JP5737909B2 - Image processing apparatus, image processing method, and program - Google Patents

Description

  The present invention relates to a technique for recognizing an object from a frame image or the like acquired from a camera.

  Conventionally, a person passing through an entrance or a passage of a store is photographed with a camera, the position of a person's face is detected from the photographed image, and the number of people passing is measured, or a person's face registered in advance A technique for recognizing whether or not there is disclosed.

  As a technique for automatically counting passers-by in such a predetermined area from a camera video, for example, there is Patent Document 1 below. In this patent document 1, a camera is installed from the upper side of the passage toward directly below. Since the shape of the head of the person viewed from above the camera is a circle, the person is detected and counted by extracting a circular object from the camera image.

  On the other hand, in recent years, a technique for detecting a face from an image has been put into practical use. Using such a technique, it is possible to count a person by installing a camera in front of the passage and detecting a face from the camera image as shown in FIG.

  Here, the more the camera captures a wider area, the more the face is photographed depending on the positional relationship between the camera and the person. And if the orientation of the face changes, the facial features will change. Therefore, recognition becomes difficult.

  In order to cope with this problem, in Patent Document 2 below, a recognition dictionary is prepared according to the orientation of the face, the frame image is divided into a plurality of regions, and the recognition dictionary applied in each region is changed. Here, the case of FIG. 11 is given as an example.

FIG. 11 is a schematic diagram for explaining that the orientation of the face to be photographed varies depending on the distance from the camera.
In FIG. 11, 1101 is the ceiling of the passage, and 1102 is the floor. Reference numeral 1103 denotes a camera, which is installed on the ceiling 1101 and photographs the passage from an oblique upper side. When a person is far from the camera 1103 as in 1104, the vertical direction of the captured face is a small angle, but when a person is near the camera 1103 as in 1105, the vertical direction of the face is The direction becomes a big angle.

FIG. 12 is a schematic diagram for explaining that facial features change depending on how they are seen.
1201 is when a person is far from the camera 1103 like a person 1104, and 1202 is when a person is near a camera 1103 like a person 1105. As can be seen from FIG. 12, the facial features change depending on how they are seen.

FIG. 13 is a schematic diagram for explaining a problem due to the prior art.
Therefore, as shown in FIG. 13, the frame image 1301 is divided into two areas 1302 and 1303, and a dictionary that recognizes a face with a small angle is used for the area 1302, and the angle is set for the area 1303. A dictionary that recognizes large faces is used. That is, in the above example, 1304 is a position in the face frame image recognized when a person is at the position of the person 1104, and a face dictionary with a small angle is used as the recognition dictionary. Reference numeral 1305 denotes a position in the face frame image recognized when the person is at the position of the person 1105, and a face dictionary having a large angle is used as the recognition dictionary.

JP-A-4-199487 JP 2007-25767 A

  However, in Patent Document 2, it is difficult to recognize if there is a face at a position corresponding to the boundary of a region to which each recognition dictionary is applied. In the above example, there is a case where a face exists at the position 1306 in FIG. The face angle at such a position is an intermediate angle. That is, the variation of the feature is the largest both in the face dictionary with a small angle and the face dictionary with a large angle. Therefore, the recognition accuracy tends to be low regardless of which recognition dictionary is used, which makes recognition difficult.

  The present invention has been made in view of such problems, and provides a mechanism that can accurately recognize an object even when the direction of the object to be recognized changes within the screen. Objective.

In order to achieve the above-described object, the present invention provides dictionary storage means for storing data of a plurality of dictionaries corresponding to different directions of the same object, and an application area in a frame image for each of the plurality of dictionaries. Setting means for setting an area that overlaps application areas of at least two dictionaries; and recognition means for recognizing the object in the application area set for the dictionary using each of the plurality of dictionaries. And the recognition means switches the dictionary to be used for each frame image in a plurality of consecutive frame images with respect to an application region that overlaps at least two dictionaries, and integrates the recognition results for each frame image. An image processing apparatus is provided.

  According to the present invention, an object can be recognized with high accuracy even when the direction of the recognized object changes in the screen.

It is a schematic diagram which shows the example of installation of the image processing apparatus which concerns on embodiment of this invention. It is a block diagram which shows an example of the hardware constitutions of the image processing apparatus which concerns on embodiment of this invention. It is a block diagram which shows an example of a function structure of the image processing apparatus which concerns on embodiment of this invention. It is a flowchart which shows an example of the process sequence of the image processing method by the image processing apparatus which concerns on embodiment of this invention. It is a schematic diagram which shows embodiment of this invention and demonstrates the angle of a face. It is the schematic which showed embodiment of this invention and showed the relationship between the angle of the face which recognition dictionary A and B can apply, and recognition accuracy. It is a schematic diagram which shows embodiment which shows embodiment of this invention and can apply the recognition dictionary A and B. FIG. It is a figure explaining the method of searching for a face pattern from an image. It is a flowchart which shows embodiment of this invention and demonstrates the operation | movement which switches a dictionary and a collation area | region. It is the schematic diagram which showed embodiment of this invention and showed an example of the production | generation and count of a locus | trajectory. It is a schematic diagram explaining that the direction of the face image | photographed according to the distance from a camera differs. It is a schematic diagram explaining that the feature of a face changes with how it looks. It is a schematic diagram for demonstrating the subject by a prior art.

  Hereinafter, embodiments (embodiments) for carrying out the present invention will be described with reference to the drawings.

In addition, embodiment mentioned below demonstrates by the example which measures the number of people who pass a passage.
FIG. 1 is a schematic diagram illustrating an installation example of an image processing apparatus according to an embodiment of the present invention.
101 is the ceiling of the passage and 102 is the floor of the passage. Reference numeral 103 denotes a person passing through the passage. An imaging unit (camera) 104 is installed on the ceiling 101 so that the person 103 can be photographed from above. Reference numeral 105 denotes a LAN cable, which transmits an image captured by the imaging unit 104. A PC 106 is an image processing apparatus that analyzes and counts video.

FIG. 2 is a block diagram illustrating an example of a hardware configuration of the image processing apparatus 106 according to the embodiment of the present invention.
In FIG. 2, reference numeral 201 denotes a CPU, which executes various controls in the image processing apparatus 106 according to the present embodiment. Reference numeral 202 denotes a ROM which stores a boot program executed when the image processing apparatus 106 is started up and various data. A RAM 203 stores a control program to be processed by the CPU 201 and provides a work area when the CPU 201 executes various controls. A keyboard 204 and a mouse 205 provide various input operation environments for the user.

  Reference numeral 206 denotes an external storage device, which includes a hard disk, flexible disk, optical disk, magnetic disk, magneto-optical disk, magnetic tape, and the like. However, the external storage device 206 is not necessarily a necessary component if the ROM 202 has all the control programs and various data. In this embodiment, it is assumed that a control program related to the processing of the present invention is stored in the ROM 202 (or the external storage device 206).

  Reference numeral 207 denotes a display device that includes a display or the like and displays a result or the like to the user. Reference numeral 208 denotes a network interface (NIC) that enables communication with the imaging unit 104 on the network via the LAN cable 105. Reference numeral 209 denotes a video interface (video I / F), which makes it possible to capture a frame image from the imaging unit 104 and the coaxial cable. Reference numeral 210 denotes a bus connecting the above-described components.

FIG. 3 is a block diagram illustrating an example of a functional configuration of the image processing apparatus 106 according to the embodiment of the present invention.
Reference numeral 10 denotes an imaging unit including an imaging lens and an imaging sensor such as a CCD or a CMOS. The imaging means 10 corresponds to the imaging unit 104 in FIG.

  Reference numeral 30 denotes an image acquisition unit that acquires image data captured by the imaging unit 10 at predetermined time intervals and outputs the data in units of a plurality of temporally continuous frames. The frame image is sent as packet data of the http protocol from the imaging unit 104 via the LAN cable 105 and acquired via the network interface 208 on the image processing apparatus 106. Alternatively, 105 may be configured by a coaxial cable and acquired by the video interface 209 on the image processing apparatus 106.

Reference numeral 40 denotes an object recognizing unit that recognizes whether a desired object is reflected in the image data acquired by the image acquiring unit 30. Specifically, the object recognition unit 40 performs processing for recognizing an object in a predetermined direction using a dictionary related to the object.
Reference numeral 50 denotes recognition result analysis / output means, which analyzes the result recognized by the object recognition means 40 and outputs the analyzed result to be displayed on the display device 207, for example.

Reference numeral 60 denotes an object dictionary storage unit, which is a memory that stores an object dictionary corresponding to a desired recognition target used by the object recognition unit 40. Specifically, the object dictionary storage unit 60 stores data of a plurality of dictionaries corresponding to different directions of the same object. The object dictionary is obtained in advance by machine learning from a number of object patterns in a predetermined direction. It is stored in the external storage device 206 and read into the RAM 203 when the program is started. In the present embodiment, it is assumed that a plurality of recognition dictionaries corresponding to the vertical angle of the face are prepared. Of course, the recognition dictionaries may be divided according to angles in different directions (such as the horizontal direction). However, in order to simplify the explanation, an example of a recognition dictionary according to the angle in the vertical direction will be described below.
Reference numeral 70 denotes a dictionary / collation area setting unit that selects a recognition dictionary to be used from a plurality of recognition dictionaries stored in the object dictionary storage unit 60 and switches the selected recognition dictionary to perform collation. Each region in the frame image is set in the object recognition means 40. That is, the dictionary / collation region setting unit 70 constitutes a switching unit that switches a plurality of dictionaries for each region of the frame image and applies them to the object recognition unit 40.

Reference numeral 80 denotes dictionary application area determination means for determining a collation area in the frame image for the recognition dictionary stored in the object dictionary storage means 60.
Reference numeral 90 denotes dictionary application area storage means, which stores the collation area in the frame image for the recognition dictionary determined by the dictionary application area determination means 80.

  The image acquisition means 30, the object recognition means 40, the dictionary / collation area setting means 70, and the dictionary application area determination means 80 in FIG. 3 are, for example, controls stored in the CPU 201 and ROM 202 (or external storage device 206) in FIG. It consists of a program and a RAM 203. The recognition result analysis / output unit 50 includes, for example, a control program stored in the CPU 201 and the ROM 202 (or the external storage device 206) in FIG. 2, the RAM 203, and the display device 207. Further, the object dictionary storage unit 60 and the dictionary application area storage unit 90 are configured in the external storage device of FIG. 2, for example.

  FIG. 4 is a flowchart showing an example of the processing procedure of the image processing method by the image processing apparatus 106 according to the embodiment of the present invention.

  First, in step S400, the dictionary application area determination unit 80 determines an area to which each of a plurality of recognition dictionaries with different face angle ranges that can be applied in a frame image can be applied. This can be determined from the relationship between the position of the face in the frame image and the angle of the face when a face is present at that position. In the present embodiment, description will be made assuming that the angle is in the vertical direction.

The angle of the face is an angle shown as Θ in FIG. 5 expressed as 0 degrees in the horizontal direction when the person is standing upright.
FIG. 5 is a schematic diagram illustrating an embodiment of the present invention and explaining a face angle.
Here, 501 in FIG. 5 is a ceiling of the passage, 502 is a floor of the passage, 503 is an imaging unit (camera), and 504 is a person.

As shown in FIG. 5, Θ is equal to an angle formed by a straight line along the ceiling and a straight line drawn from the camera 503 to the face of the person 504 (solid line in FIG. 5). Therefore, by obtaining the distance from the camera 503 to the person 504 (X in FIG. 5) and (the height of the camera 503 from the ground-the height of the face from the ground) (Y in FIG. 5), Θ is The following equation (1) can be obtained.
Θ = tan ⁻¹ ((the height of the camera 503 from the ground−the height of the face from the ground) / the distance from the camera 503 to the person) (1)

  Since the range of Θ depends on which range can be photographed, it is determined by the angle of view of the camera 503 and the angle of the camera 503 at the time of installation. In the example of FIG. 5, the dotted line is the range that can be photographed. The height of the face of the person 504 may be assumed to be, for example, a person having an average height, so that if the installation conditions of the camera 503 (camera height, camera angle, camera angle of view) are determined, the height is calculated. be able to. Therefore, the installation condition of the camera 503 can be input by the user, and the relationship between the face position and the face angle in the frame image can be obtained by Expression (1).

Also, a relational expression (Θ = f (y)) between an arbitrary position (y) in the frame image and the face angle Θ is defined, and several values are input at the time of installation to obtain f (y). It may be. For example, it is assumed that an arbitrary position (y) in the frame image and the face angle Θ can be expressed by the following linear expression (2).
Θ = ay + b (2)
Then, if the position and the face angle in two or more frame screens are measured in advance, the coefficient a and the constant b can be obtained. Accordingly, the coefficient a and the constant b may be obtained by actually standing the person 504 at two or more points in front of the camera 503 and inputting the position in each frame screen and the distance from the camera 503.

Next, the recognition dictionary will be described.
The recognition dictionary is created so that the range of applicable face angles overlaps.
FIG. 6 is a schematic diagram showing the relationship between the face angle applicable to the recognition dictionaries A and B and the recognition accuracy according to the embodiment of the present invention.

  In FIG. 6, the horizontal axis indicates the face angle to which the recognition dictionary can be applied, the vertical axis indicates the recognition accuracy, 601 indicates the recognition dictionary A, and 602 indicates the recognition dictionary B. At this time, the applicable face angle range of the recognition dictionary A (601) is 603 (Θ1) to 604 (Θ3), and the applicable face angle range of the recognition dictionary B (602) is 605 (Θ2). 606 (Θ4). Note that Θ1 <Θ2 <Θ3 <Θ4.

  As already described in [Problems to be Solved by the Invention], the accuracy of recognition decreases when it comes close to the end of the range of applicable face angles. Therefore, in the present embodiment, the two recognition dictionaries are created so that the range of applicable face angles (image regions) overlap so that the reduction in recognition accuracy can be complemented. In the example of FIG. 6, the range from 605 (Θ2) to 604 (Θ3) corresponds. The size of the overlap can be determined by how much the allowable range of the recognition accuracy is set. By using the two recognition dictionaries A and B created in this way, face detection with high recognition accuracy is always possible. The areas in the frame image to which the recognition dictionaries A and B are applied are determined so as to overlap as follows using the relationship between the face position and the face angle in the frame image.

FIG. 7 is a schematic diagram illustrating an area to which the recognition dictionaries A and B can be applied according to the embodiment of the present invention.
In the example of FIG. 7, in 701 (shaded area) in FIG. 7A, the face angle changes from Θ <b> 1 to Θ <b> 3 as the face position moves from the upper end to the lower end. Therefore, the recognition dictionary A (601) can be applied to this area.

  In 702 (shaded area) in FIG. 7B, the face angle changes from the angle Θ2 to Θ4 as the face position moves from the upper end to the lower end. Therefore, the recognition dictionary B (602) can be applied to this area. 7A and 7B are drawn as shown in FIG. 7C, the portion 704 (horizontal line region) is a region of Θ2 <Θ3, and both the recognition dictionary A and the recognition dictionary B Is applicable. That is, in the area 704, the application areas of the recognition dictionary used for the image area overlap.

  As described above, an area to which each of a plurality of recognition dictionaries having different face angle ranges can be determined. These areas are stored in the dictionary application area storage unit 90.

Here, it returns to description of FIG. 4 again.
When the process of step S400 ends, the process proceeds to step S401.
In step S401, the image processing apparatus 106 determines whether to end the process.

  As a result of the determination in step S401, if there is an instruction to end the process from the user via the power OFF or the keyboard 204 or the mouse 205, the process of this flowchart ends.

  On the other hand, if the result of determination in step S401 is that there is no instruction to end processing from the user, processing proceeds to step S402. That is, the processes in steps S402 to S406 are repeated until the user gives an instruction to end the process.

In step S402, the image acquisition unit 30 acquires a frame image from the video input to the imaging unit 10 by the method described above.
The image data read here is, for example, data of a two-dimensional array composed of 8-bit pixels, and is composed of three planes R, G, and B. At this time, if the image data is compressed by a method such as JPEG, the image data is decompressed according to a predetermined decompression method to obtain image data composed of RGB pixels. Further, in the present embodiment, RGB data is converted into luminance data, and the luminance image data is applied to subsequent processing, and is stored in an image memory (for example, the external storage device 206 in FIG. 2). When YCrCb data is input as image data, the Y component may be directly used as luminance data.

  Subsequently, in step S403, the object recognition unit 40 performs collation with the dictionary data set by the dictionary / collation region setting unit 70 from the image data transferred to the internal image memory to recognize a desired object.

First, a general object recognition method will be described.
The methods proposed in the known technique 1 and the known technique 2 are known.
For example, the known technique 1 is a technique for detecting a face pattern in an image using a neural network. The method will be briefly described below.

First, image data for face detection is read into a memory, and a predetermined area to be matched with the face is cut out from the read image. Then, one output is obtained by calculation using a neural network with the distribution of pixel values in the cut-out area as an input. At this time, the weight and threshold of the neural network are learned in advance using a face image pattern and a non-face image pattern, and for example, a face is determined if the output of the neural network is 0 or more, and a non-face is determined otherwise To do. Here, the weight and threshold value are dictionary data. Then, the face is detected from the image by scanning the cutout position of the image pattern to be collated with the face which is an input of the neural network, for example, as shown in FIG.
FIG. 8 is a diagram for explaining a method for searching for a face pattern from an image.
Specifically, the entire image 801 is scanned vertically and horizontally to extract a pattern 802 to be collated. Face discrimination processing 803 is performed on the pattern 802 to be collated.

  Further, as an example paying attention to the speeding up of processing, there is known technique 2. In this technology, AdaBoost is used to effectively combine many weak classifiers to improve the accuracy of face discrimination, while each weak classifier is configured with a Haar type rectangular feature quantity, The calculation is performed at high speed using the integral image. In addition, the discriminators obtained by AdaBoost learning are connected in series to form a cascade type face detector. This cascade type face detector first removes a pattern candidate that is clearly not a face on the spot using a simple discriminator in the previous stage. Only for the other candidates, it is determined whether or not it is a face using a later complex discriminator having higher discrimination performance. As a result, it is not necessary to make a complicated determination for all candidates, which is fast. Note that the weights and threshold values used in the discriminator are dictionary data as in the known technique 1.

Next, in this embodiment, the operation by the characteristic dictionary / collation area setting means 70 will be described with reference to the flowchart of FIG.
FIG. 9 is a flowchart illustrating an operation of switching between a dictionary and a collation area according to the embodiment of this invention.

  First, in step S900, the dictionary / collation region setting means 70 determines whether or not processing has been performed for all frames. As a result of this determination, if all the frames have been processed, the processing of this flowchart is terminated.

  On the other hand, if all the frames have not been processed, the process proceeds to step S901. That is, if the processing has not been performed for all the frames, the processing from step S901 to step S909 or S910 is repeated.

  Subsequently, in step S <b> 901, the dictionary / collation area setting unit 70 selects the recognition dictionary A from the plurality of recognition dictionaries read from the object dictionary storage unit 60 and sets the recognition dictionary A in the object recognition unit 40.

  Subsequently, in step S <b> 902, the dictionary / collation area setting unit 70 reads the collation area of the recognition dictionary A from the dictionary application area storage unit 90 and sets it in the object recognition unit 40. As described above, the collation area of the recognition dictionary A is 701 in FIG.

  Subsequently, in step S903, the object recognition unit 40 uses the recognition dictionary A set in step S901 to collate with the dictionary within the range of the collation area set in step S902.

  Subsequently, in step S <b> 904, the dictionary / collation region setting unit 70 selects the recognition dictionary B from the plurality of recognition dictionaries read from the object dictionary storage unit 60 and sets the recognition dictionary B in the object recognition unit 40.

  In step S 905, the dictionary / collation area setting unit 70 reads the collation area of the recognition dictionary B from the dictionary application area storage unit 90 and sets it in the object recognition unit 40. As described above, the collation area of the recognition dictionary B is 702 in FIG.

  Subsequently, in step S906, the object recognition unit 40 uses the recognition dictionary B set in step S904 to collate with the dictionary within the range of the collation area set in step S905.

Next, in order to deal with recognition of faces of various sizes, the frame image is reduced and matching is repeated in the subsequent processing.
First, in step S907, for example, the object recognition unit 40 (or the dictionary / collation region setting unit 70) determines whether or not the reduction has been sufficiently performed. Here, when the image pattern is reduced to the same size as the image pattern used for collation, the maximum face is detected in the frame image.

  If the result of determination in step S907 is that the reduction is not sufficient, that is, it can be reduced within a range that does not become smaller than the image pattern used for collation, the process proceeds to step S908.

  In step S908, the object recognition unit 40 reduces the frame image at a predetermined reduction rate.

  Subsequently, in step S909, the dictionary / collation area setting unit 70 reduces the collation areas of the recognition dictionary A and the recognition dictionary B at the same reduction ratio as in step S908. Then, the process returns to step S901.

  Thereafter, the processes in steps S901 to S906 are performed on the frame image reduced in step S908. Here, the area reduced in step S909 is used as the collation area set in steps S902 and S905.

  As described above, the processes in steps S901 to S909 are repeated for one frame image.

  On the other hand, as a result of the determination in step S907, when the image pattern is reduced to the same size as the image pattern used for collation, the process proceeds to step S910, and the image acquisition unit 30 acquires the next frame image, and returns to step S900.

  Through the above processing, by using the two recognition dictionaries A and B where the application range of the face angle is overlapped, it becomes possible to recognize the face with high accuracy regardless of the position in the frame image. However, when both the recognition dictionary A and the recognition dictionary B are used for each frame, the overlapping portion (704 in FIG. 7C) is double-checked with the recognition dictionary (steps S903 and S906). Become. This increases the calculation cost. Therefore, this problem can be avoided by the following method.

  The recognition dictionary and collation area to be used are changed between the Nth frame and the (N + 1) th frame (where N is a natural number). That is, in the N frame, the dictionary / collation area setting means 70 sets the recognition dictionary A, and collation is performed only on the area 701 in FIG. Then, in the N + 1 frame, the dictionary / collation region setting means 70 sets the recognition dictionary B and performs it only for the region 702 in FIG. In this way, recognition is performed while switching between the dictionary used for each successive frame and the collation area. As a result, the face is recognized in either the N frame or the N + 1 frame regardless of the position in the frame image. In addition, since the collation area for each frame is limited, the calculation cost can be reduced as compared with the case where collation is performed for all areas of the frame image.

  As a recognition result, a logical sum of N frames and N + 1 frames may be used. Even if the overlap is recognized in both the Nth frame and the (N + 1) th frame, if the time difference between the previous and next frames is sufficiently small, the position will hardly change, so it is easy to determine that they are the same. is there.

Here, it returns to description of FIG. 4 again.
When the process of step S403 ends, the process proceeds to step S404.
In step S404, the recognition result analyzing / outputting unit 50 reads out the subject area detected between the present time and a predetermined time before from the RAM 203, and generates a locus. This is a process for determining which of a plurality of faces detected within a predetermined time corresponds to the movement of the same person.

Details of this processing will be described with reference to FIG.
FIG. 10 is a schematic diagram illustrating an example of generation and counting of a trajectory according to the embodiment of the present invention.

  In FIG. 10, reference numeral 1001 denotes the entire frame being imaged. Here, the face area detected at a predetermined time is represented by a rectangle and overlaid (1003 to 1005). In the example of FIG. 10, it is assumed that three frames are overdrawn. 1003 is detected in the oldest frame, 1004 is detected in the next frame, and 1005 is detected in the next current frame. As a method for obtaining these trajectories, the centers of the respective regions may be obtained, and those having the smallest distance between the centers of the respective regions may be regarded as the same subject and connected by line segments. The locus obtained in this way is 1009 in the example of FIG.

  Subsequently, in step S405, the recognition result analysis / output unit 50 checks whether or not the trajectory created in step S404 satisfies a predetermined condition, and counts if the condition is satisfied. Here, the predetermined condition is, for example, whether or not the measurement line 1002 shown in FIG. 10 is crossed. The measurement line 1002 is set in the frame screen by the user. In the example of FIG. 10, since the trajectory 1009 crosses the measurement line 1002, it is counted as 1. If there is still a trajectory that does not cross the measurement line 1002, it is not counted at this point.

  Subsequently, in step S406, the recognition result analysis / output unit 50 displays the counted result to the user.

  Then, it returns to step S401 again.

  As described above, a plurality of dictionaries with overlapping application ranges are prepared, and the dictionary and collation range used in odd frames and even frames are switched. Thereby, even if the direction of the recognized object changes in the screen, it can be recognized with high accuracy.

  In the present embodiment, the range to be collated with the recognition dictionary used by the dictionary / collation area setting unit 70 is alternately changed for each frame, but the following method may be used. That is, the range 703 in FIG. 7C is collated using the recognition dictionary A. The overlapping region 704 is also first collated using the recognition dictionary A. At this time, a likelihood map is created based on the probability (likelihood) obtained as a result of the collation with the recognition dictionary. The likelihood is obtained from a calculation result before threshold processing at the time of matching with a dictionary.

  Next, the likelihood map is referred to and collation is performed using the recognition dictionary B only for a portion where the likelihood is a predetermined value or less. For 705 in FIG. 7C, collation is performed using the recognition dictionary B. In this way, 704 in FIG. 7C is partially double-checked. However, when the cascade type discriminator of publicly known technology 2 is used, the one with low likelihood is Since it can be determined by the discriminator in the previous stage, the time required for collation is extremely small. Therefore, the calculation cost does not increase greatly.

  In the present embodiment, the example in which the position of the face is detected has been described. However, the present invention can be applied to various parts of a person such as the entire human body, the upper body, and the head, and various objects such as a car and a bicycle. Further, the present invention can also be applied to the case of discriminating whether or not a person is a specific person from the facial features of the individual.

  In the present embodiment, the angle in the vertical direction of the face has been described. Of course, the same applies to the angle in the horizontal direction.

  Further, in the present embodiment, the example in which the recognition result analysis / output unit 50 counts the number of people passing through the passage has been described. However, the present invention can be applied to various uses such as measuring the congestion rate of a predetermined area, analyzing a flow line, and generating an alarm for a specific person.

In this embodiment, the image processing apparatus 106, which is a PC, is configured to perform recognition, counting, and display. However, the present invention is not limited to this. For example, by storing everything from the object recognition unit 40 to the dictionary application area storage unit 90 on a chip and integrating it with the imaging unit 104, only the count result is received by the image processing apparatus 106 via the LAN cable 105, and the count is performed. You may make it browse a result. Alternatively, the object recognition unit 40, the dictionary / collation region setting unit 70, the object dictionary storage unit 60, the dictionary application region determination unit 80, and the dictionary application region storage unit 90 are integrated with the imaging unit 104. Only the recognition result may be received by the image processing apparatus 106 via the LAN cable 105 and counted by the image processing apparatus 106.
Naturally, the present embodiment can also be realized by executing a program in a computer.

  Further, in this embodiment, an example of switching a plurality of dictionaries for each region of a frame image has been described. However, for example, a mode in which images are temporally continuous images and a dictionary to be used for each image is switched is also applicable. It is.

(Other embodiments)
The present invention can also be realized by executing the following processing.
That is, software (program) that realizes the functions of the above-described embodiments is supplied to a system or apparatus via a network or various storage media, and a computer (or CPU, MPU, or the like) of the system or apparatus reads the program. It is a process to be executed.
This program and a computer-readable recording medium storing the program are included in the present invention.

10 imaging means, 30 image acquisition means, 40 object recognition means, 50 recognition result analysis / output means, 60 object dictionary storage means, 70 dictionary / collation area setting means, 80 dictionary application area determination means, 90 dictionary application area storage means

Claims (3)

Dictionary storage means for storing data of a plurality of dictionaries corresponding to different directions of the same object;
Setting means for setting an application area in a frame image for each of the plurality of dictionaries, including an area overlapping with an application area of at least two dictionaries ;
Recognizing means for recognizing the object in an application area set for the dictionary using each of the plurality of dictionaries,
The recognition unit is configured to switch a dictionary to be used for each frame image in a plurality of continuous frame images with respect to an application region that overlaps at least two dictionaries, and to integrate the recognition results for the frame images. Image processing device. A setting step for setting an application area in the frame image for each of a plurality of dictionaries corresponding to different directions of the same object, including an area overlapping the application areas of at least two dictionaries ;
Recognizing the object in an application area set for the dictionary using each of the plurality of dictionaries, and
In the recognition step, for an application region that overlaps at least two dictionaries, a dictionary to be used for each frame image in a plurality of continuous frame images is switched, and a recognition result for each frame image is integrated. Image processing method. A program for causing a computer to execute each step of the image processing method according to claim 2 .

Images