Information processing apparatus, information processing method, and computer program product

Abstract

An information processing apparatus includes a detection unit that detects a feature region having a prescribed feature from a captured image, and a division unit that divides the feature region using an inertia principal axis of the feature region.

Description

Technical Field

[0001] This invention relates to a technique for segmenting regions detected from captured images. Background Technology

[0002] When detecting defined objects such as human bodies from captured images, sometimes regions wider than the defined object are detected, making it difficult to accurately detect the defined object. Here, we consider detecting regions of motion and identifying whether the objects contained within those regions are human bodies. In this case, sometimes regions containing both the object and its shadow are detected. If a region containing both the object and its shadow is detected, the shadow's influence makes it difficult to accurately identify whether the object is a human body.

[0003] Patent Document 1 discloses a technique for removing shadows from a background difference image (an image containing an object and its shadow) based on a predetermined reference brightness value.

[0004] Existing technical documents

[0005] Patent documents

[0006] Patent Document 1: Japanese Patent Application Publication No. 2008-245063 Summary of the Invention

[0007] The problem that the invention aims to solve

[0008] However, when taking photos in dark environments or against the light, the brightness value of the object is close to the brightness value of the shadow, so it is impossible to use the technology disclosed in Patent Document 1 to remove the shadow with high precision (at least a part of the object will be removed). In other words, even if the technology disclosed in Patent Document 1 is used, it is impossible to separate (segment) the area containing the object and its shadow into the area of ​​the object and the area of ​​the shadow with high precision.

[0009] The present invention was made in view of the above-mentioned actual situation, and its object is to provide a technique that can properly segment regions detected from captured images.

[0010] Methods for solving problems

[0011] To achieve the above objectives, the present invention employs the following structure.

[0012] A first aspect of the present invention provides an information processing apparatus, characterized in that it comprises: a detection unit for detecting feature regions having predetermined features from an image being captured; and a segmentation unit for segmenting the feature regions using the principal axis of inertia of the feature regions.

[0013] In most cases, the two objects that need to be separated (divided) extend in different directions. For example, when a person is standing on the ground, the person extends in a direction roughly perpendicular to the ground, while the shadow extends along the ground. Therefore, in the structure described, the feature region is divided using the principal axis of inertia of the detected feature region. This allows the feature region to be appropriately divided (in a way that two segmented regions corresponding to the two desired objects are obtained respectively).

[0014] Since the feature region has two principal axes of inertia, it is impossible to properly divide the feature region based solely on the principal axis of inertia used. Therefore, it can also be configured such that, when the centroid of the feature region is located in the first or third quadrant of the four quadrants obtained by the horizontal and vertical axes passing through the center of the circumscribed rectangle of the feature region, the segmentation unit uses the principal axis of inertia with the smaller angle relative to the horizontal axis to segment the feature region; and when the centroid of the feature region is located in the second or fourth quadrant of the four quadrants, the principal axis of inertia with the larger angle relative to the horizontal axis to segment the feature region. This allows for more appropriate segmentation of the feature region (and enables more reliable and appropriate segmentation of the feature region).

[0015] There are cases where the feature region is identical to the region of an object that should not be segmented, such as when the principal axis of inertia is detected. Furthermore, in such cases, there is a tendency for the feature region to occupy a large proportion of its bounding rectangle. Therefore, it can be configured to further include a judgment unit that determines whether to use the feature region as a segmentation unit to obtain two segmented regions, or to use the feature region itself. If the evaluation value corresponding to the proportion of the feature region to its bounding rectangle is less than a predetermined threshold, the judgment unit determines to use the two segmented regions; if the evaluation value is above the predetermined threshold, the judgment unit determines to use the feature region itself. This allows for more appropriate segmentation of the feature region (achieving high precision in using the segmentation result when the feature region should be segmented, and using the feature region (without using the segmentation result) when the feature region should not be segmented).

[0016] Alternatively, the evaluation value can be set as the ratio of the total area of ​​the bounding rectangles of the two segmented regions to the area of ​​the bounding rectangle of the feature region. Alternatively, the evaluation value can be set as the ratio of the area of ​​the logical sum of the bounding rectangles of the two segmented regions to the area of ​​the bounding rectangle of the feature region.

[0017] Alternatively, it can be configured to also include a recognition unit, which identifies whether an object contained within the bounding rectangle of the adopted region is a specified object. When using the two segmented regions, the recognition unit identifies, for each of the two segmented regions, whether the object contained within the bounding rectangle of that segmented region is a specified object. This allows for high-precision identification (detection) of specified objects. For example, when a specified object is contained within a feature region, segmentation of the feature region yields a segmented region that is closer to the specified object. Therefore, compared to using the bounding rectangle of the feature region, it is possible to identify (detect) with higher precision that the object contained within the bounding rectangle of the obtained segmented region is a specified object.

[0018] Alternatively, the specified object can be set to a human body. This allows for high-precision identification (detection) of the human body.

[0019] Alternatively, the detection unit can be configured to detect regions with motion as the feature regions. Alternatively, the detection unit can be configured to detect regions with pixel values ​​within a specified range as the feature regions. Alternatively, the detection unit can be configured to detect regions surrounded by edges as the feature regions.

[0020] A second aspect of the present invention provides an information processing method, characterized by comprising: a detection step of detecting a feature region having predetermined features from an image being captured; and a segmentation step of segmenting the feature region using the principal axis of inertia of the feature region.

[0021] Furthermore, the present invention can be understood as an information processing system, human body detection device, human body detection system, etc., having at least a part of the above-described structure or function. Additionally, the present invention can also be understood as an information processing method, human body detection method, control method of an information processing system, or control method of a human body detection system, a program for causing a computer to execute these methods, or a computer-readable recording medium that non-temporarily records such a program, including at least a part of the above-described processing. The above-described structures and processes can be combined with each other to constitute the present invention as long as they do not create technical contradictions.

[0022] Invention Effects

[0023] According to the present invention, regions detected from captured images can be appropriately segmented. Attached Figure Description

[0024] Figure 1 This is a block diagram illustrating a structural example of an information processing apparatus to which the present invention is applied.

[0025] Figure 2 This is a block diagram illustrating a structural example of the camera (information processing device) according to Embodiment 1 of the present invention.

[0026] Figure 3 This is a flowchart illustrating an example of the camera processing flow according to Embodiment 1 of the present invention.

[0027] Figure 4 This is a diagram illustrating a specific example of the operation of a camera according to Embodiment 1 of the present invention.

[0028] Figure 5 This is a diagram illustrating a specific example of the operation of a camera according to Embodiment 1 of the present invention.

[0029] Figure 6 This is a diagram illustrating a display example according to Embodiment 1 of the present invention.

[0030] Figure 7 This is a block diagram illustrating a structural example of the camera (information processing device) according to Embodiment 2 of the present invention.

[0031] Figure 8 This is a flowchart illustrating an example of the camera processing flow according to Embodiment 2 of the present invention.

[0032] Figure 9 This is a diagram illustrating a specific example of the operation of the camera according to Embodiment 2 of the present invention.

[0033] Figure 10 (A) ~ Figure 10 (C) is a diagram representing an example of region segmentation. Detailed Implementation

[0034] <Application Example>

[0035] Application examples of the present invention will be described.

[0036] In existing technologies, when detecting defined objects such as human bodies from captured images, the detected area is sometimes wider than the defined object, which sometimes prevents high-precision detection of the defined object. Here, we consider detecting areas of motion and identifying whether the object contained within the detected area is a human body. In this case, sometimes an area containing both the object and its shadow is detected. If an area containing both the object and its shadow is detected, the shadow's influence makes it difficult to accurately identify whether the object is a human body.

[0037] Existing techniques based on brightness value segmentation of detected regions are also proposed, but these techniques cannot properly segment regions. For example, in dark environments or when shooting against the light, the brightness value of an object is close to the brightness value of its shadow, so it is impossible to accurately separate (segment) the region containing the object and its shadow into the region of the object and the region of the shadow.

[0038] Figure 1 This is a block diagram illustrating a structural example of the information processing apparatus 100 to which the present invention is applied. The information processing apparatus 100 includes a region detection unit 101 and a region segmentation unit 102. The region detection unit 101 detects feature regions having predetermined characteristics from an image being captured. The region segmentation unit 102 segments the feature regions using the principal axis of inertia of the feature regions detected by the region detection unit 101. The region detection unit 101 is an example of the detection unit of the present invention, and the region segmentation unit 102 is an example of the segmentation unit of the present invention. Feature regions are, for example, regions where motion exists, regions having pixel values ​​(RGB values, brightness values, etc.) within a predetermined range, regions surrounded by edges, etc. That is, the predetermined characteristics are, "motion exists in the region", "the region has pixel values ​​within a predetermined range", "the region is surrounded by edges", etc. The predetermined range is, for example, a range of pixel values ​​that a predetermined object such as a human body can take. Since the imaging environment (e.g., light from the sun or lighting) depends on time, the predetermined range can also vary according to time.

[0039] In most cases, the two objects that need to be separated (divided) extend in different directions. For example, when a person is standing on the ground, the person extends in a direction roughly perpendicular to the ground, while the shadow extends along the ground. Therefore, in the above structure, the feature region is segmented using the principal axis of inertia of the detected feature region. This allows the feature region to be appropriately segmented (in a way that two segmented regions corresponding to the two desired objects are obtained respectively).

[0040] <Implementation Method 1>

[0041] Embodiment 1 of the present invention will be described.

[0042] Figure 2 This is a block diagram showing a structural example of the camera 200 (information processing device) according to Embodiment 1. The camera 200 includes an imaging unit 210, a control unit 220, a storage unit 230, and a display unit 240.

[0043] Furthermore, while Embodiment 1 described an example of applying the present invention to a camera, the information processing device to which the present invention is applied is not limited to a camera. For example, the present invention can also be applied to a personal computer (PC) separate from the camera (camera unit 210). The display unit 240 can be a display device (monitor) separate from the information processing device to which the present invention is applied. When the information processing device to which the present invention is applied is separate from the camera, the location where the information processing device is installed is not particularly limited. For example, the information processing device can be installed in the same room as the camera, or it can be installed in a different room. The aforementioned PC can be a computer in the cloud, or it can be a smartphone or tablet terminal, or various other terminals.

[0044] The camera unit 210 captures images and outputs them to the control unit 220. In Embodiment 1, the camera unit 210 captures moving images. The camera unit 210 sequentially captures frames of moving images and outputs them to the control unit 220. Alternatively, the camera unit 210 may also sequentially capture still images and output them to the control unit 220.

[0045] The control unit 220 includes a CPU (Central Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), etc., and performs control of various structural elements and various information processing.

[0046] Storage unit 230 stores programs executed by control unit 220, various data used by control unit 220, etc. For example, storage unit 230 is an auxiliary storage device such as hard disk drive or solid-state drive.

[0047] The display unit 240 displays information based on the information output from the control unit 220.

[0048] The control unit 220 will be described in more detail. The control unit 220 includes a region detection unit 221, an inertial axis calculation unit 222, a region segmentation unit 223, a region selection unit 224, and a human body recognition unit 225.

[0049] The region detection unit 221 detects feature regions with predetermined characteristics from the image captured by the imaging unit 210. Then, the region detection unit 221 outputs the image captured by the imaging unit 210 and information representing the detected feature regions to the inertial axis calculation unit 222. Feature regions include, for example, regions where motion exists, regions with pixel values ​​(RGB values, brightness values, etc.) within a predetermined range, and regions surrounded by edges. That is, the predetermined characteristics are, "motion exists in the region," "pixel values ​​are within a predetermined range," and "the region is surrounded by edges," etc. The predetermined range is, for example, a range of pixel values ​​that can be taken by a predetermined object (human body in Embodiment 1). Since the imaging environment (e.g., light from the sun, illumination, etc.) depends on time, the predetermined range can also vary according to time. The region detection unit 221 is an example of the detection unit of the present invention.

[0050] Furthermore, the method for detecting feature regions (methods for determining motion, edge detection, etc.) is not particularly limited, and various proposed methods can be used in feature region detection. Motion can be determined using either background subtraction or inter-frame subtraction. Background subtraction, for example, detects pixels in the captured image whose absolute difference (difference) with a predetermined background image is greater than or equal to a predetermined threshold as pixels exhibiting motion. Inter-frame subtraction, for example, detects pixels in the current captured image (current frame) whose pixel value difference with a past captured image (past frame) is greater than or equal to a predetermined threshold as pixels exhibiting motion. In inter-frame subtraction, for example, the past frame is a frame preceding the current frame by a predetermined number of frames, and this predetermined number is 1 or more. The predetermined number (the number of frames from the current frame to the past frames) can also be determined based on the frame rate processed by the control unit 220, the frame rate of the camera unit 210, etc.

[0051] The principal axis of inertia calculation unit 222 calculates the principal axis of inertia of the feature regions based on information representing the feature regions detected by the region detection unit 221. Furthermore, the principal axis of inertia calculation unit 222 outputs to the region segmentation unit 223 the image captured by the imaging unit 210, information representing the feature regions detected by the region detection unit 221, and information representing the calculated principal axis of inertia. The method for calculating the principal axis of inertia is not particularly limited, and various proposed methods can be used in the calculation of the principal axis of inertia.

[0052] The region segmentation unit 223 uses the principal axis of inertia calculated by the principal axis of inertia calculation unit 222 to segment the feature region detected by the region detection unit 221, thereby obtaining two segmented regions. Then, the region segmentation unit 223 outputs to the region selection unit 224 the image captured by the imaging unit 210, information representing the feature region detected by the region detection unit 221, and information representing the two obtained segmented regions. The region segmentation unit 223 is an example of the segmentation unit of the present invention.

[0053] The region selection unit 224 determines whether to use the two segmented regions obtained by the region segmentation unit 223 or the feature region detected by the region detection unit 221 (selecting the region to be used). The region selection unit 224 is an example of the determination unit of the present invention.

[0054] There are cases where feature regions are detected that are identical to the region of an object that should not be segmented, and where the feature region should not be segmented using the principal axis of inertia. Furthermore, in such cases, there is a tendency for the feature region to occupy a large proportion of its bounding rectangle. Therefore, if the evaluation value corresponding to the proportion of the feature region to its bounding rectangle is less than a predetermined threshold, the region selection unit 224 determines that two segmentation regions should be used. On the other hand, if the evaluation value is above the predetermined threshold, the region selection unit 224 determines that the feature region should be used. Thus, feature regions can be segmented more appropriately (enabling high-precision operation of using the segmentation result when the feature region should be segmented, and using the feature region (without using the segmentation result) when the feature region should not be segmented).

[0055] In Implementation 1, the region selection unit 224 calculates the bounding rectangle of the feature region based on information representing the feature region detected by the region detection unit 221, and calculates the bounding rectangle of each of the two segmented regions based on information representing the two segmented regions obtained by the region segmentation unit 223. Then, the region selection unit 224 calculates the ratio of the total area of ​​the bounding rectangles of the two segmented regions to the area of ​​the bounding rectangle of the feature region (the number of pixels within the bounding rectangle) as an evaluation value.

[0056] Alternatively, the evaluation value can be calculated as the ratio of the logical sum of the areas of the bounding rectangles of the two segmented regions to the area of ​​the bounding rectangle of the feature region. This allows for a more accurate calculation of the evaluation value, more closely approximating the ratio of the feature region to its bounding rectangle, and thus enabling a more appropriate decision on whether to adopt or reject a region. Of course, the evaluation value can also be calculated as the ratio of the feature region to its bounding rectangle, i.e., the ratio of the area of ​​the feature region to the area of ​​its bounding rectangle.

[0057] If a region is selected, the region selection unit 224 outputs the image captured by the imaging unit 210 and information representing the bounding rectangle of the selected region to the human body recognition unit 225.

[0058] The human body recognition unit 225 identifies whether an object contained within the bounding rectangle of the selected region is a human body (a defined object) based on information output from the region selection unit 224 (an image captured by the camera unit 210 and information representing the bounding rectangle of the selected region). When two segmented regions are used, the human body recognition unit 225 identifies whether an object contained within the bounding rectangle of each segmented region is a human body. This enables high-precision identification (detection) of human bodies. For example, when a human body is contained within a feature region, segmentation of the feature region yields a segmented region closer to the human body. Therefore, compared to using the bounding rectangle of the feature region, it is possible to identify (detect) with higher precision that the object contained within the bounding rectangle of the obtained segmented region is a human body. The human body recognition unit 225 is an example of the recognition unit of the present invention.

[0059] After completing human body recognition, the human body recognition unit 225 outputs the image captured by the imaging unit 210 and the result of human body recognition to the display unit 240. The display unit 240 can then display the result of human body recognition along with the captured image. For example, the result of human body recognition includes information representing the bounding rectangle of the area being used, and information indicating whether the object contained within the bounding rectangle is a human body. Then, the display unit 240 displays the bounding rectangle of the area being used overlaid with the captured image. The shape of the bounding rectangle (line type, brightness, color, etc.) changes depending on whether the object contained within the bounding rectangle is a human body.

[0060] Additionally, the human recognition unit 225 can also record the result of human recognition in association with the captured image in a storage medium. The storage medium may or may not be part of the camera 200.

[0061] Any algorithm can be used for human recognition based on the human recognition unit 225. For example, a recognizer that combines image features and enhancements, such as HoG or Haar-like algorithms, can also be used. Alternatively, a learned model generated through existing machine learning can be used; specifically, a learned model generated through deep learning (e.g., R-CNN, Fast R-CNN, YOLO, SSD, etc.) can also be used.

[0062] Figure 3 This is a flowchart illustrating an example of the processing flow of camera 200. Camera 200 executes this process repeatedly. Figure 3 The processing flow is as follows. The frame rate (repetition period) of the control unit 220 is not particularly limited, but in Embodiment 1, it is set to be the same as the frame rate (e.g., 30fps) of the image captured by the camera unit 210.

[0063] First, the camera unit 210 captures an image (step S301). Next, the region detection unit 221 detects feature regions with predetermined characteristics from the image captured in step S301 (step S302). Then, the principal axis of inertia calculation unit 222 calculates the principal axis of inertia of the feature regions detected in step S302 (step S303). Next, the region segmentation unit 223 segments the feature regions detected in step S302 using the principal axis of inertia calculated in step S303, thereby obtaining two segmented regions (step S304).

[0064] Then, the region selection unit 224 calculates an evaluation value based on the feature region detected in step S302 and the two segmented regions obtained in step S304, and determines whether the calculated evaluation value is less than a predetermined threshold (step S305). If it is determined that the evaluation value is less than the predetermined threshold (step S305: Yes), the process proceeds to step S306; if it is determined that the evaluation value is above the predetermined threshold (step S305: No), the process proceeds to step S307.

[0065] In step S306, the human body recognition unit 225 identifies whether the object contained in the bounding rectangle of each of the two segmented regions obtained in step S304 is a human body. In step S307, the human body recognition unit 225 identifies whether the object contained in the bounding rectangle of the feature region detected in step S302 is a human body. After the processing in step S306 or step S307 is performed, the process proceeds to step S308. In step S308, the display unit 240 displays the processing results of steps S301 to S307.

[0066] use Figure 4 Provide a specific example illustrating the actions of camera 200. Figure 4 In the example, a feature region 401 containing a human body and its shadow is detected (step S302). For high-precision human body recognition, the feature region 401 should be segmented into a region containing the human body and a region containing the shadow. The inertial axis calculation unit 222 calculates the centroid 402 of the feature region 401 and calculates the inertial axis 403 passing through the centroid 402 of the feature region 401 (step S303). Then, the region segmentation unit 223 segments the feature region 401 using the inertial axis 403, thereby obtaining two segmented regions 404 and 405 (step S304). Figure 4It can be seen that segmentation region 404 corresponds to the human body, and segmentation region 405 corresponds to the shadow. Therefore, segmentation regions 404 and 405 should be used for human body recognition. The region selection unit 224 calculates the circumscribed rectangle 406 of feature region 401, the circumscribed rectangle 407 of segmentation region 404, and the circumscribed rectangle 408 of segmentation region 405. Then, the region selection unit 224 calculates the ratio of the sum of the areas of circumscribed rectangle 407 and circumscribed rectangle 408 (total area) to the area of ​​circumscribed rectangle 406 as an evaluation value. Figure 4 As can be seen, the evaluation value is relatively small (less than the specified threshold) (step S305: Yes). Therefore, human body recognition can be performed using segmented regions 404 and 405 (step S306).

[0067] use Figure 5 Other specific examples illustrating the actions of camera 200. Figure 5 In the example, a feature region 501 identical to the human body region is detected (step S302). For high-precision human body recognition, feature region 501 should not be segmented. The inertial axis calculation unit 222 calculates the centroid 502 of feature region 501 and calculates the inertial axis 503 passing through the centroid 502 of feature region 501 (step S303). Then, the region segmentation unit 223 uses the inertial axis 503 to segment feature region 501, thereby obtaining two segmented regions 504 and 505 (step S304). Figure 5 It is known that segmentation region 504 corresponds to the upper body of the human body, and segmentation region 505 corresponds to the lower body. If segmentation regions 504 and 505 are used for human body recognition, the human body cannot be detected, or only a part of the human body may be detected. Therefore, segmentation regions 504 and 505 should not be used for human body recognition; instead, feature region 501 should be used. The region selection unit 224 calculates the circumscribed rectangle 506 of feature region 501, the circumscribed rectangle 507 of segmentation region 504, and the circumscribed rectangle 508 of segmentation region 505. Then, the region selection unit 224 calculates the ratio of the sum of the areas of circumscribed rectangle 507 and circumscribed rectangle 508 (total area) to the area of ​​circumscribed rectangle 506 as an evaluation value. Figure 5 It is known that the evaluation value is relatively large (a value above the specified threshold) (step S305: No). Therefore, human body recognition can be performed using feature region 501 instead of segmentation regions 504 and 505 (step S306).

[0068] Figure 6 This diagram illustrates an example of a display based on display unit 240. Figure 6The captured image contains a human body 601, a car 602, and their shadows. For the human body 601, since the evaluation value is less than a threshold, a box 603 (the bounding rectangle of the segmented region) containing the human body 601 and a box 604 (the bounding rectangle of the segmented region) containing the shadow of the human body 601 are displayed. For the car 602, since the evaluation value is above the threshold, a box 605 (the bounding rectangle of the feature region) containing the car 602 and its shadow is displayed. The object contained in box 603 is identified as a human body, while the objects contained in boxes 604 and 605 are not identified as human bodies; therefore, box 603 is displayed with a solid line, and boxes 604 and 605 are displayed with dashed lines.

[0069] As described above, according to Embodiment 1, the feature region is segmented using the principal axis of inertia of the feature region. Therefore, the feature region can be appropriately segmented (in a way that two segmented regions corresponding to the two desired objects are obtained respectively). Furthermore, based on an evaluation value corresponding to the proportion of the feature region to the bounding rectangle of the feature region, it is determined whether to use the two segmented regions obtained by segmenting the feature region, or to use the feature region itself. Therefore, the feature region can be segmented more appropriately (it is possible to achieve with high precision the operation of using the segmentation result when the feature region should be segmented, and using the feature region (without using the segmentation result) when the feature region should not be segmented).

[0070] <Implementation Method 2>

[0071] Embodiment 2 of the present invention will be described. Since there are two principal axes of inertia for the feature region, the feature region cannot be properly divided according to the principal axis of inertia used. In Embodiment 2, the principal axis of inertia used is selected in a manner that allows for more appropriate division of the feature region (allowing for more reliable and appropriate division of the feature region).

[0072] Figure 7 This is a block diagram illustrating a structural example of the camera 700 (information processing device) according to Embodiment 2. The camera 700 includes an image capturing unit 210, a control unit 720, a storage unit 230, and a display unit 240. The control unit 720 has a connection to the control unit 220 (… Figure 2 The control unit 720 has the same functions as the region detection unit 221, the inertial axis calculation unit 222, the inertial axis selection unit 721, the region segmentation unit 223, the region selection unit 224, and the human body recognition unit 225.

[0073] The principal axis of inertia calculation unit 222 outputs to the principal axis of inertia selection unit 721 an image captured by the imaging unit 210, information representing the feature region detected by the region detection unit 221, and information representing the calculated principal axis of inertia. In embodiment 2, the principal axis of inertia calculation unit 222 calculates two principal axes of inertia for the feature region, but this is not always the case. For example, the principal axis of inertia calculation unit 222 may calculate one of the two principal axes of inertia, and the principal axis of inertia selection unit 721 may correct the principal axis of inertia calculated by the principal axis of inertia calculation unit 222 as needed to obtain the other of the two principal axes of inertia.

[0074] The principal axis selection unit 721 selects one of the two principal axes of inertia for the feature region. When the center of gravity of the feature region is located in the first or third quadrant of the four quadrants obtained by the horizontal and vertical axes passing through the center of the circumscribed rectangle of the feature region, the principal axis of inertia with the smaller angle relative to the horizontal axis is selected. When the center of gravity of the feature region is located in the second or fourth quadrant of the four quadrants, the principal axis of inertia with the larger angle relative to the horizontal axis is selected. Furthermore, the principal axis selection unit 721 outputs to the region segmentation unit 223 the image captured by the imaging unit 210, information indicating the feature region detected by the region detection unit 221, and information indicating the selected principal axis of inertia. The region segmentation unit 223 segments the feature region detected by the region detection unit 221 using the principal axis of inertia selected by the principal axis selection unit 721. This allows for more appropriate segmentation of the feature region (enabling more reliable and appropriate segmentation of the feature region).

[0075] Figure 8 This is a flowchart illustrating an example of the processing flow of camera 700. Camera 700 executes this process repeatedly. Figure 8 The processing flow is as follows. The frame rate (repetition period) of the control unit 720 is not particularly limited, but in Embodiment 2, it is set to be the same as the frame rate of the camera unit 210 (e.g., 30fps).

[0076] exist Figure 8 In the processing flow, step S800 is performed between the processing in step S303 and the processing in step S304. In step S800, the inertial axis selection unit 721 selects one of the two inertial axes calculated in step S303. In step S304, the region segmentation unit 223 uses the inertial axis selected in step S800 to segment the feature region detected in step S302.

[0077] use Figure 9 Provide a specific example illustrating the actions of camera 700.

[0078] exist Figure 9In the example, a feature region 901 containing a human body and its shadow is detected (step S302). For high-precision human body recognition, the feature region 901 should be segmented into a region containing the human body and a region containing the shadow. The inertial axis calculation unit 222 calculates the centroid 902 of the feature region 901 and calculates the inertial axes 903 and 904 of the feature region 901 passing through the centroid 902 (step S303). The deflection angle θ1 from the horizontal direction to the inertial axis 903 is smaller than the deflection angle θ2 from the horizontal direction to the inertial axis 904. Since the inertial axes 903 and 904 are orthogonal to each other, the deflection angle θ2 is the deflection angle θ1 + 90°.

[0079] The inertia principal axis selection unit 721 selects one of the inertia principal axes 903 and 904 (step S800). From Figure 9 It can be seen that even if the feature region 901 is segmented using the principal axis of inertia 903, the feature region 901 will not be segmented into the area of ​​the human body and the area of ​​the shadow. However, by segmenting the feature region 901 using the principal axis of inertia 904, the feature region 901 can be segmented into the area of ​​the human body and the area of ​​the shadow. Therefore, the principal axis of inertia selection unit 721 needs to select the principal axis of inertia 904.

[0080] The inertial principal axis selection unit 721 calculates the centroid 902 of the feature region 901 and the center 905 of the circumscribed rectangle of the feature region 901 based on information representing the feature region detected by the region detection unit 221. Then, based on the coordinates of the centroid 902 and the center 905, the inertial principal axis selection unit 721 determines the quadrant in which the centroid 902 is located among the four quadrants obtained by the horizontal and vertical axes passing through the center 905. Here, it considers that as the... Figure 9 As the arrow travels along the horizontal axis, the horizontal coordinate (position coordinate in the horizontal direction) increases; similarly, as the arrow travels along the vertical axis, the vertical coordinate (position coordinate in the vertical direction) increases. In this case, if the horizontal coordinate of the centroid 902 is greater than the horizontal coordinate of the center 905, and the vertical coordinate of the centroid 902 is also greater than the vertical coordinate of the center 905, then the centroid 901 is located in the first quadrant. If the horizontal coordinate of the centroid 902 is less than the horizontal coordinate of the center 905, and the vertical coordinate of the centroid 902 is greater than the vertical coordinate of the center 905, then the centroid 901 is located in the second quadrant. If the horizontal coordinate of the centroid 902 is less than the horizontal coordinate of the center 905, and the vertical coordinate of the centroid 902 is less than the vertical coordinate of the center 905, then the centroid 901 is located in the third quadrant. If the horizontal coordinate of the centroid 902 is greater than the horizontal coordinate of the center 905, and the vertical coordinate of the centroid 902 is less than the vertical coordinate of the center 905, then the centroid 901 is located in the fourth quadrant.

[0081] Then, the principal axis selection unit 721 selects one of the principal axes of inertia 903 and 904 based on the determination result of the quadrant in which the center of gravity 902 is located. If the center of gravity 902 is located in the first or third quadrant, the principal axis selection unit 721 selects the principal axis 903, which has a smaller angle of deviation relative to the horizontal axis (horizontal direction). If the center of gravity 902 is located in the second or fourth quadrant, the principal axis selection unit 721 selects the principal axis 904, which has a larger angle of deviation relative to the horizontal axis. Figure 9 In this configuration, the center of gravity 902 is located in the fourth quadrant, therefore the principal axis of inertia selection unit 721 can select the principal axis of inertia 904. Furthermore, the feature region 901 can be divided into the region of the human body and the region of the shadow.

[0082] As described above, according to Embodiment 2, the principal axis of inertia is selected from the four quadrants obtained by the horizontal and vertical axes passing through the center of the circumscribed rectangle of the feature region, based on the quadrant in which the centroid of the feature region is located. This allows for more appropriate segmentation of the feature region (and enables more reliable and appropriate segmentation of the feature region).

[0083] <Other>

[0084] The above embodiments are merely illustrative examples illustrating the structure of the present invention. The present invention is not limited to the specific embodiments described above, and various modifications can be made within the scope of its technical concept. For example, the identified object may not be a human body, but may be a car or building, a finger, an arm, a torso, an animal, etc. Furthermore, an example of dividing the feature region into a region of the human body and a region of the shadow has been described, but the invention is not limited thereto. According to the present invention, as... Figure 10 As shown in (A), the feature region can also be divided into two buildings (two regions corresponding to the two buildings respectively). For example... Figure 10 As shown in (B), the feature region can also be divided into the area of ​​the utility pole and the area of ​​its shadow. For example... Figure 10 As shown in (C), the feature area can also be divided into a thumb area and an index finger area. Alternatively, the feature area (the area of ​​the human body) can be divided into a torso area and an arm area. Thus, the present invention can be used for a variety of purposes.

[0085] <Postscript 1>

[0086] Information processing apparatus (100, 200, 700), characterized in that it has:

[0087] Detection units (101, 221) detect feature regions with prescribed characteristics from the captured image; and

[0088] The segmentation unit (102, 223) uses the principal axis of inertia of the feature region to segment the feature region.

[0089] <Appendix 2>

[0090] The information processing method is characterized by having:

[0091] Detection step (S302): Detecting feature regions with prescribed characteristics from the captured image; and

[0092] The segmentation step (S304) involves segmenting the feature region using the principal axis of inertia of the feature region.

[0093] Label Explanation

[0094] 100: Information processing device

[0095] 101: Regional Testing Department

[0096] 102: Region Segmentation Department

[0097] 200: Camera (Information Processing Device)

[0098] 210: Camera Department

[0099] 220: Control Department

[0100] 230: Storage Department

[0101] 240: Display Section

[0102] 221: Regional Testing Department

[0103] 222: Inertia Principal Axis Calculation Unit

[0104] 223: Region Segmentation Department

[0105] 224: Area Selection Department

[0106] 225: Human Recognition Department

[0107] 700: Camera (Information Processing Device)

[0108] 720: Control Department

[0109] 721: Inertial spindle selection unit.

Claims

1. An information processing device, characterized in that, have: The detection unit detects feature regions with specified characteristics from the captured image; as well as The segmentation unit segments the feature region using the principal axis of inertia of the feature region. When the centroid of the feature region is located in either the first or third quadrant of the four quadrants obtained by the horizontal and vertical axes passing through the center of the circumscribed rectangle of the feature region, the segmentation unit uses the principal axis of inertia of the feature region with the smaller angle of deviation relative to the horizontal axis to segment the feature region. When the center of gravity of the feature region is located in the second or fourth quadrant of the four quadrants, the segmentation unit uses the principal axis of inertia of the feature region that has the larger angle of deviation relative to the horizontal axis to segment the feature region.

2. The information processing device according to claim 1, characterized in that, It also includes a judgment unit that determines whether to use the segmentation unit to segment the feature region to obtain two segmented regions, or to use the feature region itself. If the evaluation value corresponding to the proportion of the feature region to the bounding rectangle of the feature region is less than a predetermined threshold, the judgment unit determines that the two segmented regions should be used. If the evaluation value is above the specified threshold, the judgment unit determines that the feature region should be used.

3. The information processing device according to claim 2, characterized in that, The evaluation value is the ratio of the total area of ​​the outer rectangles of the two segmented regions to the area of ​​the outer rectangle of the feature region.

4. The information processing apparatus according to claim 2, characterized in that, The evaluation value is the ratio of the area of ​​the logical sum of the bounding rectangles of the two segmented regions to the area of ​​the bounding rectangle of the feature region.

5. The information processing apparatus according to claim 2, characterized in that, It also includes a recognition unit for identifying whether an object contained within the bounding rectangle of the adopted area is a specified object. When the two segmented regions are used, the identification unit identifies, for each of the two segmented regions, whether the object contained in the outer rectangle of that segmented region is a specified object.

6. The information processing apparatus according to claim 5, characterized in that, The specified object is the human body.

7. The information processing apparatus according to claim 1, characterized in that, The detection unit detects the areas where movement occurs as the feature regions.

8. The information processing apparatus according to claim 1, characterized in that, The detection unit detects regions with pixel values ​​within a specified range as the feature regions.

9. The information processing apparatus according to claim 1, characterized in that, The detection unit detects the region surrounded by the edge as the feature region.

10. An information processing method, characterized in that, have: The detection step involves detecting feature regions with specified characteristics from the captured image; as well as The segmentation step utilizes the principal axis of inertia of the feature region to segment the feature region. In the segmentation step, When the centroid of the feature region is located in either the first or third quadrant of the four quadrants obtained by the horizontal and vertical axes passing through the center of the circumscribed rectangle of the feature region, the feature region is divided using the principal axis of inertia of the feature region that has the smaller angle of deviation relative to the horizontal axis. When the center of gravity of the feature region is located in the second or fourth quadrant of the four quadrants, the feature region is divided using the principal axis of inertia of the feature region that has the larger angle of deviation relative to the horizontal axis.

11. A computer program product, Used to cause a computer to perform each step of the information processing method of claim 10.

Images