Information processing apparatus, information processing method, and program

Abstract

To accelerate detection of a peak position from a score map.SOLUTION: An information processing apparatus detects a peak position from a score map. The information processing apparatus determines whether an attention position indicates the peak position, and selects a new attention position on the basis of a relationship between a score at the attention position and a score at a first position included in an area of a predetermined size around the attention position.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to an information processing device, an information processing method, and a program, and more particularly to a technique for detecting a peak position from a score map generated using a neural network. [Background technology]

[0002] In the field of information processing, a process of detecting peak positions from a score map is sometimes used. For example, a hierarchical computation method (a pattern recognition method based on deep learning technology) represented by CNN has attracted attention as an object detection method that is robust against object fluctuations. In order to perform robust object detection, peak positions of a score map output by the final layer of CNN may be extracted. This score map may indicate a score for each position. In addition, this score may indicate, for example, the likelihood that an object exists at the corresponding position. In this case, the extracted peak position may indicate an estimation result of the object's position.

[0003] Patent Document 1 and Non-Patent Document 1 disclose methods for detecting peak positions from a score map. Patent Document 1 proposes a method for extracting a pixel of interest as a peak pixel when it is determined that the value of the pixel of interest is greater than the values ​​of all surrounding pixels or equal to the largest value among the surrounding pixels. Non-Patent Document 1 also proposes a method for quickly searching for peak points by reducing the area to be matched by utilizing a color histogram and the upper limit of a similarity measure. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2005-221241 A [Non-patent literature]

[0005] [Non-Patent Document 1] IEICE Papers D-II Vol. J81-D-II No.9 pp.2035-2042 Summary of the Invention [Problem to be solved by the invention]

[0006] In recent years, the size of the score map has tended to increase, which has led to a demand for faster peak position detection processing.

[0007] According to the method of Patent Document 1, a process is performed to determine whether or not each pixel of interest is a peak pixel. Therefore, the process time required by the method of Patent Document 1 is relatively long. With the method of Patent Document 1, it is difficult to speed up the process, especially when there are many peak points in the score map. Also, with the method of Non-Patent Document 1, a part of the process is skipped based on the similarity with a previously specified image. As described above, the method of Non-Patent Document 1 requires similar images to be prepared in advance, so that the applicable cases are limited. In particular, the method of Non-Patent Document 1 is not suitable for recognition processing using machine learning.

[0008] An object of the present invention is to increase the speed of detecting peak positions from a score map. [Means for solving the problem]

[0009] An information processing device according to an embodiment of the present invention has the following configuration. An information processing device for detecting a peak position from a score map, A determination means for determining whether or not the position of interest indicates a peak position; a selection means for selecting a new position of interest based on a relationship between a score at the position of interest and a score at a first position included in an area of ​​a predetermined size around the position of interest; Equipped with. Effect of the Invention

[0010] It is possible to speed up the detection of peak positions from the score map. [Brief description of the drawings]

[0011] [Figure 1] FIG. 2 is a diagram showing an example of the hardware configuration of an information processing apparatus according to an embodiment. [Diagram 2] FIG. 13 is a diagram showing an example of a feature map. [Diagram 3] 1 is a flowchart of an information processing method according to an embodiment. [Figure 4] 1 is a flowchart of an information processing method according to an embodiment. [Diagram 5] 1 is a flowchart of an information processing method according to an embodiment. [Figure 6] 5A to 5C are diagrams for explaining an example of a peak determination method. [Figure 7] 5A to 5C are diagrams for explaining an example of a peak determination method. [Figure 8] 1 is a flowchart of an information processing method according to an embodiment. [Figure 9] 5A and 5B are diagrams for explaining a method of selecting a pixel of interest. [Figure 10] 5A and 5B are diagrams for explaining a method of selecting a pixel of interest. [Figure 11] 11A and 11B are diagrams for explaining a method of selecting a pixel of interest when reduction processing is used. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, the embodiments will be described in detail with reference to the attached drawings. Note that the following embodiments do not limit the invention according to the claims. Although the embodiments describe a number of features, not all of these features are essential to the invention, and the features may be combined in any manner. Furthermore, in the attached drawings, the same reference numbers are used for the same or similar configurations, and duplicated descriptions are omitted.

[0013] (Overall composition) An information processing device according to an embodiment detects a peak position from a score map. The information processing device according to an embodiment may be an imaging device or a smartphone. The functions of each processing unit of the information processing device shown in Figs. 1(B) and (C) can be realized using a processor. However, at least a part of the processing units may be realized by dedicated hardware. The information processing device may be composed of a plurality of information processing devices connected via a network, for example.

[0014] FIG. 1(A) illustrates an example of a hardware configuration of an information processing device according to an embodiment. The CNN processing unit 101 generates a score map. For example, the CNN processing unit 101 can generate the score map by performing an object detection process on an image to be processed. The CNN processing unit 101 can generate the score map based on the likelihood that an object exists at each position of the input image. The CNN processing unit 101 may generate a feature map other than the score map.

[0015] The image input unit 102 acquires an image to be processed (input image). The image input unit 102 may be a photoelectric conversion device. The photoelectric conversion device may have an optical system such as a lens and a sensor. The photoelectric conversion device may also have a driver circuit for controlling the sensor, an AD converter, and the like. The sensor may be, for example, a Charge-Coupled Device (CCD) or a Complimentary Metal Oxide Semiconductor (CMOS).

[0016] A central processing unit (CPU) 103 controls the entire information processing device. A read only memory (ROM) 104 can store instructions and parameters that define the operation of the CPU 103. A random access memory (RAM) 105 is a work memory used by the CPU 103 for operation. The RAM 105 can be, for example, a large-capacity dynamic random access memory (DRAM).

[0017] The user interface unit 106 acquires input from a user of the information processing device. The user interface unit 106 also outputs information to the user of the information processing device. The user interface unit 106 may include, for example, a display device that displays a processing result such as an object detection process. The user interface unit 106 may also be a button or a touch panel that receives user input. The user interface unit 106 may acquire input from the user or output information to the user via a Graphical User Interface (GUI). For example, the user may specify a task for object detection processing using the GUI.

[0018] The data bus 107 is a data transfer path between the above-mentioned components.

[0019] In this way, a processor such as the CPU 103 executes a program stored in a memory such as the ROM 104 or the RAM 105, thereby realizing the functions of each unit shown in Figures 1(B) and (C) etc. Note that the CNN processing unit 101 may be dedicated hardware. Alternatively, the functions of the CNN processing unit 101 may be realized by a processor such as the CPU 103.

[0020] In the following, an example will be mainly described in which an information processing device detects a detection target object (hereinafter simply referred to as an object) from a processing target image. In the example of FIG. 1(A), a CNN processing unit 101 generates a feature map by performing processing in a neural network (e.g., CNN) on the processing target image. The feature map is, for example, feature data that can be represented using two-dimensional coordinates. The CNN processing unit 101 can execute a specified CNN operation according to an instruction from the CPU 103. The CNN processing unit 101 stores the generated feature map in the RAM 105. Then, the CPU 103 performs various detection tasks in object detection based on the feature map generated by the CNN processing unit 101.

[0021] 2(A) to (C) show examples of feature maps output by the CNN processing unit 101. The CNN processing unit 101 performs processing in CNN to output a feature map corresponding to an image 204. In this example, the CNN processing unit 101 outputs a score map 201, a region width map 202, and a region height map 203 as feature maps. In this way, the CNN processing unit 101 can generate multiple feature maps corresponding to objects. The CNN used by the CNN processing unit 101 is trained to output these feature maps.

[0022] In the example of Fig. 2(A) to (C), the score map 201, the region width map 202, and the region height map 203 have 8 columns x 6 rows of values. In this way, each map is represented by a plurality of values ​​arranged vertically and horizontally. Hereinafter, one position on the map may be referred to as a pixel. One pixel corresponds to one position on the map. Also, one pixel corresponds to a specific position on the image 204. In this example, the positional relationship between two pixels on the map is the same as the positional relationship between two corresponding positions on the image 204.

[0023] In this embodiment, the score map 201 indicates the likelihood that a detection target object exists at each position of the image. Each value (score) of the score map 201 indicates the reliability that an object exists at the corresponding position of the image 204. Specifically, the higher the probability that an object exists at the corresponding position, the larger the value of the score map 201. Here, when the value of the score map is larger than a predetermined threshold, it can be determined that an object exists at the position on the image 204 corresponding to this value. Also, it can be determined that the center of the object exists at the position on the image 204 corresponding to the peak position of the score map (the position where the score is maximized). For example, in the example of FIG. 2(A), it can be determined that the positions 201a and 201b having a score of 255 in the score map 201 correspond to the center position of the object. As described above, one position on the score map may be called a pixel. Also, the score at one position on the score map may be called the score of the corresponding pixel.

[0024] Each value of the region width map 202 and the region height map 203 indicates an estimated value of the region width and region height of an object when the object is present at the corresponding position in the image 204. Therefore, the region width and region height (W1, H1) of the object corresponding to the position 201a, and the region width and region height (W2, H2) of the object corresponding to the position 201b can be obtained from the region width map 202 and the region height map 203. In this way, the region (position and size) of an object (e.g., a person's head) can be detected from the image 204 using the feature maps.

[0025] (Object detection processing) Next, a configuration of an information processing device that performs object detection processing according to an embodiment will be described. FIG. 1(B) shows an example of a functional configuration of an information processing device 120 that performs object detection processing. In an embodiment, the information processing device 120 is the same device as the information processing device shown in FIG. 1(A). That is, the information processing device 120 detects an object region by processing a feature map generated by the CNN processing unit 101. However, the information processing device 120 may be a device different from the information processing device shown in FIG. 1(A). In addition, the information processing device 120 may perform processing on a feature map generated by a device different from the information processing device shown in FIG. 1(A). As shown in FIG. 1(B), the information processing device 120 has a peak detection unit 121, a list processing unit 122, and a region determination unit 123.

[0026] The peak detection unit 121 detects peak positions from a score map included in the feature map. This process makes it possible to exclude positions that do not correspond to the center position of the object from subsequent processing targets. Then, the peak detection unit 121 generates a list indicating the detected peak positions (hereinafter, referred to as a peak list). The process performed by the peak detection unit 121 will be described in detail later.

[0027] The list processing unit 122 performs processing on the peak list generated by the peak detection unit 121. For example, the list processing unit 122 can rearrange the peak positions included in the peak list in score order (for example, descending order). Furthermore, the list processing unit 122 can exclude some peak positions from the peak list based on the scores shown in the score map 201. For example, the list processing unit 122 can exclude peak positions having a score equal to or less than a threshold (i.e., object positions with low reliability) from the peak list. This processing can simplify subsequent processing such as integration processing of detection results.

[0028] The region determination unit 123 determines the position and region of the object on the input image based on the peak position included in the peak list and the region width map and region height map included in the feature map. For example, the region determination unit 123 can determine the position on the input image corresponding to the peak position as the center position of the object. Furthermore, the region determination unit 123 can acquire the width and height of the object region corresponding to the peak position by referring to the region width map and the region height map. Then, the region determination unit 123 can determine the position and region of the object on the input image based on the peak position and the width and height of the object region corresponding to the peak position. At this time, the region determination unit 123 may determine the position and region of the object on the input image by performing scale conversion or linear correction on the position on the peak map and the width and height of the object region indicated by the region width map and the region height map. For example, the region determination unit 123 can calculate (W1, H1) or (W2, H2) shown in FIG. 2(A). In this way, the region determination unit 123 can determine the position of the object on the input image for the peak position.

[0029] Furthermore, the area determination unit 123 may integrate the object detection results. For example, there may be cases where an object appears large in an image, or where the score threshold used in peak detection is low. In such cases, even if there is only one object, multiple positions close to each other may be obtained as the estimated object position. In such cases, in order to narrow down the object detection result to one, the area determination unit 123 may combine multiple detection results (e.g., center positions) into one. For example, Non-Maximum Suppression (NMS) technology may be used to combine the detection results.

[0030] Next, an object detection method according to an embodiment will be described. Fig. 3 shows an example of a flowchart of an object detection task. According to this process, an object and its region are detected based on a feature map output from the final layer of the CNN processing unit 101.

[0031] In S301, the peak detection unit 121 performs a process of detecting peak positions from the score map. As described above, the peak detection unit 121 can generate a peak list indicating the detected peak positions. This process can speed up the processes of S302 to S304. That is, in S302 to S304 described later, only the processes for the peak positions stored in the peak list can be performed. Details of S301 will be described later.

[0032] In S302, list processing unit 122 performs processing on the peak list obtained in S301 as described above. In S303, area determination unit 123 determines the position and area of ​​an object on the input image for each peak position shown in the peak list as described above. In S304, area determination unit 123 performs integration processing on the object detection results obtained in S303 as described above. In this way, by the processing of S301 to S304, an object can be detected from the input image.

[0033] (Peak detection processing) As described above, in the object detection process, a process of detecting a peak position in a score map is performed. This process will be described in more detail below. Note that the target of the peak position detection process described below is not limited to the score map for object detection as described above. The following peak position detection process can be performed on any type of score map. In addition, the score map may be a two-dimensional map or a multidimensional map.

[0034] The configuration of an information processing device that performs peak detection processing according to an embodiment will be described below. FIG. 1(C) shows an example of a functional configuration of an information processing device 150 that performs peak detection processing. In an embodiment, the information processing device 150 corresponds to the peak detection unit 121 shown in FIG. 1(B). In an embodiment, the information processing device 150 may be a device different from the information processing device shown in FIG. 1(A). For example, the information processing device 150 may perform processing on a score map generated by a device different from the information processing device shown in FIG. 1(A). As shown in FIG. 1(C), the information processing device 150 has a candidate extraction unit 151, a comparison unit 152, and a peak determination unit 153.

[0035] The candidate extraction unit 151 extracts peak position candidates. The processes by the comparison unit 152 and the peak determination unit 153 are performed on a position of interest (or a part thereof) described below. The peak position candidates extracted by the candidate extraction unit 151 are later used as positions of interest. In this way, it can be said that the candidate extraction unit 151 selects positions of interest.

[0036] In this embodiment, the candidate extraction unit 151 selects a peak position candidate (i.e., a new position of interest) based on the relationship between the score at the position of interest and the score at a first position included in an area of ​​a predetermined size around the position of interest. This position of interest is a peak position candidate previously extracted by the candidate extraction unit 151. In this way, the candidate extraction unit 151 can select a new position of interest in the vicinity of the position of interest.

[0037] The comparison unit 152 compares the score at the position of interest with a threshold. For example, the comparison unit 152 can determine whether the score indicated by the score map for the position of interest is smaller than a threshold h. The comparison unit 152 can perform such processing to determine whether to exclude the position of interest from the target of the peak determination processing.

[0038] The peak determination unit 153 determines whether the position of interest indicates a peak position. Hereinafter, this process may be referred to as peak determination process. The peak determination unit 153 can determine whether the position of interest indicates a peak position according to a determination criterion. This determination criterion is that the score at the position of interest and the scores at each position included in an area of ​​a predetermined size around the position of interest satisfy a predetermined relationship. When it is determined that such a determination criterion is satisfied, the peak determination unit 153 can determine that the position of interest indicates a peak position. In this way, the peak determination unit 153 can determine a peak pixel indicating a peak position by comparing the score of the pixel of interest indicating the position of interest with the scores of its surrounding pixels.

[0039] Here, the peak determination unit 153 skips the determination of positions that have not been selected as positions of interest (peak position candidates). That is, the peak determination unit 153 can determine that positions that have not been selected as positions of interest (peak position candidates) by the candidate extraction unit 151 do not indicate peak positions. Therefore, by reducing the number of positions of interest (peak position candidates) selected by the candidate extraction unit 151, the time required for peak determination processing can be shortened.

[0040] Furthermore, the peak determination unit 153 can skip the determination for the position of interest in accordance with the determination by the comparison unit 152. For example, the peak determination unit 153 can determine whether or not the peak determination process for the position of interest indicates a peak position in response to the score at the position of interest being determined to be equal to or greater than the threshold h. On the other hand, the peak determination unit 153 can skip the determination as to whether or not the position of interest indicates a peak position in response to the score at the position of interest being determined to be smaller than the threshold h.

[0041] With this configuration, peak detection is performed so that peak positions having a small score are not detected. In this embodiment, the peak determination process for positions having a small score can also be skipped. With this configuration, the processing time can be shortened compared to when the peak determination unit 153 performs the peak determination process for all positions in the score map. However, it is not essential that the peak determination unit 153 skips the determination for the position of interest in accordance with the determination by the comparison unit 152. In addition, it is not essential that the information processing device 150 has the comparison unit 152.

[0042] 6(A) to (E) show an example of a peak pixel determination method. For peak determination processing for the target position, the peak determination unit 153 can use a 3×3 pixel neighborhood filter 601 shown in FIG. 6(A). That is, the peak pixel can be determined based on the relationship between the score of the target pixel S and the scores of the neighboring pixels of the target position. In this example, the neighboring pixels of the target pixel S are pixels S1 to S8. FIG. 6(B) shows a condition 602 regarding the relationship between the score of the target pixel S and the scores of its neighboring pixels S1 to S8, which is satisfied when the target pixel S indicates a peak position.

[0043] FIG. 6C shows a score map 603 of 6×8 pixels that is the target of the peak position detection process. In FIG. 6C, an area R in the neighborhood filter corresponding to the pixel of interest S is shown by a dotted rectangle. The condition 602 varies depending on the pixel position of the pixel of interest on the score map 603. The symbols A to E attached to each pixel in FIG. 6C correspond to the type in FIG. 6B, and indicate the condition used in the peak determination process for the pixel of this symbol. For example, the symbol A is attached to the pixel of interest S shown in FIG. 6C. Therefore, the condition of type=A shown in FIG. 6B is used in the peak determination process for the pixel of interest S. That is, the condition for determining the pixel of interest S as a peak pixel is (the score of the pixel of interest≧the score of each of the neighboring pixels S1 to S4) and (the score of the pixel of interest S>the score of each of the neighboring pixels S5 to S8). In this example, the pixels (hatched dot pixel area) on the periphery of the score map 603 are excluded from the determination of the peak pixel.

[0044] The conditions for determining the peak pixel are not particularly limited. On the other hand, by using the condition 602 shown in FIG. 6B, even if multiple pixels having the same score are adjacent, one of the multiple pixels can be extracted as a peak pixel. For example, the condition 604 shown in FIG. 6D can be used. In the condition 604, the pixel of interest S is determined to be a peak pixel when the score of the pixel of interest S is greater (>) than each of the scores of the neighboring pixels S1 to S8. On the other hand, by using the condition 602, it becomes easier to detect a peak pixel when the scores of the pixel of interest and the neighboring pixels tend to be the same (for example, when detecting a large face). In addition, the condition 605 shown in FIG. 6E can be used. In the condition 605, the pixel of interest S is determined to be a peak pixel when the score of the pixel of interest S is greater than or equal to each of the scores of the neighboring pixels S1 to S8 (≧). On the other hand, by using the condition 602, it becomes easier to suppress the detection of multiple peak pixels representing the same object when the scores of the pixel of interest and the neighboring pixels tend to be the same (for example, when detecting a large face). By reducing the number of peak pixels detected, the processing speed at the subsequent stages can be improved.

[0045] When the condition 602 is used, the condition of type=A is used for many pixels. On the other hand, even if the score of the pixel of interest is the same as the score of an adjacent pixel on the periphery of the score map 603, the conditions of types=B to E are used for pixels in the vicinity of the periphery so that the pixel of interest is extracted as a peak pixel.

[0046] Hereinafter, a peak position detection method according to an embodiment will be specifically described with reference to the flowchart of Fig. 4. For example, the process of S301 can be performed as follows.

[0047] In S401, the candidate extraction unit 151 sets a row from which the pixel of interest is selected. In the first processing of S401, the candidate extraction unit 151 selects the first row of the score map as the row from which the pixel of interest is selected. In the second and subsequent processing of S401, the candidate extraction unit 151 moves the row from which the pixel of interest is selected in the sub-scanning direction of the score map.

[0048] In S402, the candidate extraction unit 151 sets a column from which the pixel of interest is selected. In S402, immediately after the row from which the pixel of interest is selected is set in S401, the candidate extraction unit 151 selects the first column of the score map as the column from which the pixel of interest is selected. In other cases, the candidate extraction unit 151 moves the column from which the pixel of interest is selected in the main scanning direction of the score map. The amount of movement at this time is indicated by a movement amount dcol, which will be described later. In this way, in S401 and S402, the row and column from which the pixel of interest is selected are set, and the pixel of interest according to this row and column is selected. Then, the processes of S410, S408, and S411 are performed on the selected pixel of interest.

[0049] In this way, the candidate extraction unit 151 selects a plurality of attention positions located along the main scanning direction of the score map. As described later, the candidate extraction unit 151 determines the interval (movement amount dcol) between the attention position and the new attention position. Then, the candidate extraction unit 151 selects the new attention position based on this interval (movement amount dcol). As a result, the peak determination unit 153 determines whether or not each of the plurality of attention positions (at least a portion of the plurality of attention positions selected by the candidate extraction unit 151) located along the main scanning direction of the score map indicates a peak position.

[0050] However, if the pixel of interest according to the set row and column is a pixel that is not determined to be a peak pixel such as a pixel on the periphery, the candidate extraction unit 151 can further move the column in which the pixel of interest is selected in the main scanning direction of the score map. The amount of movement dcol at this time may be set to 1. Also, in S402, if the column in which the pixel of interest is selected cannot be moved in the main scanning direction of the score map (i.e., if the last column has been set), the loop of S402 ends and the process returns to S401. In this case, the candidate extraction unit 151 can also set the amount of movement dcol to 1.

[0051] S410 is a pre-processing for S411 (peak determination processing) described later. In S410, the candidate extraction unit 151 selects a peak position candidate (i.e., a new focus position). S410 includes S403 to S405. In these processes, a shift amount dcol used to determine the focus pixel in S402 is set based on a comparison result between the score of the focus pixel S and the score of its adjacent pixel S4 to the right.

[0052] In S403, the candidate extraction unit 151 compares the score at the attention position with the score at a first position included in an area of ​​a predetermined size around the attention position. This first position is located, for example, downstream of the attention position in the main scanning direction of the score map. In this example, the first position is a position adjacent to the attention position. That is, the candidate extraction unit 151 compares the score of the attention pixel S with the score of the pixel S5 adjacent to the attention pixel S on the right. However, the first position may be downstream of the position adjacent to the attention position, as in the case of combining the process of detecting a peak position and the reduction process described later.

[0053] In this embodiment, the candidate extraction unit 151 determines whether the score at the target position is greater than the score at the first position. If the candidate extraction unit 151 determines that the score of the target pixel S>the score of pixel S5, the process proceeds to S404. If the candidate extraction unit 151 determines that the score of the target pixel S≦the score of pixel S5, the process proceeds to S405.

[0054] S404 is performed in response to a determination that the score at the attention position is greater than the score at the first position. In S404, the candidate extraction unit 151 sets the shift amount dcol to 2. In this case, in S402, which is performed after the subsequent S408 and S411, the column in which the attention pixel is selected is shifted by two columns, so that the pixel S5 is not selected as the attention pixel. That is, the pixel S5 is excluded from the target of the peak determination process in S411. In other words, the candidate extraction unit 151 can select a second position (the right side of the pixel S5) different from the first position (pixel S5) as a peak position candidate. In this case, the second position is located downstream of the attention position (attention pixel S) and the first position (pixel S5) in the main scanning direction of the score map. In this example, the second position is a position adjacent to the first position. However, the second position may be downstream of a position adjacent to the first position, as in the case of using a 5×5 pixel neighborhood filter described later and in the case of combining the process of detecting the peak position with the reduction process.

[0055] S404 is performed in response to a determination that the score at the focus position is equal to or less than the score at the first position. In S405, the candidate extraction unit 151 sets the shift amount dcol to 1. In this case, in S402, which is performed after the subsequent S408 and S411, the column in which the focus pixel is selected shifts by one column, and pixel S5 is selected as the focus pixel. In this way, the candidate extraction unit 151 can select the first position (pixel S5) as a new peak position candidate.

[0056] The setting of the above-mentioned shift amount dcol will be further explained. As described above, the pixel of interest shifts in the main scanning direction. The score of the pixel of interest T being greater than the score of the pixel T+1 (i.e., pixel S5) on the right means that the score of the pixel T+1 is smaller than the pixel T (i.e., pixel S4) on the left. Therefore, the pixel T+1 is not determined to indicate a peak position according to the condition 602. Therefore, the peak determination process for the pixel T+1 on the right of the pixel of interest can be skipped. In this way, the shift amount dcol is set to 2 in order to skip the determination of the peak position for the pixel to the right of the pixel of interest. On the other hand, if the score of the pixel of interest T is equal to or less than the score of the pixel T+1 (i.e., pixel S5) on the right, there is a possibility that the pixel T+1 is determined to indicate a peak position according to the condition 602. Therefore, the shift amount dcol is set to 1 in order not to skip the determination of the peak position for the pixel T+1 on the right of the pixel of interest.

[0057] Then, the processes of S408 and S411 are performed. In this embodiment, the processes of S408 and S411 are performed on the target position. The new target position (peak position candidate) selected in S410 is used in the next loop.

[0058] In S408, the comparison unit 152 performs a determination exclusion process to determine whether or not to exclude the position of interest from the peak determination. The comparison unit 152 compares the score of the pixel of interest S with a threshold value h. If the comparison unit 152 determines that the score of the pixel of interest S is less than the threshold value h, the process returns to S402 and the next pixel of interest is selected. In other words, the process of S411 is not performed on the pixel of interest. This configuration suppresses the extraction of peak pixels having small scores.

[0059] In S411, the peak determination unit 153 performs peak determination processing on the position of interest. In this way, the peak determination unit 153 extracts the peak position. S411 includes S406 and S407. In S411, the peak determination unit 153 determines whether the pixel of interest indicates a peak position as described above. For example, the peak determination unit 153 compares the score of the pixel of interest S with the scores of its neighboring pixels S1 to S8. If all comparison results satisfy the condition 602, the peak determination unit 153 determines that the pixel of interest is a peak pixel, and the process proceeds to S407. If there is a comparison result that does not satisfy the condition 602, the pixel of interest is not determined to be a peak pixel, and the process returns to S402.

[0060] In S407, the peak determination unit 153 stores information on the pixel of interest determined to be the peak pixel in S406 in a peak list. The information on the pixel of interest may include, for example, the score of the pixel of interest and the position information of the pixel of interest on the score map. The peak list may be implemented as an array in memory.

[0061] The above process allows the generation of a peak list. Based on the determination results by peak determination unit 153, which are shown in the peak list, region determination unit 123 can estimate the position of the detection target object in the input image, as described above.

[0062] In the example of FIG. 4, the pre-processing S410, the determination exclusion process S408, and the peak determination process S411 are performed in this order. In this way, in order to improve the processing speed, the determination exclusion process S408 can be performed before the peak determination process S411. Also, in order to improve the processing speed, in a loop before the processing for the pixel of interest, the pre-processing S410 can be performed to determine whether or not to perform peak determination processing for this pixel of interest (i.e., whether or not to extract the pixel as a peak position candidate). As described above, in the pre-processing S410, it can be determined to skip the peak determination process for the next pixel. Also, in the determination exclusion process S408, it can be determined to skip the peak determination process for the pixel of interest. In this way, the peak determination process S411, which has a large processing load, can be skipped by the pre-processing S410 and the determination exclusion process S408, and therefore the processing speed is improved.

[0063] According to such a method, the processing speed can be improved compared to the case where the peak determination process is performed for all pixels. For example, when the number of objects detected from the input image is small and most of the pixels are excluded from the peak determination process by the determination exclusion process S408, the processing speed can be effectively improved.

[0064] As described above, the peak determination unit 153 can determine that the position of interest indicates a peak position when the score at the position of interest and the scores at each position included in a region of a predetermined size around the position of interest satisfy a predetermined relationship. This predetermined relationship can include a relationship determined for each position included in the region of a predetermined size. For example, the predetermined relationship can include a first relationship between the score at the position of interest (e.g., a pixel of interest S) and the score at a first position (e.g., pixel S5) included in the region of a predetermined size. In the example of FIG. 6(B), the first relationship (type=A) is score of pixel of interest S>score of pixel S5.

[0065] The candidate extraction unit 151 can select the peak position candidate (i.e., a new focus position) by, for example, the following method. In the peak determination process for the focus position, the candidate extraction unit 151 can select the first position as a peak position candidate in response to a determination that the score at the focus position and the score at the first position do not satisfy the first relationship. This corresponds to setting the movement amount dcol to 1 when the score of the focus pixel S≦the score of pixel S5 in the example of FIG. 4. In addition, in the peak determination process for the first position, the candidate extraction unit 151 can select the first position as a peak position candidate in response to a determination that the score at the first position and the score at the focus position satisfy the second relationship. Here, the above-mentioned predetermined relationship can include a second relationship between the score at the focus position (e.g., the focus pixel S) and the score at a second position (e.g., pixel S4) included in an area of ​​a predetermined size. In the example of FIG. 6(B), the second relationship is the score of the focus pixel S≧the score of pixel S4. According to condition 602 (type=A) shown in FIG. 6B, satisfying the second relationship in the peak determination process for the first position is equivalent to not satisfying the first relationship in the peak determination process for the position of interest.

[0066] Furthermore, the candidate extraction unit 151 can select a second position different from the first position as a peak position candidate in response to a determination that the score at the focus position and the score at the first position satisfy the first relationship in the peak determination process for the focus position. This corresponds to setting the movement amount dcol to 2 in the case where the score of the focus pixel S is greater than the score of the pixel S5 in the example of FIG. 4. Furthermore, the candidate extraction unit 151 can select the second position as a peak position candidate in response to a determination that the score at the first position and the score at the focus position do not satisfy the above-mentioned second relationship in the peak determination process for the first position. According to the condition 602 (type=A) shown in FIG. 6(B), not satisfying the second relationship in the peak determination process for the first position is equivalent to satisfying the first relationship in the peak determination process for the focus position.

[0067] In this way, in pre-processing S410, the candidate extraction unit 151 selects a new focus position so as to satisfy part of the criteria used in the peak determination process. In other words, the candidate extraction unit 151 performs a determination on a peak position candidate (new focus position) in advance regarding part of the conditions used in the peak determination process by the peak determination unit 153 for the focus position. Specifically, the candidate extraction unit 151 performs a determination on some of the conditions used in the peak determination process for pixels in the periphery of the focus pixel. Then, the candidate extraction unit 151 excludes positions that do not satisfy some of the conditions used in the peak determination process from the targets of the peak determination process. With this configuration, the number of peak determinations is reduced.

[0068] As described above, the candidate extraction unit 151 can select the second position as a peak position candidate without selecting the first position as a peak position candidate in response to a determination that the score at the position of interest is greater than the score at the first position. In one embodiment, the determination criterion used in the peak determination process at least requires that the score at the position of interest is equal to or greater than each of the scores at each position included in an area of ​​a predetermined size around the position of interest. As already described, the above-mentioned predetermined relationship can include a relationship determined for each position included in an area of ​​a predetermined size. Here, the relationship determined for a specific position can require that the score at the position of interest is greater than the score at the specific position or is equal to or greater than the score at the specific position. For example, in order to satisfy the condition 602, regardless of the type, it is necessary that at least the score of the pixel of interest S is equal to or greater than the scores of the other pixels (S1 to S8). Therefore, regardless of the type, if the score at the position of interest is greater than the score at the first position, the first position is not determined to indicate a peak position. Therefore, the candidate extraction unit 151 does not need to select the first position as a peak position candidate.

[0069] In an example such as the above condition 602, it is possible to determine whether or not to select the first position as a peak position candidate by performing peak determination for the position of interest (for example, comparing the score of the pixel of interest S with the score of pixel S5). In such a configuration, the candidate extraction unit 151 may select a peak position candidate (or determine the movement amount dcol) based on the score comparison result by the peak determination unit 153. As another example, the peak determination unit 153 may refer to the score comparison result performed by the candidate extraction unit 151 to select a peak candidate position based on the score at the position of interest, for peak determination for the position of interest.

[0070] On the other hand, as shown in FIG. 4, pre-processing S410 for selecting a peak candidate position based on the score at the target position can be performed before the determination exclusion process S408 for the target position. This pre-processing S410 can be performed independently of the peak determination process S411 for the target position. The determination exclusion process S408 makes it possible to omit subsequent processes for the target position. On the other hand, by performing pre-processing S410 before the determination exclusion process S408, even if the score at the target position is less than the threshold, the peak candidate position is selected in S410 (for example, the movement amount dcol is determined). Therefore, even if the score at the target position is less than the threshold, it is possible to determine whether or not to skip the peak determination process for the pixel next to the target pixel (for example, the pixel adjacent to the right).

[0071] So far, we have described a peak detection method using a 3x3 pixel neighborhood filter. However, a similar method can be adopted when a different neighborhood filter is used. In the following, we will explain the case where a 5x5 pixel neighborhood filter is used.

[0072] Fig. 7(A) shows a 5×5 pixel neighborhood filter 701 that can be used by the peak determination unit 153. Fig. 7(B) shows a condition 702 regarding the relationship between the score of a pixel of interest S and the scores of its neighborhood pixels S1 to S24, which is satisfied when the pixel of interest S indicates a peak position. Fig. 7(C) shows a score map 703 that is the target of the peak position detection process. The score map 703 is the same as the score map 603 shown in Fig. 6(C).

[0073] The peak detection process using a 5×5 pixel neighborhood filter can be performed according to the flowchart shown in Fig. 5. For example, the process of S301 can be performed as follows.

[0074] S501 and S502 are performed in the same manner as S401 and S402 shown in FIG.

[0075] S510 is pre-processing for S511 (peak determination processing) described later. S510 includes S503 to S505 and S518 to S519. In S510, a shift amount dcol is set based on a comparison result between the score of the pixel of interest S and the scores of the pixels S5 and S17 consecutive to the right of the pixel of interest S.

[0076] In S503, the candidate extraction unit 151 compares the score of the pixel of interest S with the score of its neighboring pixel S5 to the right, as in S403. If the candidate extraction unit 151 determines that the score of the pixel of interest S is greater than the score of pixel S5, the process proceeds to S518. In S518, the candidate extraction unit 151 sets the amount of movement dcol to 1. If the candidate extraction unit 151 determines in S503 that the score of the pixel of interest S is less than or equal to the score of pixel S5, the process proceeds to S505.

[0077] In S518, the candidate extraction unit 151 compares the score of the pixel of interest S with the score of pixel S17, which is two pixels to the right of the pixel of interest S, in the same manner as in S403. If the candidate extraction unit 151 determines that the score of the pixel of interest S is greater than the score of pixel S17, the process proceeds to S504. In S504, the candidate extraction unit 151 sets the shift amount dcol to 2. If the candidate extraction unit 151 determines in S518 that the score of the pixel of interest S is less than or equal to the score of pixel S5, the process proceeds to S519. In S519, the candidate extraction unit 151 sets the shift amount dcol to 3.

[0078] In this way, the candidate extraction unit 151 compares the score at each of the N consecutive positions following the focus position along the main scanning direction of the score map with the score at the focus position. Then, the candidate extraction unit 151 can determine the movement amount dcol to be N+1 in response to a determination that the score at the focus position is greater than each of the scores at the N consecutive positions.

[0079] Note that the score of the pixel of interest T being greater than the score of pixel T+2 two pixels to the right (i.e., pixel S17) means that the score of pixel T+2 is smaller than the score of pixel T two pixels to the left (i.e., pixel S16). Therefore, pixel T+2 is not determined to indicate a peak position according to condition 702. Therefore, the peak determination process for pixel T+2, which is adjacent to the pixel of interest to the right, can be skipped. In S519, the determination in S503 indicates that pixel T+1 is not determined to indicate a peak position according to condition 702, so the shift amount dcol can be set to 3.

[0080] S508 is performed in the same manner as S408 shown in Fig. 4. S511 is also performed in the same manner as S411 shown in Fig. 4, except that the neighborhood filter 701 and the condition 702 are used.

[0081] In this way, by using a larger neighborhood filter, the number of pixels that can be skipped in the peak determination process (that is, the amount of movement dcol) can be increased, and the peak position can be detected even faster.

[0082] In the above embodiment, the score of the pixel of interest S is compared with the score of the pixel to the right (e.g., S5) to determine whether or not to skip the peak determination process for the pixel to the right. On the other hand, when the main scanning direction is the vertical direction of the score map, the score of the pixel of interest is compared with the score of the pixel below it to determine whether or not to skip the peak determination process for the pixel below it.

[0083] 3, a process S302 for the peak list is performed separately from the process S301 for detecting the peak position. However, a sorting process may be performed in S407. In this case, the sorting process in S302 can be omitted.

[0084] Moreover, the process of detecting the peak position and the reduction process can be combined. For example, the peak position can be detected from a plurality of positions selected from the score map to have an interval M between them along the main scanning direction. A case where the reduction is performed by 1 / 2 in the main scanning direction (M=2) will be described with reference to FIG. 11. FIG. 11 shows a score map 1101. In addition to the pixels (dark hatching) on ​​the outer periphery that are excluded from the determination target of the peak pixel, the score map 1101 shows pixels (light hatching) that are not referenced for the reduction process. When the pixel of interest S is pixel X, the pixel S5 to the right of the pixel of interest S is pixel X+2, and the next pixel in the main scanning direction is pixel X+4. Therefore, the shift amount dcol is set to 4 in S404. Also, the shift amount dcol is set to 2 in S405.

[0085] In general, when performing a 1 / M reduction process, the candidate extraction unit 151 can set the amount of movement dcol to 2×M in response to a determination that the score at the position of interest is greater than the score of the position following the position of interest. In general, when performing a 1 / M reduction process, the candidate extraction unit 151 can set the amount of movement dcol to M×(N+1) in response to a determination that the score at the position of interest is greater than each of the scores at N positions following the position of interest. Here, the positions following the position of interest are included in a plurality of positions selected from the score map to have an interval M between each other along the main scanning direction. According to such an embodiment, peak positions can be detected more quickly, especially from a score map in which many peak positions exist.

[0086] (Modification of Peak Detection Process) In the above embodiment, pre-processing S410 was performed in units of one pixel. Specifically, a new peak position candidate was selected in the vicinity of the target position by setting a movement amount in the main scanning direction. In the following example, pre-processing is performed in units of 2×2 pixels. Then, one peak pixel candidate is selected from a 2×2 pixel block. In this case, the 2×2 pixel block to be processed is moved by a fixed movement amount (i.e., two pixels in the main scanning direction and two rows in the sub-scanning direction).

[0087] The peak detection process in this modified example can be performed according to the flowchart shown in Fig. 8. In S801 and S802, the candidate extraction unit 151 sets a 2 x 2 pixel block to be processed. While repeating the process, the candidate extraction unit 151 can move the pixel block.

[0088] For example, in S801, the candidate extraction unit 151 moves the pixel block in the sub-scanning direction. The candidate extraction unit 151 can move the pixel block by two rows. As shown in FIG. 9, the candidate extraction unit 151 can move the pixel block from line A to line B. Note that the score map 603 shown in FIG. 9 is the same as that shown in FIG. 6. Also, in S802, the candidate extraction unit 151 moves the pixel block in the main scanning direction. The candidate extraction unit 151 can move the pixel block by two pixels. As shown in FIG. 9, the candidate extraction unit 151 can move the pixel block from position 901 to position 902.

[0089] The pre-processing S810 includes S821 and S822. In S810, the candidate extraction unit 151 can select one or more peak position candidates (i.e., positions of interest) from the fixed-size region set in S801 and S802. In the following example, the candidate extraction unit 151 selects a peak position candidate from a 2×2 pixel block. In this modification, the candidate extraction unit 151 can also select a peak position candidate using the condition 602 shown in FIG. 6, which is used in the peak determination process. The neighborhood filter used in this example is the neighborhood filter 601 shown in FIG. 6.

[0090] In S821, the candidate extraction unit 151 obtains the score of each pixel included in the 2×2 pixel block.

[0091] In S822, the candidate extraction unit 151 extracts peak position candidates from the 2×2 pixel block based on the score acquired in S821. The candidate extraction unit 151 can determine the peak position candidates according to the flowchart shown in FIG. 10(B).

[0092] In S1011, the candidate extraction unit 151 determines whether to select the top left pixel of the 2×2 pixel block as a peak position candidate. Here, the candidate extraction unit 151 can select a peak position candidate (i.e., a position of interest) so as to satisfy some of the determination criteria used in the peak determination process. For example, the candidate extraction unit 151 can make this determination by comparing the score of the top left pixel of the 2×2 pixel block with the scores of each of the other pixels in the 2×2 pixel block. The candidate extraction unit 151 can make this determination in accordance with condition 602.

[0093] The process in S1011 will be described with reference to condition 1003A shown in Fig. 10(A). Conditions 1003A to 1003D indicate conditions when type = A. Condition 1003A indicates the pixel positions of other pixels when neighborhood filter 601 is applied to the top left pixel. That is, when the top left pixel is the pixel of interest S, the top right, bottom left, and bottom right pixels are pixels S5, S7, and S8, respectively.

[0094] Moreover, condition 1003A indicates a condition regarding the relationship between the score of the top left pixel and the scores of each of the other pixels for determining the top left pixel as a peak position candidate. According to condition 1003A, the condition for determining the top left pixel as a peak position candidate is the score of the top left pixel>the scores of the top right pixel, the bottom left pixel, and the bottom right pixel. This condition 1003A is a part of condition 602 used in the peak determination process. For example, according to condition 602, the condition for determining the target pixel S as a peak position candidate is the score of the top left pixel (target pixel S)>the scores of the top right pixel (pixel S5), the bottom left pixel (pixel S7), and the bottom right pixel (pixel S8). When this condition is satisfied, the scores of the top right pixel (pixel S5), the bottom left pixel (pixel S7), and the bottom right pixel (pixel S8) are smaller than the score of the top left pixel (target pixel S). Therefore, the top right pixel, the bottom left pixel, and the bottom right pixel are not determined as peak pixels. Therefore, the top right pixel, bottom left pixel, and bottom right pixel can be excluded from the peak position candidates. Therefore, the candidate extraction unit 151 can select the top left pixel as the peak position candidate. Conversely, if the condition 1003A is not satisfied, the top left pixel is not determined as the peak pixel. Therefore, the candidate extraction unit 151 can exclude the top left pixel from the peak position candidates.

[0095] On the other hand, if the condition 1003A is not satisfied, the process proceeds to S1012. In S1012, the candidate extraction unit 151 determines whether or not to select the top right pixel of the 2×2 pixel block as a peak position candidate. If the condition 1003B is satisfied, the candidate extraction unit 151 can select the top right pixel as a peak position candidate. This condition 1003B is also a part of the condition 602 used in the peak determination process. The condition 1003B is satisfied when the score of the top right pixel>the score of the bottom left pixel, the score of the top right pixel>the score of the bottom right pixel, and the score of the top right pixel≧the score of the top left pixel. In this case, the scores of the bottom left pixel and the bottom right pixel are smaller than the score of the top right pixel, so they are not determined as peak pixels. Also, in S1011, the top left pixel has already been removed from the peak pixel candidate. Therefore, only the top right pixel can be determined as a peak pixel in the 2×2 pixel block. Therefore, the candidate extraction unit 151 can select the top right pixel as a peak position candidate. Conversely, if the condition 1003B is not satisfied, the upper right pixel is not determined to be a peak pixel, and the candidate extraction unit 151 can therefore exclude the upper right pixel from the peak position candidates.

[0096] Furthermore, if condition 1003B is not satisfied, the process proceeds to S1013. In S1012, the candidate extraction unit 151 determines whether or not to select the bottom left pixel of the 2×2 pixel block as a peak position candidate. If condition 1003C is satisfied, the candidate extraction unit 151 can select the top right pixel as a peak position candidate. This condition 1003C is also part of the condition 602 used in the peak determination process. The condition 1003C is satisfied when the score of the bottom left pixel>the score of the bottom right pixel, the score of the bottom left pixel≧the score of the top left pixel, and the score of the bottom left pixel≧the score of the top right pixel. In this case, the score of the bottom right pixel is smaller than the score of the bottom left pixel, so it is not determined as a peak pixel. Also, in S1011 and S1012, the top left pixel and the top right pixel have already been removed from the peak pixel candidates. For this reason, only the bottom left pixel can be determined as a peak pixel in the 2×2 pixel block. Therefore, the candidate extraction unit 151 can select the bottom left pixel as a peak position candidate. Conversely, if the condition 1003C is not satisfied, the bottom left pixel is not determined to be a peak pixel. Therefore, the candidate extraction unit 151 can exclude the bottom left pixel from the peak position candidates.

[0097] Furthermore, if the condition 1003C is not satisfied, the process proceeds to S1014. In S1012, the candidate extraction unit 151 selects the bottom right pixel of the 2×2 pixel block as a peak position candidate. Since the top left pixel, the top right pixel, and the bottom left pixel are already excluded from the peak pixel candidates, the candidate extraction unit 151 can select the bottom right pixel as the peak position candidate. In this case, it is not necessary to determine whether the condition 1003D is satisfied. Note that the condition 1003D is part of the condition 602 used in the peak determination process, and is a necessary condition for the bottom right pixel to be determined as a peak pixel. On the other hand, if the conditions 1003A to C are not satisfied, the condition 1003D is satisfied.

[0098] In this way, when the condition 602 is used, the candidate extraction unit 151 can select one pixel that can be determined as a peak pixel from the 2×2 pixel block as a peak position candidate. In this case, it is confirmed that the three pixels not selected as the peak position candidate are not determined as peak pixels. That is, by performing the peak determination process on the peak position candidate, the peak extraction for the 2×2 pixel block is completed. Therefore, it is not necessary to change the amount of movement in S801 and S802. That is, a fixed amount of movement dcol=2 can be used. As described above, the candidate extraction unit 151 can select a peak position candidate (i.e., a position of interest) according to the conditions 1003A to 1003D that correspond to each position in a fixed-size area and that define a part of the condition 602 used for peak determination.

[0099] The processes of S808 and S811 are performed in the same manner as S408 and S411 shown in FIG. 4. However, the processes of S808 and S811 are performed on the peak position candidate (pixel of interest) selected in S810. The order of S810 and S808 may be reversed. In this case, in S808, the comparison unit 152 may compare the maximum score of the 2×2 pixels with a threshold value h. If the maximum score of the 2×2 pixels is smaller than the threshold value h, the processes of S810 and S811 for this 2×2 pixel block can be omitted. With this configuration as well, it is possible to skip peak determination processes for some pixels for each 2×2 pixel block.

[0100] According to this modification, the number of times the peak determination process is performed can be reduced by selecting peak position candidates in units of 2×2 pixels, thereby speeding up the detection of peak positions.

[0101] (Other Examples) The present invention can also be realized by a process in which a program for implementing one or more of the functions of the above-described embodiments is supplied to a system or device via a network or a storage medium, and one or more processors in a computer of the system or device read and execute the program. The present invention can also be realized by a circuit (e.g., ASIC) that implements one or more of the functions.

[0102] The disclosure of this specification includes the following information processing device, information processing method, and program. (Item 1) An information processing device for detecting a peak position from a score map, A determination means for determining whether or not the position of interest indicates a peak position; a selection means for selecting a new position of interest based on a relationship between a score at the position of interest and a score at a first position included in an area of ​​a predetermined size around the position of interest; An information processing device comprising: (Item 2) 2. The information processing device according to item 1, wherein the selection means selects the first position as the new focus position when it is determined that the score at the focus position is equal to or lower than the score at the first position. (Item 3) 3. The information processing device according to any one of items 1 to 2, wherein the first position is located downstream of the focus position in a main scanning direction of the score map. (Item 4) The information processing device according to any one of items 1 to 3, characterized in that the selection means selects a second position different from the first position as the new focus position in response to a determination that the score at the focus position is greater than the score at the first position. (Item 5) 5. The information processing device according to item 4, characterized in that the first position is located downstream of the focus position in the main scanning direction of the score map, and the second position is located downstream of the first position in the main scanning direction of the score map. (Item 6) 6. The information processing device according to any one of items 1 to 5, wherein the determining means determines that a position not selected as the target position by the selecting means does not indicate a peak position. (Item 7) The determining means determines whether the target position indicates a peak position according to a determination criterion; 7. The information processing device according to any one of items 1 to 6, wherein the judgment criterion is that the score at the focus position and the score at each position included in the area of ​​the specified size around the focus position satisfy a specified relationship. (Item 8) the predetermined relationship includes a first relationship between a score at the target position and a score at the first position included in the region of the predetermined size; 8. The information processing device according to item 7, wherein the selection means selects the first position as a new focus position when it is determined that the score at the focus position and the score at the first position do not satisfy the first relationship. (Item 9) Item 9. The information processing device according to item 8, characterized in that when it is determined that the score at the focus position and the score at the first position satisfy the first relationship, the selection means selects a second position different from the first position as the new focus position. (Item 10) 10. The information processing device according to item 9, wherein the first relationship is that the score at the target position is greater than the score at the first position. (Item 11) The information processing device according to any one of items 9 to 10, characterized in that the first position is located downstream of the focus position in the main scanning direction of the score map, and the second position is located downstream of the first position in the main scanning direction of the score map. (Item 12) the determining means determines whether or not each of a plurality of focus positions located along a main scanning direction of the score map indicates a peak position; 12. The information processing device according to any one of items 1 to 11, wherein the selection means determines an interval between the focus position and a new focus position, and selects the new focus position based on the interval. (Item 13) The information processing device described in item 12, characterized in that the selection means determines the interval to be N+1 when it is determined that the score at the focus position is greater than each of the scores at N consecutive positions following the focus position along the main scanning direction of the score map. (Item 14) the information processing device detects peak positions from among a plurality of positions selected from the score map at an interval M from each other along a main scanning direction; The information processing device described in item 12, characterized in that when it is determined that the score at the focus position is greater than each of the scores at N positions following the focus position among the multiple positions, the selection means determines the interval to be M×(N+1). (Item 15) A comparison means for comparing the score at the position of interest with a threshold value is further provided, 15. The information processing device according to any one of items 1 to 14, wherein the determination means determines whether the focus position indicates a peak position when it is determined that the score at the focus position is equal to or greater than a threshold. (Item 16) the score map indicates the likelihood that a detection target object exists at each position of the image; 16. The information processing device according to any one of items 1 to 15, further comprising an estimation unit that estimates a position of the detection target object in the image based on a determination result by the determination unit. (Item 17) An information processing device for detecting a peak position from a score map, A selection means for selecting a position of interest; A comparison means for comparing the score at the position of interest with a threshold value; a determination means for determining whether the position of interest indicates a peak position in accordance with a determination criterion in response to the score at the position of interest being determined to be equal to or greater than a threshold value, the determination criteria include a relationship between the score at the focus position and each of the positions included in an area of ​​a predetermined size around the focus position, the relationship being determined for each of the positions; The information processing apparatus is characterized in that the selection means selects the focus position so as to satisfy a part of the judgment criteria. (Item 18) Item 18. The information processing device according to item 17, wherein the selection means selects one of the focus positions from an area of ​​a fixed size. (Item 19) A generating means for generating the score map based on a likelihood that an object exists at each position of an input image; a determining means for determining a position of the object on the input image with respect to the peak position; 19. The information processing device according to any one of items 1 to 18, further comprising: (Item 20) An information processing method for detecting a peak position from a score map, comprising: determining whether the position of interest indicates a peak position; selecting a new position of interest based on a relationship between the score at the position of interest and a score at a first position included in an area of ​​a predetermined size around the position of interest; 13. An information processing method comprising: (Item 21) An information processing method for detecting a peak position from a score map, comprising: selecting a position of interest; comparing the score at the position of interest with a threshold; and determining whether or not the position of interest indicates a peak position in accordance with a determination criterion in response to the score at the position of interest being determined to be equal to or greater than a threshold value, the determination criteria include a relationship between the score at the focus position and each of the positions included in an area of ​​a predetermined size around the focus position, the relationship being determined for each of the positions; 20. A method for processing information, comprising the steps of: selecting the focus position so as to satisfy a portion of the criteria; (Item 22) 20. A program for causing a computer to function as the information processing device according to any one of items 1 to 19.

[0103] The invention is not limited to the above-described embodiments, and various modifications and variations are possible without departing from the spirit and scope of the invention. Accordingly, the following claims are appended to apprise the public of the scope of the invention. [Explanation of symbols]

[0104] 150: information processing device, 151: candidate extraction unit, 152: comparison unit, 153: peak determination unit

Claims

1. An information processing device for detecting a peak position from a score map, A determination means for determining whether or not the position of interest indicates a peak position; a selection means for selecting a new focus position based on a relationship between a score at the focus position and a score at a first position included in an area of ​​a predetermined size around the focus position; An information processing device comprising:

2. 2 . The information processing apparatus according to claim 1 , wherein said selection means, when it is determined that the score at said focus position is equal to or lower than the score at said first position, selects said first position as said new focus position.

3. The information processing apparatus according to claim 1 , wherein the first position is located downstream of the target position in a main scanning direction of the score map.

4. 3. The information processing apparatus according to claim 2, wherein said selection means, when it is determined that the score at said focus position is greater than the score at said first position, selects a second position different from said first position as said new focus position.

5. 5. The information processing device according to claim 4, wherein the first position is located downstream of the focus position in the main scanning direction of the score map, and the second position is located downstream of the first position in the main scanning direction of the score map.

6. 2. The information processing apparatus according to claim 1, wherein said determining means determines that a position not selected as said target position by said selecting means does not represent a peak position.

7. The determining means determines whether the target position indicates a peak position according to a determination criterion; The information processing apparatus according to claim 1 , wherein the determination criterion is that a score at the focus position and a score at each position included in the area of ​​the predetermined size around the focus position satisfy a predetermined relationship.

8. the predetermined relationship includes a first relationship between a score at the target position and a score at the first position included in the region of the predetermined size; 8. The information processing apparatus according to claim 7, wherein said selection means, when it is determined that the score at said focus position and the score at said first position do not satisfy said first relationship, selects said first position as a new focus position.

9. 9. The information processing device according to claim 8, characterized in that, when it is determined that the score at the focus position and the score at the first position satisfy the first relationship, the selection means selects a second position different from the first position as the new focus position.

10. The information processing apparatus according to claim 8 , wherein the first relationship is that a score at the target position is greater than a score at the first position.

11. 10. The information processing device according to claim 9, wherein the first position is located downstream of the focus position in the main scanning direction of the score map, and the second position is located downstream of the first position in the main scanning direction of the score map.

12. the determining means determines whether or not each of a plurality of focus positions located along a main scanning direction of the score map indicates a peak position; 2. The information processing apparatus according to claim 1, wherein said selection means determines an interval between said focus position and a new focus position, and selects said new focus position based on said interval.

13. 13. The information processing device according to claim 12, characterized in that the selection means determines the interval to be N+1 when it is determined that the score at the target position is greater than each of the scores at N consecutive positions following the target position along the main scanning direction of the score map.

14. the information processing device detects peak positions from among a plurality of positions selected from the score map to have an interval M between each other along a main scanning direction; 13. The information processing device according to claim 12, characterized in that the selection means, when it is determined that the score at the target position is greater than each of the scores at N positions following the target position among the plurality of positions, determines the interval to be M×(N+1).

15. A comparison means for comparing the score at the position of interest with a threshold value is further provided, The information processing apparatus according to claim 1 , wherein the determining means determines whether or not the focus position indicates a peak position when it is determined that the score at the focus position is equal to or greater than a threshold value.

16. the score map indicates the likelihood that a detection target object exists at each position of the image; The information processing apparatus according to claim 1 , further comprising an estimation unit that estimates a position of the detection target object in the image based on a result of the determination by the determination unit.

17. An information processing device for detecting a peak position from a score map, A selection means for selecting a position of interest; A comparison means for comparing the score at the position of interest with a threshold value; a determination means for determining whether the position of interest indicates a peak position in accordance with a determination criterion in response to the score at the position of interest being determined to be equal to or greater than a threshold value, the determination criteria include a relationship between the score at the focus position and each of the positions included in an area of ​​a predetermined size around the focus position, the relationship being determined for each of the positions; The information processing apparatus is characterized in that the selection means selects the focus position so as to satisfy a part of the judgment criteria.

18. 18. The information processing apparatus according to claim 17, wherein said selection means selects one of said focus positions from an area of ​​a fixed size.

19. A generating means for generating the score map based on a likelihood that an object exists at each position of an input image; a determining means for determining a position of the object on the input image with respect to the peak position; The information processing device according to claim 1 , further comprising:

20. An information processing method for detecting a peak position from a score map, comprising: determining whether the position of interest indicates a peak position; selecting a new focus position based on a relationship between the score at the focus position and a score at a first position included in an area of ​​a predetermined size around the focus position; 13. An information processing method comprising:

21. An information processing method for detecting a peak position from a score map, comprising: selecting a position of interest; comparing the score at the position of interest with a threshold; and determining whether or not the position of interest indicates a peak position in accordance with a determination criterion in response to the score at the position of interest being determined to be equal to or greater than a threshold value, the determination criteria include a relationship between the score at the focus position and each of the positions included in an area of ​​a predetermined size around the focus position, the relationship being determined for each of the positions; 20. A method for processing information, comprising the steps of: selecting the focus position so as to satisfy a portion of the criteria;

22. A program for causing a computer to function as the information processing device according to any one of claims 1 to 19.

Images