the legs.

  The first feature information generation unit 104 is a unit that generates first feature information 300 indicating the features of an image in which a predetermined filter is set in the first filter set 200 by the first filter setting unit 103. . Here, generation of the first feature information 300 will be described with reference to FIG. As shown in FIG. 5, the first feature information generation unit 104 generates first feature information 300 using a linear or nonlinear function for each pixel of a candidate image that has been filtered. For example, in the case of a nonlinear function, a tangent hyperbolic function as shown in FIG. 5 is used. Here, p is a pixel value of a candidate image to be filtered, q is a pixel value of the first feature information 300, and f (p) is a linear or nonlinear function. In this example, a tangent hyperbolic function is used for f (p). In this case, although -1 <= f (p) <= 1, it may be a function having a range in which output values are different. Accordingly, the first feature information generation unit 104 generates, for example, the first feature information 301 in the image in which the filter 201 is set, and the generated first feature information 301 is used as the second filter setting unit 105. To hand over. Note that the number of filters selected is one here, but the number is not limited to one. The first feature information 300 is image information generated by setting a filter, and is also referred to as a first feature map 300.

  The second filter setting unit 105 is characterized by each of the first feature information 300 generated by the first feature information generation unit 104 in order to further recognize a desired object (person). This is means for setting a predetermined filter in the second filter set 400 of a filter (convolution kernel) having a natural property. Here, like the first filter set 200, the second filter set 400 is a set of filters generated in advance based on information (data) given to detect a person. Although the number is 12 here, the number is not limited to this.

  In addition, each filter having a characteristic property is, for example, a filter for extracting a feature in the direction of movement of a person, such as a filter that is good at extracting a feature in the vertical direction or a feature in the horizontal direction. A filter that is good at extracting. In FIG. 2, for example, the filter 401 corresponds to a filter that is good at extracting vertical features, and the filter 402 is equivalent to a filter that is good at extracting horizontal features. Specifically, the second filter setting unit 105 sets the filter 401 for the first feature information 301 generated by the first feature information generation unit 104, and the first feature in which the filter 401 is set. The information 301 is transferred to the second feature information generation unit 106.

  The second feature information generating unit 106 is a unit that generates second feature information 500 indicating the feature of the first feature information in which a predetermined filter is set in the second filter set 400. Here, generation of the second feature information 500 will be described with reference to FIG. As shown in FIG. 6, the second feature information generation unit 106 uses the linear or nonlinear function for each pixel of the first feature information filtered by the second filter setting unit 105. Feature information 500 is generated. For example, in the case of a nonlinear function, a tangent hyperbolic function is used. Here, p is the pixel value of the candidate image that has been filtered, q is the pixel value of the second feature information 500, and f (p) is a linear or nonlinear function. In this example, a tangent hyperbolic function is used for f (p). In this case, although -1 <= f (p) <= 1, it may be a function having a range in which output values are different. When generating the second feature information 500, it is possible to generate the second feature information 500 using one or a plurality of first feature information that has already been calculated. From this, the second feature information generation unit 106 generates, for example, the first feature information 301 in which the filter 401 is set and the second feature information 501 in the first feature information 302 in which the filter 402 is set. Then, the generated second feature information 501 is transferred to the recognition unit 107. The second feature information 500 is image information generated by setting a filter, and is also referred to as a second feature map 500.

  Based on the second feature information generated by the second feature information generating unit 106, the recognizing unit 107 recognizes whether or not what is shown in the acquired image is a desired object (person). To do. The recognition by the recognition unit 107 will be described later. Specifically, the recognition means 107 is a desired object (person) that is reflected in the acquired image based on the second feature information 501 generated by the second feature information generation means 106. Recognize whether there is. If the image can be recognized, the recognition process for the image is terminated and the next image is recognized.

  On the other hand, when the second feature information 501 alone cannot be recognized, the first filter setting unit 103 sets the filter 202 to the image cut out by the cut-out image extraction unit 102, and the first feature information generation unit 104. Generates first feature information 302 indicating the features of the image for which the filter 202 is set. Then, the second filter setting unit 105 sets the filter 402 to the first feature information 302 generated, and the second feature information generation unit 106 sets the first feature information 301 and the filter 401 of the second filter set. , 402 is used to generate second feature information 502 indicating the features of the first feature information 302. Then, the recognition unit 107 recognizes based on the generated second feature information 502 and the already generated second feature information 501. If this cannot be recognized, the above-described process is repeated until the recognition is possible. As a result, it is possible to reduce the processing load as compared to setting all the filters in the image in advance and processing the set image.

  Here, the recognition means 107 will be described. As shown in FIG. 2, the recognition means 107 outputs a recognition result based on the results output from a plurality of units called hidden layers. As the recognition means 107, for example, a neural network can be used. The same applies to the recognition means (recognition device) shown in FIG. 9, FIG. 12, FIG. 13, FIG. Each unit is a node for recognizing a predetermined part of a person, for example, a node for recognizing a face or a node for recognizing a foot. These nodes are, for example, weighted and linearly coupled. For example, if the result of the linear combination exceeds a predetermined threshold, it is determined that the person is a person. Although the unit is 40 here, it is not limited to 40. Note that the details of the recognition unit 107 are well-known techniques, and thus detailed description thereof is omitted.

  Next, an object recognition flow in the object recognition apparatus according to the first embodiment of the present invention will be described with reference to FIGS. 7A and 7B. As shown in FIGS. 7A and 7B, first, the image data acquisition unit 101 acquires an image shot by an external shooting unit 109 (such as a camera) (step S701). Next, the cutout image extraction unit 102 changes the image acquired by the image data acquisition unit 101 to the size of scale_max described above (step S702). Then, a scaled image is acquired from the resized image (step S703). The next candidate image is cut out from the scaled image, and the candidate image is acquired (step S704). Then, f = 1 is set (step S705). Here, f indicates the number of repetitions of the object recognition flow. Next, the first filter setting unit 103 sets the filter Fil1_f of the first filter set 200 as an extracted candidate image (step S706). Here, Fil1_f means the f-th filter of the first filter set. The first feature information generation unit 104 generates the first feature information FM1_f using the candidate image and the filter Fil1_f of the first filter set (step S707). Here, FM1_f means the f-th first feature information.

  The second filter setting unit 105 applies the second filter set 400 to the first feature information FM1_f generated by the first feature information generation unit 104 and / or the first feature information generated in the past. The filter Fil2_f is set (step S708). The second feature information generation means 106 uses the first feature information FM1_1, FM1_2,..., FM1_f and at least one of the filters Fil2_1, Fil2_2,. 2 feature information FM2_f is generated (step S709). The recognition unit 107 calculates an output value for recognition based on the second feature information FM2_1, FM2_2,..., FM2_f generated by the second feature information generation unit 106 (step S710). The recognizing unit 107 has an output value larger than a first threshold value for performing a predetermined recognition (threshold value for determining whether a candidate image is a person (pedestrian)). Whether or not (step S711). If it is determined that the output value is greater than the first threshold, the candidate image is recognized as a person (pedestrian).

  On the other hand, when it is determined that the output value is equal to or lower than the first threshold, the output value is determined to be a second threshold (a candidate image is not a person (pedestrian)). It is judged whether it is smaller than (threshold value) (step S712). When it is determined that the output value is smaller than the second threshold, it is recognized that the person is not a person (pedestrian). On the other hand, if it is determined that the output value is greater than or equal to the second threshold value, it is determined whether f is smaller than F (step S713). Here, F is the maximum number of first and second feature information. When f is smaller than F, 1 is added to f to make it f (step S714), and the process returns to step S706. On the other hand, when f is F or more, it recognizes that the candidate image is not a person (pedestrian). Then, it is determined whether or not there is a candidate image in the scaled image (step S715). If it is determined that there is, the process returns to step 704. If it is determined that there is no scale_factor, it is determined whether or not scale_factor is larger than scale_min (step S716). If it is determined that scale_factor is larger than scale_min, scale_factor is multiplied by scale_step to set it as scale_factor (step S717), and the process returns to step S703. On the other hand, if it is determined in step S716 that scale_factor is less than or equal to scale_min, the process ends and preparation for image acquisition is performed.

<Second Embodiment>
Next, an object recognition apparatus according to a second embodiment of the present invention will be described with reference to FIGS. 8 to 10B. FIG. 8 is a block diagram showing the configuration of the object recognition apparatus according to the second embodiment of the present invention. FIG. 9 is a diagram for explaining a processing flow of the object recognition apparatus according to the second embodiment of the present invention. FIG. 10A is a flowchart for explaining a part of the object recognition flow in the object recognition apparatus according to the second embodiment of the present invention. FIG. 10B is a flowchart for explaining a part of the object recognition flow in the object recognition apparatus according to the second embodiment of the present invention. As shown in FIG. 8, the object recognition apparatus 600 according to the second embodiment has the same configuration as the object recognition apparatus 100 according to the first embodiment, but some components. The processing method is different. Below, the process of the target object recognition apparatus which concerns on 2nd Embodiment is demonstrated using FIG.

  The processing of the image data acquisition unit 601 and the cutout image extraction unit 602 in the object recognition apparatus 600 according to the second embodiment is the same as that of the first embodiment. The first filter setting unit 603 sets each of the filters (filters 211, 212, etc.) of the first filter set 210 shown in FIG. 9 with respect to the candidate image extracted from the cut-out image extraction unit 602. . Then, the first feature information generation unit 604 generates first feature information 310 (first feature information 311, 312, etc.) indicating the feature of the image in which each filter is set. Regarding the generation of the first feature information, the first embodiment is referred to. The second filter setting unit 605 sets the filters (filters 411, 412 and the like) of the second filter set 410 with respect to the first feature information 310 generated by the first feature information generation unit 604. The second feature information generating unit 606 is configured to provide second feature information 510 (second feature information) indicating the feature of the first feature information in which the filters (filters 411, 412 and the like) of the second filter set 410 are set. 511, 512, etc.). For the generation of the second feature information, the first embodiment is referred to.

  Unlike the first embodiment, the recognizing unit 607 first recognizes only one node 700 in the hidden layer, and if it cannot recognize the other one node in the hidden layer. Recognition is performed based on 701 and the node 700. If it still cannot be recognized, the node 702 is added for recognition. Thus, it recognizes by adding a node until it can recognize. Thus, by changing the amount of calculation according to the difficulty of recognizing the cut-out image, it is possible to reduce the load on the entire recognition unit. Note that the first embodiment and the second embodiment may be combined. That is, recognition is performed based on a result output by one node based on one second feature information.

  Next, an object recognition flow in the object recognition apparatus according to the second embodiment of the present invention will be described with reference to FIGS. 10A and 10B. As shown in FIGS. 10A and 10B, first, the image data acquisition unit 601 acquires an image shot by an external shooting unit 109 (such as a camera) (step S1001). Next, the cutout image extraction unit 102 changes the image acquired by the image data acquisition unit 601 to the size of scale_max described above (step S1002). Then, a scaled image is acquired from the resized image (step S1003). The next candidate image is cut out from the scaled image, and the candidate image is acquired (step S1004). Next, the first filter setting unit 603 sets the filters Fil1_1 to Fil1_n of the first filter set 210 to the extracted candidate images (step S1005). Here, n of Fil1_n is the number of first filter sets 210. The first feature information generation unit 604 generates first feature information FM1_1 to FM1_n using the image that is a candidate and the filters Fil1_1 to Fil1_n of the first filter set (step S1006).

  The second filter setting unit 605 sets the filters Fil2_1 to Fil2_n of the second filter set 410 for the first feature information generated by the first feature information generation unit 604 (step S1007). The second feature information generating unit 606 generates second feature information FM2_1 to FM2_n using the generated first feature information FM1_1 to FM1_n and the filters Fil2_1 to Fil2_n of the second filter set 410 ( Step S1008). Then, g = 1 is set (step S1009). Here, g is a number indicating a node of a hidden layer to be described later. That is, g = 1 means the first node. Based on the second feature information FM2_1 to FM2_n generated by the second feature information generating unit 606, the recognizing unit 607 calculates an output value from the nodes Node_1 to Node_g of the hidden layer (step S1010). Here, Node_g means a node of the g-th hidden layer. And the recognition means 607 is larger than the 1st threshold value (threshold value for judging whether the image made into a candidate is a person (pedestrian)) for the recognition whose output value was decided beforehand. Whether or not (step S1011). If it is determined that the output value is greater than the first threshold, the candidate image is recognized as a person (pedestrian).

  On the other hand, when it is determined that the output value is equal to or lower than the first threshold, the output value is determined to be a second threshold (a candidate image is not a person (pedestrian)). It is judged whether it is smaller than (threshold value) (step S1012). When it is determined that the output value is smaller than the second threshold, it is recognized that the person is not a person (pedestrian). On the other hand, if it is determined that the output value is greater than or equal to the second threshold, it is determined whether g is smaller than G (step S1013). Here, G indicates the number of nodes in the hidden layer. When g is smaller than G, 1 is added to g (step S1014), and the process returns to step S1010. On the other hand, when g is G or more, the candidate image is recognized as not a person (pedestrian). Then, it is determined whether or not there is a candidate image in the scaled image (step S1015). If it is determined that there is, the process returns to step 1004. If it is determined that there is no scale_factor, it is determined whether or not scale_factor is larger than scale_min (step S1016). When it is determined that scale_factor is larger than scale_min, scale_factor is multiplied by scale_step to set it as scale_factor (step S1017), and the process returns to step S1003. On the other hand, if it is determined in step S1016 that scale_factor is less than or equal to scale_min, the process ends and preparation for image acquisition is performed.

<Third Embodiment>
Next, an object recognition apparatus according to a third embodiment of the present invention will be described with reference to FIGS. 11 to 14B. FIG. 11 is a block diagram showing the configuration of the object recognition apparatus according to the third embodiment of the present invention. FIG. 12 is a diagram for explaining a processing flow of the object recognition apparatus according to the third embodiment of the present invention. FIG. 13 is a diagram for explaining the processing flow of the object recognition apparatus according to the third embodiment of the present invention. FIG. 14A is a flowchart for explaining a part of the object recognition flow in the object recognition apparatus according to the third embodiment of the present invention. FIG. 14B is a flowchart for explaining a part of the object recognition flow in the object recognition apparatus according to the third embodiment of the present invention. As shown in FIG. 11, the object recognition apparatus 800 according to the third embodiment has the same configuration as the object recognition apparatus 100 according to the first embodiment, but some components. The processing method is different. Below, the process of the target object recognition apparatus which concerns on 3rd Embodiment is demonstrated using FIG.12 and FIG.13.

  The processing of the image data acquisition unit 801 and the cutout image extraction unit 802 in the object recognition apparatus 800 according to the third embodiment is the same as that of the first embodiment. The first filter setting unit 803 applies each of the filters (filters 221, 222, etc.) of the first filter set 220 shown in FIG. 12 to a part of the candidate images extracted from the cutout image extraction unit 802. Set. Then, the first feature information generation unit 804 includes first feature information 320 (first feature information 321, 322, etc.) indicating the features of the image set as a part of the image for which each filter is a candidate. Is generated. Regarding the generation of the first feature information, the first embodiment is referred to. The second filter setting unit 805 applies the first filter to the part of the first feature information 320 generated by the first feature information generation unit 804 (the part where the filter of the first filter set 220 is set). The filters (filters 421, 422, etc.) of the second filter set 420 are set. The second feature information generation unit 806 is configured to provide second feature information 520 (second feature information) indicating the feature of the first feature information in which the filters (filters 421, 422, etc.) of the second filter set 420 are set. 521, 522, etc.). For the generation of the second feature information, the first embodiment is referred to.

  Based on the second feature information 520 generated by the second feature information generating unit 806, the recognizing unit 807 outputs the result output from the linearly coupled nodes as in the first embodiment. Recognize based on. Note that, as in the second embodiment, recognition may be performed based on output results from individual nodes. If it cannot be recognized, the first filter setting means 803 sets the filters of the first filter set 220 in parts other than the part where the first filter set 220 has already been set for the image. Then, as shown in FIG. 13, the first feature information generation unit 804 again includes first feature information 320 indicating the features of the image in which each filter of the first filter set 220 is set as a part of the image. '(First feature information 321', 322 ', etc.) is generated.

  The second filter setting means 805 sets the filter of the second filter set 420 for the part of the generated first feature information 320 ′ where the filter of the first filter set 220 is newly set. . As shown in FIG. 13, the second feature information generation unit 806 again includes second feature information 520 ′ (representing the feature of the image in which each filter of the second filter set 420 is set as a part of the image. Second feature information 521 ′, 522 ′, etc.) is generated. Then, the recognition unit 807 recognizes based on the result output from the linearly coupled nodes based on the generated second feature information 520 ′. If this cannot be recognized, the above processing is continued until it can be recognized. As a result, it is possible to recognize more efficiently than setting and processing a filter on the entire image from the beginning.

  Next, an object recognition flow in the object recognition apparatus according to the third embodiment of the present invention will be described with reference to FIGS. 14A and 14B. As shown in FIGS. 14A and 14B, first, the image data acquisition unit 801 acquires an image shot by an external shooting unit 109 (camera or the like) (step S1401). Next, the cutout image extraction unit 802 changes the image acquired by the image data acquisition unit 801 to the size of scale_max described above (step S1402). Then, a scaled image is acquired from the resized image (step S1403). The next candidate image is cut out from the scaled image, and the candidate image is acquired (step S1404). Here, h = 1 is set (step S1405). h indicates the number of the divided area when the candidate image is divided into H pieces. That is, h = 1 indicates the first divided area among the H divided areas. Next, the first filter setting unit 803 sets the filters Fil1_1 to Fil1_n of the first filter set 220 as a part of the extracted candidate images (step S1406). Here, n of Fil1_n is the number of first filter sets 220. The first feature information generation unit 804 generates first feature information FM1_1 to FM1_n using the candidate image and the filters Fil1_1 to Fil1_n of the first filter set (step S1407).

  The second filter setting unit 805 sets the filters Fil2_1 to Fil2_n of the second filter set 420 for the first feature information generated by the first feature information generation unit 804 (step S1408). The second feature information generating unit 806 generates second feature information FM2_1 to FM2_n using the generated first feature information FM1_1 to FM1_n and the filters Fil2_1 to Fil2_n of the second filter set 420 ( Step S1409). The recognition unit 807 calculates an output value for recognition based on the second feature information FM2_1 to FM2_n generated by the second feature information generation unit 806 (step S1410). The recognizing unit 807 has an output value larger than a first threshold value for performing a predetermined recognition (threshold value for determining whether or not a candidate image belongs to a person (pedestrian)). It is determined whether or not (step S1411). If it is determined that the output value is greater than the first threshold, the candidate image is recognized as a person (pedestrian).

  On the other hand, when it is determined that the output value is equal to or lower than the first threshold, the output value is determined to be a second threshold (a candidate image is not a person (pedestrian)). It is determined whether it is smaller than the threshold value (step S1412). When it is determined that the output value is smaller than the second threshold, it is recognized that the person is not a person (pedestrian). On the other hand, if it is determined that the output value is greater than or equal to the second threshold, it is determined whether h is smaller than H (step S1413). Here, H indicates the number of divisions when the candidate image is divided. If h is smaller than H, 1 is added to h (step S1414), and the process returns to step S1406. On the other hand, if h is H or more, the candidate image is recognized as not a person (pedestrian). Then, it is determined whether or not there is a candidate image in the scaled image (step S1415). If it is determined that there is, the process returns to step 1404. If it is determined that there is no scale_factor, it is determined whether or not scale_factor is larger than scale_min (step S1416). When it is determined that scale_factor is larger than scale_min, scale_factor is multiplied by scale_step, which is set as scale_factor (step S1417), and the process returns to step S1403. On the other hand, if it is determined in step S1416 that scale_factor is less than or equal to scale_min, the process ends and preparation for image acquisition is performed.

<Fourth embodiment>
Next, an object recognition apparatus according to a fourth embodiment of the present invention will be described with reference to FIGS. 15 to 18B. FIG. 15: is a block diagram which shows the structure of the target object recognition apparatus which concerns on the 4th Embodiment of this invention. FIG. 16 is a diagram for explaining a process flow of the object recognition apparatus according to the fourth embodiment of the present invention. FIG. 17 is a diagram for explaining the processing flow of the object recognition apparatus according to the fourth embodiment of the present invention. FIG. 18A is a flowchart for explaining a part of the object recognition flow in the object recognition apparatus according to the fourth embodiment of the present invention. FIG. 18B is a flowchart for explaining a part of the object recognition flow in the object recognition apparatus according to the fourth embodiment of the present invention. As shown in FIG. 15, the object recognition apparatus 900 according to the fourth embodiment has the same configuration as the object recognition apparatus 100 according to the first embodiment, but some components The processing method is different. Below, the process of the target object recognition apparatus which concerns on 4th Embodiment is demonstrated using FIG.16 and FIG.17.

  The processing of the image data acquisition unit 901 and the cutout image extraction unit 902 in the object recognition apparatus 900 according to the fourth embodiment is the same as that of the first embodiment. The first filter setting unit 903 sets the filters (filters 231 and 232, etc.) of the first filter set 230 shown in FIG. 16 for the image extracted from the cutout image extraction unit 902. Then, the first feature information generation unit 904 reduces the image in which each filter is set, and the first feature information 330 (first feature information 331, 332, etc.) indicating the feature of the reduced image. Generate. The second filter setting unit 905 sets the filters (filters 431, 432, etc.) of the second filter set 430 for the first feature information 330 generated by the first feature information generation unit 904. The second feature information generation unit 906 includes second feature information 530 (second feature information) indicating the feature of the first feature information in which the filters (filters 431, 432, etc.) of the second filter set 430 are set. 531, 532, etc.).

  Based on the second feature information 530 generated by the second feature information generating unit 906, the recognizing unit 907 is based on the result output from the linearly coupled nodes as in the first embodiment. recognize. Note that, as in the second embodiment, recognition may be performed based on output results from individual nodes. If it cannot be recognized, the first feature information generation unit 904 enlarges the reduced image set with the previous filter at a predetermined rate, as shown in FIG. 17, and shows the features of the enlarged image. First feature information 330 ′ (first feature information 331 ′, 332 ′, etc.) is generated. Here, enlarging at a predetermined rate means, for example, that when the size of the first reduced image is 1, the size is enlarged by 2 times, 3 times,.

  The second filter setting unit 905 sets the filters (filters 431, 432, etc.) of the second filter set 430 for the first feature information 330 ′ generated by the first feature information generation unit 904. . The second feature information generating unit 906 is configured to provide second feature information 530 ′ (second feature) indicating the feature of the first feature information in which the filters (filters 431, 432, etc.) of the second filter set 430 are set. Information 531 ′, 532 ′, etc.). Then, the recognition unit 907 recognizes based on the result output from the linearly coupled node based on the generated second feature information 530 ′. If this cannot be recognized, the above processing is continued until it can be recognized. As a result, the resolution of the reduced image is reduced, the processing load is reduced correspondingly, and the entire image can be recognized efficiently.

  Next, an object recognition flow in the object recognition apparatus according to the fourth embodiment of the present invention will be described with reference to FIGS. 18A and 18B. As shown in FIGS. 18A and 18B, first, the image data acquisition unit 901 acquires an image captured by an external imaging unit 109 (camera or the like) (step S1801). Next, the cutout image extraction unit 902 changes the image acquired by the image data acquisition unit 901 to the size of scale_max described above (step S1802). Then, a scaled image is acquired from the resized image (step S1803). The next candidate image is cut out from the scaled image, and the candidate image is acquired (step S1804). Here, j = 1 is set (step S1805). j indicates one image among a plurality of images (J images) obtained by reducing candidate images at different rates, and j = 1 indicates the image with the lowest magnification among the plurality of images. ing. Next, the first filter setting unit 903 sets the filters Fil1_1 to Fil1_n of the first filter set 230 to the extracted candidate images (step S1806). Here, n of Fil1_n is the number of first filter sets 230. The first feature information generation unit 904 uses the image as a candidate and the filters Fil1_1 to Fil1_n of the first filter set to obtain the first feature information FM1_1 to FM1_n reduced at a magnification corresponding to the jth. Generate (step S1807).

  The second filter setting unit 905 sets the filters Fil2_1 to Fil2_n of the second filter set 430 for the first feature information generated by the first feature information generation unit 904 (step S1808). The second feature information generation unit 906 generates the second feature information FM2_1 to FM2_n using the generated first feature information FM1_1 to FM1_n and the filters Fil2_1 to Fil2_n of the second filter set 430 ( Step S1809). The recognition unit 907 calculates an output value for recognition based on the second feature information FM2_1 to FM2_n generated by the second feature information generation unit 906 (step S1810). And the recognition means 907 is larger than the 1st threshold value (threshold value for judging whether the image made into a candidate is a person (pedestrian)) for the recognition whose output value was decided beforehand. It is determined whether or not (step S1811). If it is determined that the output value is greater than the first threshold, the candidate image is recognized as a person (pedestrian).

  On the other hand, when it is determined that the output value is equal to or lower than the first threshold, the output value is determined to be a second threshold (a candidate image is not a person (pedestrian)). It is judged whether it is smaller than (threshold value) (step S1812). When it is determined that the output value is smaller than the second threshold, it is recognized that the person is not a person (pedestrian). On the other hand, if it is determined that the output value is equal to or greater than the second threshold value, it is determined whether j is smaller than J (step S1813). Here, J indicates the number of images when the candidate images are reduced at different rates. When j is smaller than J, 1 is added to j (step S1814), and the process returns to step S1806. On the other hand, if j is greater than or equal to J, the candidate image is recognized as not a person (pedestrian). Then, it is determined whether or not there is a candidate image in the scaled image (step S1815). If it is determined that there is, the process returns to step 1804. If it is determined that there is no scale_factor, it is determined whether or not scale_factor is larger than scale_min (step S1816). When it is determined that scale_factor is larger than scale_min, scale_factor is multiplied by scale_step to set it as scale_factor (step S1817), and the process returns to step S1803. On the other hand, if it is determined in step S1816 that scale_factor is less than or equal to scale_min, the process ends and preparation for image acquisition is performed.

  The object recognition apparatus and the object recognition method according to the present invention can recognize whether or not an object shown in a photographed image with a reduced calculation load is a desired object. This is useful for an object recognition apparatus and an object recognition method for recognizing whether or not an object is a desired object.

It is a block diagram which shows the structure of the target object recognition apparatus which concerns on the 1st Embodiment of this invention. It is a figure for demonstrating the flow of the process in the target object recognition apparatus which concerns on the 1st Embodiment of this invention. It is a figure for demonstrating the cutting-out process of the image in the target object recognition apparatus which concerns on the 1st Embodiment of this invention. It is a figure for demonstrating extraction of the image used as the candidate in the target object recognition apparatus which concerns on the 1st Embodiment of this invention. It is a figure for demonstrating the production | generation of the 1st feature information in the target object recognition apparatus which concerns on the 1st Embodiment of this invention. It is a figure for demonstrating the production | generation of the 2nd feature information in the target object recognition apparatus which concerns on the 1st Embodiment of this invention. It is a flowchart for demonstrating a part of target object recognition flow in the target object recognition apparatus which concerns on the 1st Embodiment of this invention. It is a flowchart for demonstrating a part of target object recognition flow in the target object recognition apparatus which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the structure of the target object recognition apparatus which concerns on the 2nd Embodiment of this invention. It is a figure for demonstrating the flow of the process in the target object recognition apparatus which concerns on the 2nd Embodiment of this invention. It is a flowchart for demonstrating a part of target object recognition flow in the target object recognition apparatus which concerns on the 2nd Embodiment of this invention. It is a flowchart for demonstrating a part of target object recognition flow in the target object recognition apparatus which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows the structure of the target object recognition apparatus which concerns on the 3rd Embodiment of this invention. It is a figure for demonstrating the flow of the process in the target object recognition apparatus which concerns on the 3rd Embodiment of this invention. It is a figure for demonstrating the flow of the process in the target object recognition apparatus which concerns on the 3rd Embodiment of this invention. It is a flowchart for demonstrating a part of target object recognition flow in the target object recognition apparatus which concerns on the 3rd Embodiment of this invention. It is a flowchart for demonstrating a part of target object recognition flow in the target object recognition apparatus which concerns on the 3rd Embodiment of this invention. It is a block diagram which shows the structure of the target object recognition apparatus which concerns on the 4th Embodiment of this invention. It is a figure for demonstrating the flow of the process in the target object recognition apparatus which concerns on the 4th Embodiment of this invention. It is a figure for demonstrating the flow of the process in the target object recognition apparatus which concerns on the 4th Embodiment of this invention. It is a flowchart for demonstrating a part of target object recognition flow in the target object recognition apparatus which concerns on the 4th Embodiment of this invention. It is a flowchart for demonstrating a part of target object recognition flow in the target object recognition apparatus which concerns on the 4th Embodiment of this invention.

Explanation of symbols

100, 600, 800, 900 Object recognition device 101, 601, 801, 901 Image data acquisition means 102, 602, 802, 902 Cut image extraction means 103, 603, 803, 903 First filter setting means 104, 604, 804, 904 First feature information generating means 105, 605, 805, 905 Second filter setting means 106, 606, 806, 906 Second feature information generating means 107, 607, 807, 907 Recognizing means 109 Imaging means ( camera)
200, 210, 220, 230 First filter set 10, 11, 12, 13, 201, 202, 211, 212, 221, 222, 231, 232, 401, 402, 411, 412, 421, 422, 431, 432 Filter 300, 301, 302, 310, 311, 312, 320, 320 ', 321, 321', 322, 322 ', 330, 330', 331, 331 ', 332, 332' First feature information (feature map)
400, 410, 420, 430 Second filter set 500, 501, 502, 510, 511, 512, 520, 520 ′, 521, 521 ′, 522, 522 ′, 530, 530 ′, 531, 531 ′, 532 532 ′, 540 Second feature information (feature map)
700, 701, 702 nodes

Claims (8)

Image data acquisition means for acquiring image data indicating an image;
Image processing means for performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data;
Based on information of the image data subjected to a plurality of image processing procedures at different levels, it is desirable that the image is reflected in the image according to output results output from a plurality of recognizers each having a characteristic property An object recognition device comprising: a recognition means for recognizing whether or not the object is
The image processing means includes
In order to enable recognition of a desired object, a first filter that sets a predetermined filter in a first filter set of filters each having a characteristic property with respect to the acquired image data. Setting means;
First feature information generating means for generating first feature information indicating a feature of the image data in which a predetermined filter is set in the first filter set;
Filters each having a characteristic property with respect to the first feature information generated and / or the first feature information generated in the past in order to further enable recognition of the desired object. Second filter setting means for setting a predetermined filter in the second filter set;
A second feature for generating the image data subjected to the image processing procedure, which is second feature information indicating a feature of the first feature information in which a predetermined filter is set in the second filter set. Information generating means,
Said recognition means, pre-outs image processing procedure to select a portion of image data among the image data that has been subjected, the one that is reflected in the image is desired on the basis of the image data selected Recognizing whether or not it is an object, until it can be recognized, select some image data of the image data subjected to the image processing procedure other than the selected image data until it can be recognized, An object recognition apparatus characterized in that recognition is continued based on the image data selected in the past and the newly selected image data. Image data acquisition means for acquiring image data indicating an image;
Image processing means for performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data;
Based on information of the image data subjected to a plurality of image processing procedures at different levels, it is desirable that the image is reflected in the image according to output results output from a plurality of recognizers each having a characteristic property An object recognition device comprising: a recognition means for recognizing whether or not the object is
The image processing means includes
First filter setting means for setting a filter of a first filter set of filters each having a characteristic property for the acquired image data in order to enable recognition of a desired object; ,
First feature information generating means for generating first feature information indicating a feature of the image data in which a filter of the first filter set is set;
In order to further enable recognition of the desired object, a second filter set of a second filter set of filters each having a characteristic property is set for the generated first feature information. Filter setting means;
Second feature information generating means for generating the image data subjected to the image processing procedure, which is second feature information indicating the feature of the first feature information in which the filter of the second filter set is set And
The recognizing unit recognizes whether or not an object reflected in the image is a desired object based on an output result of a part of the plurality of recognizing devices. Until possible, the recognition is performed based on the output results of some of the plurality of recognizers other than the recognizer in which the output result is used and the output results used in the past. Object recognition device. Image data acquisition means for acquiring image data indicating an image;
Image processing means for performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data;
Based on information of the image data subjected to a plurality of image processing procedures at different levels, it is desirable that the image is reflected in the image according to output results output from a plurality of recognizers each having a characteristic property An object recognition device comprising: a recognition means for recognizing whether or not the object is
The image processing means includes
In order to enable recognition of a desired object, a first filter set of a first filter set of filters each having a characteristic property is set for each part of the acquired image data. Filter setting means;
First feature information generating means for generating first feature information indicating a feature of the image data in which a filter of the first filter set is set;
In order to further enable recognition of the desired object, a filter of a second filter set of filters each having a characteristic property is set for a part of the generated first characteristic information. A second filter setting means;
Second feature information generating means for generating the image data subjected to the image processing procedure, which is second feature information indicating the feature of the first feature information in which the filter of the second filter set is set And
It said recognition means, pre-outs image processing procedure on the basis of the information of the image data subjected to the part by each resulting output to output a plurality of recognizers having characteristic properties, appearing in the image A part of the image data subjected to the plurality of image processing procedures and the plurality of image processing procedures until it can be recognized. And recognizing based on information of a part newly subjected to the plurality of image processing procedures in addition to a part of the image data subjected to. Image data acquisition means for acquiring image data indicating an image;
Image processing means for performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data;
Based on information of the image data subjected to a plurality of image processing procedures at different levels, it is desirable that the image is reflected in the image according to output results output from a plurality of recognizers each having a characteristic property An object recognition device comprising: a recognition means for recognizing whether or not the object is
The image processing means includes
First filter setting means for setting a filter of a first filter set of filters each having a characteristic property for the acquired image data in order to enable recognition of a desired object. When,
First feature information generating means for reducing the image in which the filter of the first filter set is set, and generating first feature information indicating the feature of the reduced image;
In order to further enable recognition of the desired object, a second filter set of a second filter set of filters each having a characteristic property is set for the generated first feature information. Filter setting means;
Second feature information generating means for generating the image data subjected to the image processing procedure, which is second feature information indicating the feature of the first feature information in which the filter of the second filter set is set And
Desirably, the recognition means is reflected in the image based on output results output from a plurality of recognizers each having a characteristic property based on information of the image data subjected to the image processing procedure. Whether or not it is the object, and if it cannot be recognized, the reduced image is recognized based on image data information of images sequentially enlarged by the image processing means. Feature object recognition device. Obtaining image data representing an image;
Performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data;
Based on information of the image data subjected to a plurality of image processing procedures at different levels, it is desirable that the image is reflected in the image according to output results output from a plurality of recognizers each having a characteristic property Recognizing whether or not the object is a target object,
The step of applying the image processing procedure includes:
Setting a predetermined filter in a first filter set of filters each having a characteristic property for the acquired image data in order to enable recognition of a desired object;
Generating first characteristic information indicating characteristics of the image data in which a predetermined filter is set in the first filter set;
Filters each having a characteristic property with respect to the first feature information generated and / or the first feature information generated in the past in order to further enable recognition of the desired object. Setting a predetermined filter in the second filter set of:
Generating the image data subjected to the image processing procedure, which is second feature information indicating the feature of the first feature information in which a predetermined filter is set in the second filter set. ,
In the step of recognizing, selecting a part of the image data among the image data pre-outs image processing procedure has been performed, what is reflected in the image based on the image data selected desired If it cannot be recognized, select some image data of the image data subjected to the image processing procedure other than the selected image data until it can be recognized. Then, recognition is continued based on the image data selected in the past and the newly selected image data. Obtaining image data representing an image;
Performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data;
Based on information of the image data subjected to a plurality of image processing procedures at different levels, it is desirable that the image is reflected in the image according to output results output from a plurality of recognizers each having a characteristic property Recognizing whether or not the object is a target object,
The step of applying the image processing procedure includes:
Setting a filter of a first filter set of filters each having a characteristic property for the acquired image data to enable recognition of the desired object;
Generating first feature information indicating features of the image data in which a filter of the first filter set is set;
Setting a filter of a second filter set of filters each having a characteristic property for the generated first feature information to further enable recognition of the desired object;
Generating the image data subjected to the image processing procedure that is the second feature information indicating the feature of the first feature information in which the filter of the second filter set is set,
In the step of recognizing, if it is not possible to recognize whether or not what is reflected in the image is the desired object based on the output result of some of the plurality of recognizers Is recognized based on the output results of some recognizers of the plurality of recognizers other than the recognizer in which the output result is used and the output results used in the past until it can be recognized. The object recognition method. Obtaining image data representing an image;
Performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data;
Based on information of the image data subjected to a plurality of image processing procedures at different levels, it is desirable that the image is reflected in the image according to output results output from a plurality of recognizers each having a characteristic property Recognizing whether or not the object is a target object,
Steps to perform the image processing procedure,
Setting a filter of a first filter set of filters each having a characteristic property for a part of the acquired image data in order to enable recognition of a desired object;
Generating first feature information indicating features of the image data in which a filter of the first filter set is set;
In order to further enable recognition of the desired object, a filter of a second filter set of filters each having a characteristic property is set for a part of the generated first characteristic information. Steps,
Generating the image data subjected to the image processing procedure that is the second feature information indicating the feature of the first feature information in which the filter of the second filter set is set,
In the step of recognizing, by pre-outs image processing procedure on the basis of the information of the image data subjected to some, each of which outputs a plurality of recognizers with characteristic properties output result, the image The part of the image data that has been subjected to the plurality of image processing procedures and the plurality of images until it can be recognized if it can not be recognized. An object recognition method comprising: recognizing based on information on a part newly subjected to the plurality of image processing procedures in addition to a part of the image data subjected to the processing procedure. Obtaining image data representing an image;
Performing a plurality of image processing procedures at different levels in order to recognize a desired object with respect to the acquired image data;
Based on information of the image data subjected to a plurality of image processing procedures at different levels, it is desirable that the image is reflected in the image according to output results output from a plurality of recognizers each having a characteristic property Recognizing whether or not the object is a target object,
Steps to perform the image processing procedure,
Setting each filter of a first filter set of filters each having a characteristic property for the acquired image data in order to enable recognition of a desired object;
Reducing the image in which the filter of the first filter set is set, and generating first feature information indicating features of the reduced image;
Setting a filter of a second filter set of filters each having a characteristic property for the generated first feature information to further enable recognition of the desired object;
Generating the image data subjected to the image processing procedure that is the second feature information indicating the feature of the first feature information in which the filter of the second filter set is set,
In the recognizing step, based on the information of the image data subjected to the image processing procedure , the image is reflected in the image by output results output from a plurality of recognizers each having a characteristic property Is recognized based on the information of the image data of the images sequentially enlarged by the image processing means until it can be recognized if it cannot be recognized. An object recognition method characterized by the above.

Images