JP5218345B2 - Image detection device - Google Patents

Description

  The present invention relates to an image detection apparatus that detects a specific image from an image captured by a digital camera, a digital video camera, or the like.

  Conventionally, in the field of digital cameras and digital video cameras, a method for detecting a specific image (hereinafter referred to as a target image) such as a face image from a photographed image has been proposed. Non-Patent Document 1 discloses a Viola-Jones algorithm that detects a specific image by a detector generated by machine learning using a sample image as one of standard image detection techniques. A face detection algorithm using a neural network has also been proposed. Since these methods use a group of sample images with similar top and bottom directions for learning, the in-plane rotation angle of the target image that can be detected by the generated detector may be determined by the top and bottom direction of the sample images used for learning. Many. In such a case, the above method cannot detect a target image rotated in a direction different from the vertical direction of the sample image.

  Therefore, various methods have been proposed that can detect a target image even in such a case. For example, a method for detecting an input image by rotating it at a certain rotation angle, a method for preparing learning data used by the detector at each rotation angle in advance, or a target image rotated at a certain angle can be detected. A method of converting learning data learned in accordance with the angle to be detected has been proposed.

  In these methods, a target image is detected by inputting a region (selected region) in which a part of an input image is selected to a detector. However, since the size of the selection area is fixed, the size of the target image that can be detected often matches the size of the selection area. Therefore, in order to detect target images of different sizes, there are a method of detecting the input image by enlarging and reducing it, and a method of converting the learning data used by the detector so as to correspond to detection of target images of different sizes. Proposed.

  Thus, in order to detect the target image without omission, it is necessary to perform detection in consideration of parameters (detection parameters) such as the in-plane rotation angle and size of the target image included in the input image. However, when performing detection in consideration of a large number of detection parameters, it is necessary to perform detection for one input image many times while changing the detection parameters, and the detection time increases in proportion to the number of detection parameters. There is. In addition, if the detectors to which different detection parameters are assigned are parallelized, the detection process can be speeded up, but there is a problem that the scale of the detector increases.

  Therefore, as described in Patent Document 1, a method of performing detection at high speed by limiting detection parameters has been proposed. In the method of Patent Document 1, the rotation angle of the face image once detected is set as the rotation angle used in the subsequent detection, and detection of other rotation angles is omitted, thereby speeding up the detection process.

JP 2007-110575 A

P. Viola and M. Jones, “Rapid object detection using a boosted cascade of simple features”, Proc. Of CVPR, 2001.

  However, in the conventional method of Patent Document 1, the detection can be speeded up by limiting the detection parameters based on the detection result, but the target image that can be detected with the omitted detection parameters. Therefore, there is a problem that omission of detection of the target image occurs. The present invention has been made to solve such a problem, and an object of the present invention is to provide an image detection apparatus capable of detecting a target image with little leakage and at a high speed.

  In order to solve the above-described problem, the image detection apparatus according to the present invention has an image selection unit that selects a selection image that is a part of an input image, and a detection condition that includes a detection position of a target image to be detected. Based on the detected parameters, an image detection unit that detects the target image from the selected image selected by the image selection unit, and a priority region that includes the target image detected by the image detection unit And a priority application detection parameter storage unit that stores a priority application detection parameter preferentially set to the detection parameter of the image detection unit, and a detection result of the target image detected by the image detection unit. Based on the priority application detection parameter for updating the priority area and the priority application detection parameter stored in the priority application detection parameter storage means. When a detection parameter candidate obtained by changing the detection parameter of the image update unit and the meter detection unit is generated, and a detection position of the detection parameter candidate is included in the priority area stored in the priority application detection parameter storage unit, The priority application detection parameter associated with the priority area is set as the detection parameter, and when the detection position of the detection parameter candidate is not included in the priority area, the detection parameter candidate is set as the detection parameter. Thus, a detection parameter control means for controlling the detection parameter of the image detection means is provided.

According to the present invention, the priority application detection parameter list is provided so that the detection parameter can be specified for each priority area, so that the detection areas listed in the priority application detection parameter list are registered. By using detection parameters, the detection process can be speeded up, and detection areas that are not listed are detected using various detection parameters, so that detection with less leakage can be performed. This has the effect of becoming
In addition, when the target image cannot be detected in the priority area, it is possible to continue to detect the target image once detected by changing the detection parameter and follow the target to be detected. There is an effect that it is possible to detect the target without interruption.

It is a block diagram which shows an example of the image detection apparatus in Embodiment 1 of this invention. 4 is a diagram illustrating a flowchart of image detection processing (first frame) in the image detection apparatus according to Embodiment 1. FIG. It is a figure which shows an example of the selection image which the image selection means 1 selected from the input image of the 1st frame. It is a figure which shows an example of the result of having detected the object image from the selection image (ABCD). FIG. 5 is a diagram showing an example of a priority application detection parameter list updated based on the detection result of the target image shown in FIG. 4. It is a figure which shows an example of the "priority area | region" registered into a priority application detection parameter list | wrist in the 1st frame. 6 is a diagram illustrating a flowchart of image detection processing (second and subsequent frames) in the image detection apparatus according to Embodiment 1. FIG. It is a figure which shows an example of the detection parameter which the detection parameter control means 6 changed in the 2nd frame. It is a figure which shows an example of the result of having detected the target image from the selection image (EFGH). It is a figure which shows an example of the priority application detection parameter list updated based on the detection result of the target image shown in FIG. It is a figure which shows an example of the "priority area | region" registered into a priority application detection parameter list in the 2nd frame. It is a figure which shows an example of the detection parameter which the detection parameter control means 6 changed in the 3rd frame. FIG. 11 is a diagram showing two priority areas registered in the priority application detection parameter list shown in FIG. 10. FIG. 10 is a diagram illustrating a flowchart of an image detection process that follows a target in the image detection apparatus according to the second embodiment.

1 image selection means, 2 image detection means, 3 detection result output means, 4 priority application detection parameter storage means, 5 detection parameter control means.

Embodiment 1 FIG.
FIG. 1 is a configuration diagram illustrating an example of an image detection apparatus according to Embodiment 1 of the present invention.
In FIG. 1, the image selection unit 1 selects a partial region of the input image according to detection parameter values such as “detection position”, “detection region size”, and “detection angle” specified by the detection parameter control unit 6. Then, the selected area is cut out as a selected image, and the selected image is transferred to the image detection means 2.

  The image detection unit 2 detects a target image from the selected image transferred by the image selection unit 1 by a detector including learning data generated by machine learning using a sample image in advance.

  The detection result output unit 3 outputs the detection result to a display unit (not shown) based on the target image detected by the image detection unit 2 and also outputs the detection result to a priority application detection parameter update unit 5 described later.

  The priority application detection parameter storage means 4 has detection parameters (priority application detection parameters) that are preferentially applied in the detection of the target image for detection parameters such as “detection position”, “detection region size”, and “detection angle”. The stored priority application detection parameter list is stored. Details of the priority application detection parameter list will be described later.

  The priority application detection parameter update unit 5 updates the priority application detection parameter list stored in the priority application detection parameter storage unit 4 based on the detection result output by the detection result output unit 3.

  The detection parameter control means 6 refers to the priority application detection parameter list stored in the priority application detection parameter storage means 4 and executes a detection sequence for detecting the target image while changing the detection parameters. The detection parameters are transferred to the image selection unit 1 and the image detection unit 2, respectively, and used for the cut-out process of the selected image of the image selection unit 1 and the detection process of the image detection unit 2.

  Next, the operation of the image detection apparatus according to the first embodiment will be described with reference to FIGS.

FIG. 2 is a flowchart of image detection processing (first frame) in the image detection apparatus according to the first embodiment.
When a target image is detected using each frame of a moving image such as a digital video camera as an input image, the detection parameter control means 6 controls the image selection means 1 and the image detection means 2 for each frame, and FIG. The image detection processing sequence is executed.

  First, in step S101, the detection parameter control means 6 detects three detections of “detection position”, “detection region size”, and “detection angle” in order to generate detection parameter candidates used when detecting the target image. Change the parameter. The order in which the detection parameters are changed is arbitrary. Hereinafter, these detection parameters will be described in detail.

  “Detection position”, which is one of the detection parameters, sets coordinates for specifying the position of the area to be detected. For example, the upper left coordinates of the area to be detected and the coordinates of the center of the area are set. As a method of changing coordinates, a method of updating coordinates in the order of raster scanning, a method of updating coordinates in a spiral shape from the center of the screen, or the like can be used.

  “Detection area size” which is one of the detection parameters is set to specify the size (size) of the area to be detected. For example, the length of one side of the area to be detected, the ratio to the size of the input image, and the like are set. Further, the shape of the area to be detected need not be a square. For example, when the shape of the region is a rectangle, the lengths of the two sides of the long side and the short side are set, and when the shape of the region is a circle, the length of the radius may be set. In order to detect target images having different sizes, several types of detection area sizes are set. The order of setting these detection area sizes is arbitrary.

  “Detection angle” which is one of the detection parameters is a difference angle between the vertical direction of the input image and the vertical direction of the target image assumed at the time of detection. This difference angle is set for each fixed rotation angle. For example, when setting in units of 45 degrees, any one of 0 degree, 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees, and 315 degrees is set. The order in which these angles are set is arbitrary.

  Based on the detection parameter settings such as “detection position”, “detection area size”, and “detection angle” described above, the detection parameter control means 6 performs the following operations on the image selection means 1 and the image detection means 2. The three changed detection parameters are designated, and the execution of the image selection process of the image selection unit 1 and the image detection process of the image detection unit 2 is controlled.

  Note that the detection parameters need not be limited to the above three, and there may be other detection parameters. Further, it is not essential that the above three detection parameters are provided. For example, when the detector used in the image detection means 2 is a detector that can detect a target image rotated at any angle, the “detection angle” need not be specified and may be removed from the detection parameters.

  Next, in step S102, the image selection unit 1 determines a part of the input image according to the detection parameter values such as “detection position”, “detection region size”, and “detection angle” specified by the detection parameter control unit 6. An area is selected, the selected area is cut out as a selected image, and the selected image is transferred to the image detection means 2.

FIG. 3 is a diagram illustrating an example of a selected image selected from the input image of the first frame by the image selection unit 1.
In FIG. 3, the detection parameter control means 6 designates the coordinates A (X1, Y1) as the “detection position” and the length of one side of the detection area ABCD set as a square as the “detection area size”. This indicates that the image selection means 1 has cut out the detection area ABCD as a selected image when a certain L1 is designated and 0 degree is designated as the “detection angle” assuming that the Y axis is the vertical direction. is there. The image detection means 2 transfers the selected image cut out in this way to the image detection means 2.

  Next, in step S103, the image detection unit 2 detects a target image from the selected image transferred by the image selection unit 1. For the detection of the target image, a detector including learning data generated by machine learning using a sample image in advance is used. As the detection method, the aforementioned Viola-Jones algorithm, a face detection algorithm using a neural network, or the like is used.

FIG. 4 is a diagram illustrating an example of a result of detecting a target image from a selected image (ABCD).
In FIG. 4, a result of detecting a face image from the area ABCD of the selected image by the image detection unit 2 using the face image as a target image is schematically expressed.

  Here, an operation in the case of detecting a target image rotating in the input image plane from step S101 to step S103 will be described. In order to detect a rotating target image, for example, the detection parameter control means 6 designates a “detection angle” for each fixed rotation angle to the image selection means 1, and the image selection means 1 According to the “detection angle” designated by the detection parameter control means 6, the selected image is rotated.

  Alternatively, the detection parameter control unit 6 designates a “detection angle” for each fixed rotation angle with respect to the image detection unit 2, and the image detection unit 2 selects from the learning data prepared in advance for each rotation angle. Learning data matching the designated “detection angle” may be selected and used. Furthermore, the learning data learned so that the image detection means 2 can detect the target image rotated at a certain angle may be used by converting it according to the designated “detection angle”.

  Next, in step S104, the detection result output unit 3 detects “detection position”, “detection region size”, “detection angle” when the target image is detected based on the target image detected by the image detection unit 2. And the like as a detection result to a display means (not shown). The detection result output from the detection result output unit 3 is also transferred to the priority application detection parameter update unit 5.

  In addition, when the detection result output means 3 outputs a detection result, you may perform the process which deletes the duplication of the detection result with respect to the same target image. Furthermore, a process of deleting the detection of the target image that does not occur redundantly as detection may be performed.

  Next, in step S105, the priority application detection parameter update unit 5 receives the detection result output from the detection result output unit 3, and updates the priority application detection parameter list stored in the priority application detection parameter storage unit 4.

  The priority application detection parameter list stores detection parameters (priority application detection parameters) that are preferentially applied in the detection of target images after the next frame. In the detection process of the target image, the detection sequence may not end during the display period for one frame, the detection sequence may be interrupted in the middle, and the selected image to be detected may be updated to the selected image of the next frame. is there. For this reason, it is desirable that the detection parameter control unit 6 sequentially applies detection parameters that are highly likely to be detected so that the target image is detected as early as possible in the detection sequence. Such detection parameters are stored as priority application detection parameters in the priority application detection parameter list, and the target application detection is performed by preferentially using the priority application detection parameters in the detection of the target image after the next frame. Can be performed efficiently.

  As an example of a detection parameter that is highly likely to be detected, a detection parameter of a target image that is detected most recently can be used. This is because the detection target often does not move for a while, and thus the detected target image is highly likely to be detected with the same detection parameter in subsequent frames. For this reason, if detection is performed by setting the “detection position”, “detection region size”, and “detection angle” to the same or close values, there is a high possibility that the same detection target can be continuously detected. . Therefore, the detection parameters at the time of detection are stored in the priority application detection parameter list and used for detection in the subsequent frames.

  Alternatively, if the detected area size of the detected target image is large, it means that the detection target exists near the camera, and it is considered that it is often a target to be detected with priority. The detection parameter of the target image whose “detection area size” is equal to or larger than a predetermined threshold may be preferentially added and updated in the priority application detection parameter list.

FIG. 5 is a diagram showing an example of the priority application detection parameter list updated based on the detection result of the target image shown in FIG. FIG. 6 is a diagram illustrating an example of a “priority area” registered in the priority application detection parameter list in the first frame.
In the “priority area” of the priority application detection parameter list, an area including the vicinity of the position where the target image is detected, for example, a rectangular area is registered. In the case of the detection result of FIG. 4, since the target image is detected in the area ABCD having (X coordinate, Y coordinate) = (X1, Y1) as the upper left vertex, a large area including the area ABCD, For example, as shown in FIG. 6, rectangular areas from (X0, Y0) to (X2, Y2) are registered. Further, “detection angle” and “detection region size” when the target image is detected are registered in “detection angle” and “detection region size”, respectively.

  The priority application detection parameter update unit 5 registers all detection parameters of the detected target image in the priority application detection parameter list defined as described above. When the number of detected target images exceeds the number that can be registered in the priority application detection parameter list, the target images to be registered are narrowed down based on some priority, for example, the certainty of the detection result.

  The priority application detection parameter list is updated by the priority application detection parameter update unit 5, for example, every time the detection result is output by the detection result output unit 3. Alternatively, each time the input image is updated, the priority application detection parameter list may be initialized and re-registered.

  Next, in step S106, the detection parameter control means 6 determines whether or not all detection parameters have been applied to the plurality of detection parameters set in step S101. If NO, the process returns to step S101. Detection with the remaining detection parameters is executed. If YES, the detection process is terminated.

  In order to shorten the execution time of the detection sequence, the image selection unit 1 and the image detection unit 2 may be arranged in parallel. When these are arranged in parallel, a plurality of detections with different detection parameters can be performed simultaneously. For example, when the image detection means 2 is in parallel, the “detection angle” is set to 0 degree in the first image detection means 2 and the “detection angle” is set to 90 degrees in the other detection unit. If detection is performed, detection with “detection angle” of 0 degrees and 90 degrees can be performed simultaneously.

  Next, a case will be described in which the detection process for the input image of the first frame is completed and the detection process is continuously executed for the input images of the second and subsequent frames.

  FIG. 7 is a diagram illustrating a flowchart of image detection processing (second and subsequent frames) in the image detection apparatus according to the first embodiment.

First, in step S201, the detection parameter control means 6 generates three detection parameters “detection position”, “detection region size”, and “detection angle” in order to generate detection parameter candidates used when detecting the target image. To change.
FIG. 8 is a diagram showing an example of detection parameters changed by the detection parameter control means 6 in the second frame.
As shown in FIG. 8, the coordinates E (X3, Y3) is designated as the “detection position”, and L2 that is the length of one side of the detection area EFGH set by the square is designated as the “detection area size”. Furthermore, it is assumed that the detection parameter is changed so that 90 degrees is designated as the “detection angle” assuming that the Y axis is the vertical direction.

  Next, in step S202, the detection parameter control means 6 determines whether or not the “detection position” is included in the priority area stored in the priority application detection parameter list. Since the priority area stored in the priority application detection parameter list is only the rectangular area from (X0, Y0) to (X2, Y2) as shown in FIG. The coordinates E (X3, Y3) that is the “detection position” specified in S201 is determined not to be included in the priority area. As a result, the process proceeds to step S203 which is a branch destination of NO.

  Next, in step S203, the image selection unit 1 determines the input image of the second frame in accordance with detection parameter values such as “detection position”, “detection region size”, and “detection angle” specified by the detection parameter control unit 6. A detection area EFGH, which is a partial area of the image, is cut out as a selected image, and this selected image is transferred to the image detection means 2.

Next, in step S <b> 204, the image detection unit 2 detects a target image from the selected image transferred by the image selection unit 1.
FIG. 9 is a diagram illustrating an example of a result of detecting a target image from a selected image (EFGH).
In FIG. 9, coordinates E (X3, Y3) are designated as “detection position”, L2 which is the length of one side of the detection area EFGH is designated as “detection area size”, and the Y axis is The result of detecting the target image when 90 degrees is designated as the “detection angle” assuming the vertical direction is shown.

  Next, in step S205, the detection result output unit 3 detects “detection position”, “detection region size”, “detection angle” when the target image is detected based on the target image detected by the image detection unit 2. And the like as a detection result to a display means (not shown). The detection result output from the detection result output unit 3 is also transferred to the priority application detection parameter update unit 5.

  Next, in step S206, the priority application detection parameter update unit 5 receives the detection result output from the detection result output unit 3, and updates the priority application detection parameter list stored in the priority application detection parameter storage unit 4.

FIG. 10 is a diagram illustrating an example of the priority application detection parameter list updated based on the detection result of the target image illustrated in FIG. 9. FIG. 11 is a diagram illustrating an example of a “priority area” registered in the priority application detection parameter list in the second frame.
As shown in FIG. 10, detection parameter values of “detection position”, “detection region size”, and “detection angle” detected in step S205 are updated by additionally registering them in the priority application detection parameter list. . In FIG. 11, rectangular areas from (X4, Y4) to (X5, Y5) are shown as larger areas including the area EFGH, and the “detection angle” at the time of detecting the target image is 90 degrees, L2 of “detection area size” is shown.

  Next, in step S207, the detection parameter control means 6 determines whether or not all detection parameters have been applied to the plurality of detection parameters set in step S201. If NO, the process returns to step S201. Detection with the remaining detection parameters is executed. If YES, the detection process is terminated.

Next, detection processing for the input image of the third frame will be described.
First, in step S201, the detection parameter control means 6 changes three detection parameters of “detection position”, “detection area size”, and “detection angle”.
FIG. 12 is a diagram showing an example of detection parameters changed by the detection parameter control means 6 in the third frame.
As shown in FIG. 12, the coordinates E ′ (X6, Y6) is designated as the “detection position”, and one side of the detection area E′F′G′H ′ set as a square as the “detection area size” And the detection parameter is changed so that 90 degrees is designated as the “detection angle” assuming that the Y axis is the vertical direction.

  Next, in step S202, the detection parameter control means 6 determines whether or not the “detection position” of the changed detection parameter is included in the priority area stored in the priority application detection parameter list. In the priority application detection parameter list at the start of processing of the third frame, as shown in FIG. 10, the priority area is a rectangular area from (X0, Y0) to (X2, Y2), and (X4, Y4) to ( Two preferential application parameters with the rectangular area up to X5, Y5) are registered. FIG. 13 is a diagram showing two priority areas registered in the priority application detection parameter list shown in FIG.

  The detection parameter control means 6 refers to all the priority areas registered in the priority application detection parameter list, and includes the coordinates E ′ (X6, Y6) which is the “detection position” specified in step S201. It is determined whether or not there is a priority area. In the example shown in FIGS. 10 and 13, the coordinates E ′ (X6, Y6) are determined to be included in the priority area of the rectangular area from (X4, Y4) to (X5, Y5). As a result, the process proceeds to step S208, which is a branch destination of YES. On the other hand, if it is determined that the coordinate E ′ (X6, Y6) is not included in any priority area, the process proceeds to step S203, which is a branch destination of NO, and the following description is based on the second frame. Similarly to the above, the detection process is performed.

  Next, in step S208, the detection parameter control means 6 determines whether or not the changed detection parameter value is the same as or close to the priority application detection parameter value. In the example of FIG. 13, the “detection angle” of the changed detection parameter is prioritized in the priority region from (X4, Y4) to (X5, Y5) determined to include the coordinates E ′ (X6, Y6). It is the same as the application detection parameter of 90 degrees. At this time, if the “detection area size” L3 of the changed detection parameter is determined to be a value close to L2 of the priority application detection parameter, it is considered highly likely that the target image can be detected in the third frame. The process proceeds to step S203, which is the branch destination of YES. Since step S203 and subsequent operations are the same as those described in the second frame, description thereof is omitted.

  On the other hand, if the value of the changed detection parameter is not the same as or close to the value of the priority application detection parameter, it is considered unlikely that the target image can be detected. May be omitted. For example, when it is determined that the “detection area size” L3 of the changed detection parameter is not a value close to L2 of the priority application detection parameter, the process proceeds to step S207, which is a branch destination of NO, and from step S203 to step S206. This detection process is omitted. Thus, by omitting the detection process when it is considered that the target image is unlikely to be detected, the detection process can be speeded up.

  As described above, the detection parameter control means 6 continuously executes the processing from step S201 to step S207 in the third and subsequent frames as in the above processing.

  In updating the priority application detection parameter list, it is more efficient to update the priority application detection parameter list every time a detection result is output in step S205. This is because when a target image is detected in a certain area, the detection process using other detection parameters in that area is omitted in subsequent detection processes for the same frame. Therefore, if the detection process is performed in descending order of the “detection area size” and the detection result is registered in the priority application detection parameter list, the area where the detection process can be omitted is increased from the initial stage of the detection process. be able to. In this way, the detection process can be made more efficient.

  Further, in step S101 and step S201, when the detection parameter is changed and the search parameter candidate is generated, the detection parameter may be changed with reference to the priority application detection parameter list. For example, the priority area listed in the priority application detection parameter list is highly likely to be able to detect the target image. Therefore, the detection process is performed before the area not listed. In other words, detection parameters determined to be highly detectable based on past detection results are preferentially applied in the detection process. As a result, the target image can be detected as early as possible in the detection sequence, and even when the detection processing time is limited, it is possible to contribute to the efficiency of the detection processing.

  The contents listed in the priority application detection parameter list can also be used in the detection process when the area is not listed in the priority application detection parameter list. For example, the most frequently registered “detection angle” among the listed “detection angles” is highly likely to be detected by this “detection angle” even in an area that is not yet listed. Conceivable. This is because the angles of the other target images are likely to be the same as the most registered “detection angles”. In this way, the detection parameters of the areas not listed can be ranked in a majority manner based on the priority application detection parameter list.

  As described above, according to the first embodiment, the priority application detection parameter list is provided, and the detection parameters can be designated for each priority area, so that the detections listed in the priority application detection parameter list can be specified. By using registered detection parameters for a region, the detection process can be accelerated, and for detection regions that are not listed, detection is performed using various detection parameters. Therefore, there is an effect that detection without leakage can be performed.

Embodiment 2. FIG.
In Embodiment 1 described above, the detection parameters used for detection in the priority area are limited to increase the speed of the detection process. Next, in the detection using the priority application detection parameter in the priority area, the target is detected. Embodiment 2 relating to an image detection apparatus that determines that a detection target has moved when an image cannot be detected, continues detection while changing detection parameters, and follows the detection target will be described.

  Since the detection target often does not move for a while, if detection parameters such as `` detection position '', `` detection area size '' and `` detection angle '' are set to the same or close values, There is a high possibility that the same detection object can be detected continuously. However, with this method, the detection target may not be detected for a while at the moment the detection target moves. For example, when the detection target moves so as to approach the camera direction, the ratio of the target image to the input image increases, and if the “detection area size” is not set larger than before, the detection target is continuously detected. Can not. Once the detection target cannot be detected, the detection target is excluded from the priority application detection parameter list. Therefore, detection is performed again with detection processing using a detection parameter that can detect the detection target after movement in the subsequent frame detection. Until detected, the detection target is in an undetected state. Therefore, in the second embodiment, if the target cannot be detected by the detection using the priority application detection parameter in the priority area, it is determined that the target has moved, and the detection is continued while changing the detection parameter in the priority area. Follow the subject.

Hereinafter, the operation of following the target in the second embodiment will be described.
FIG. 14 is a diagram illustrating a flowchart of an image detection process for tracking a target in the image detection apparatus according to the second embodiment.

In FIG. 14, the processes other than step S309 and step S310 are the same as those in the flowchart of FIG.
In step S309, the detection parameter control means 6 controls the image detection means 2 and detects the target image from the selected image transferred by the image selection means 1 in step S308 when it is determined as the priority area in step S302. To do. As a result, when the target image cannot be detected, the detection parameter control means 6 detects the three detection parameter values of “detection position”, “detection region size”, and “detection angle” as detection parameters that have not yet been attempted to be detected. The value is changed to a value and the target image is detected again. The detection parameter control means 6 continues this process, ends the detection process when the target image is detected, and proceeds to step S310. On the other hand, if the target image cannot be detected even when a predetermined combination of detection parameter values is applied, there is a high possibility that the target has moved out of the priority area, so the detection process ends and the process proceeds to step S310.

  Next, in step S310, when the target image is detected, the detection parameter control means 6 advances the process to step S305 by branching YES, and when the target image is not detected, step S307 is performed by branching NO. Proceed to the process. The subsequent processing is the same as that in FIG.

  In step S309, a history of detection results of the target image may be left, and the priority application detection parameter list may be updated while predicting the operation of the detection target with reference to the history. For example, when it is known from the history of detection results that the detection target is moving in the right direction, a process of shifting the detection parameter application area in the right direction may be performed.

  As described above, according to the second embodiment, when the target image cannot be detected in the priority area, the target image detected once is emphasized by following the target to be detected while changing the detection parameter. Thus, it is possible to continuously detect a small number of objects and to detect a small number of objects without interruption.

Claims (7)

Image selection means for selecting a selection image that is part of the input image;
An image detecting unit for detecting the target image from the selected image selected by the image selecting unit based on a detection parameter in which a detection condition including a detection position of the target image to be detected is set;
A priority application detection parameter that associates and stores a priority area that is an area including the target image detected by the image detection means and a priority application detection parameter that is preferentially set to the detection parameter of the image detection means Storage means;
Based on the detection result of the target image detected by the image detection means, priority application detection parameter update means for updating the priority area and the priority application detection parameter stored in the priority application detection parameter storage means;
When a detection parameter candidate is generated by changing the detection parameter of the image detection means, and the detection position of the detection parameter candidate is included in the priority area stored in the priority application detection parameter storage means, this priority area is supported If the detection application candidate is not included in the priority area, the detection parameter candidate is set as the detection parameter. An image detection apparatus comprising detection parameter control means for controlling the detection parameter of the means. The image detection means detects the target image from an input image based on a detection parameter in which a detection condition including a detection region size indicating a size of a region for detecting the target image is set,
The detection parameter control unit is configured to detect the priority application based on the detection result of the target image whose detection area size is equal to or larger than a predetermined threshold among the detection results of the target image detected by the image detection unit. The image detection apparatus according to claim 1, wherein the priority application detection parameter stored in the parameter storage unit is updated. The image detection apparatus according to claim 2, wherein the detection parameter control means sets the detection parameter candidates as the detection parameters of the image detection means in descending order of the detection area size. The detection parameter control means preferentially sets the detection parameter candidate changed to the detection position included in the priority area of the priority application detection parameter stored in the priority application detection parameter storage means to the detection parameter,
The said image selection means selects the selection image which is a part of input image based on the said detection parameter set by the said detection parameter control means preferentially. Image detection device. When the detection position of the detection parameter candidate obtained by changing the detection parameter of the image detection unit is not included in the priority area stored in the priority application detection parameter storage unit, the detection parameter control unit performs the priority application detection. The image detection according to any one of claims 1 to 4, wherein a detection condition of the detection parameter candidate is changed based on an appearance frequency of each detection condition of the parameter, and the detection parameter candidate is set as the detection parameter. apparatus. When the detection position of the detection parameter candidate is included in the priority area, the detection parameter control unit sets the priority application detection parameter when the detection result of the target image by the image detection unit is undetected. The detection condition of the detected parameter is changed to an unused detection condition in the detection in the priority area,
The image detection device according to claim 1, wherein the image detection unit redetects the target image from the priority area based on the detection parameter changed by the detection parameter control unit. The detection parameter control means calculates a predicted movement position that predicts a position after movement of the target image based on a history of past detection results of the target image detected by the image detection means, and the detection parameter The image detection apparatus according to claim 6, wherein the detection position is changed to the predicted movement position.

Images