JP2014199506A - Object detection device, object method of detection, and program - Google Patents

Object detection device, object method of detection, and program Download PDF

Info

Publication number
JP2014199506A
JP2014199506A JP2013073936A JP2013073936A JP2014199506A JP 2014199506 A JP2014199506 A JP 2014199506A JP 2013073936 A JP2013073936 A JP 2013073936A JP 2013073936 A JP2013073936 A JP 2013073936A JP 2014199506 A JP2014199506 A JP 2014199506A
Authority
JP
Japan
Prior art keywords
detection
region
human
hidden
detector
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
JP2013073936A
Other languages
Japanese (ja)
Inventor
睦凌 郭
Bokuryo Kaku
睦凌 郭
矢野 光太郎
Kotaro Yano
光太郎 矢野
直嗣 佐川
Naotada Sagawa
直嗣 佐川
Original Assignee
キヤノン株式会社
Canon Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by キヤノン株式会社, Canon Inc filed Critical キヤノン株式会社
Priority to JP2013073936A priority Critical patent/JP2014199506A/en
Publication of JP2014199506A publication Critical patent/JP2014199506A/en
Pending legal-status Critical Current

Links

Images

Abstract

An object of the present invention is to accurately detect a shielded human body and improve the detection speed. A front detector 104 detects a head Ω by applying a head Ω detection model from each pyramid image. Next, the hidden area setting unit 106 hides the hidden area based on the position and size of the head Ω in the layer to be processed and the size (Hd2, Wd2) of the whole body detection model used in the post-stage detector 110. Set Γ1. Then, the latter-stage detector application determination unit 108 determines whether or not the application area excluding the hidden area Γ1 is the whole body detection model application area, and determines whether or not the latter-stage detector 110 is a human body. [Selection] Figure 1

Description

  The present invention particularly relates to an object detection apparatus, an object detection method, and a program suitable for use in detecting a shielded object.

  In recent years, intelligent video equipment has become widespread. For example, a human body detection function is installed in a surveillance camera, and a function for counting the number of people, analyzing a customer's intention, detecting an abnormal operation, or detecting an intrusion into a dangerous area has been proposed. In addition, a function for automatically controlling the focus and exposure of a camera by identifying and tracking the position of a person from an image captured by an imaging device such as a digital camera is also attracting attention. In addition to people, attention is also paid to general objects such as dogs, cats, and flowers, and a function for automatically controlling the camera is also popular.

  The foundation of intelligent video equipment is object detection technology based on machine learning. In machine learning, a feature amount that distinguishes an object from a non-object is extracted from a large number of objects and non-object learning samples, and a recognition model is created. When detecting an object from an image, a pyramid image layer is created by scaling the size of the original image. Then, a raster scan is performed on each pyramid image layer, and objects of different sizes are detected by combining the discriminator responses of each feature amount described in the recognition model.

  On the other hand, when the objects overlap each other, the detection rate of the learned object detector for the shielded objects decreases, and false detection also increases. To solve this problem, the method described in Non-Patent Document 1 uses the whole body of the human body as a detection target and divides the whole body region into 7 × 15 blocks. Then, in each block, the HOG feature value and the LBP feature value are obtained, and the SVM coefficient and offset are obtained. When it is determined by the whole body detector that the human body is not a human body, the likelihood of the human body is obtained for each block, and a likelihood map is created. And shielding is determined based on a likelihood map, and it is determined whether it is a human body with the human body part detector.

  In the method described in Patent Document 1, when detecting a human body, it is first determined whether or not the human body is based on the upper body feature amount. When it is determined that the human body is based on the upper body feature amount, it is not determined whether the human body is based on the whole body feature amount. On the other hand, if the feature quantity of the upper body cannot be determined as the human body, it is re-determined whether the human body is the feature quantity of the whole body.

  Further, in the method described in Patent Document 2, the second object is obtained by extracting the object by the first extraction unit that satisfies the first condition and relaxing the first condition for a region (shielding region) including a part of the extracted region. Conditions are set and the object is determined by the second extraction means. At this time, the degree to which the condition is relaxed is set by the ratio of the overlapping area to the hidden candidate area. In this way, the detection rate of the shielding area can be improved.

JP 2011-53959 A JP 2011-186633 A

X. Wang, T .; X. Han, and S.H. Yan. "An HOG-LBP Human Detector with Partial Occlusion Handling," In ICCV, 2009. NavneetDalal and Bill Triggs, "Histogram of Oriented Gradients for Human Detection," CVPR2005.

  In the detection method described in Non-Patent Document 1, after applying a whole-body detector to a window for scanning, if it is not determined to be a human body, a likelihood map is created to determine shielding. And when it determines with shielding, it is determined with a human body parts detector whether it is a human body. As described above, the detection method described in Non-Patent Document 1 has a problem that the detection speed is slow because the whole body detector is applied to all the scan windows.

  Further, in the detection method described in Patent Document 2, when the human body cannot be detected with the upper body feature amount in a state where the human body is shielded, it is determined whether the human body is based on the whole body feature amount. In this method, the detection speed is reduced with respect to a shielded human body, and erroneous detection may be caused. Furthermore, in the detection method described in Patent Document 3, the detection rate is improved because the condition of the detector is relaxed with respect to the shielded human body, so that the detection rate can be improved. However, there is a problem that false detection increases. is there.

  The present invention has been made in view of the above-described problems, and an object of the present invention is to detect a shielded human body with high accuracy and to improve the detection speed.

  The object detection device according to the present invention is set by a preceding detection means for detecting a specific partial region of an object, a setting means for setting a hidden region based on a distribution of detection results of the preceding detection means, and the setting means A post-stage detection means for detecting an area different from the specific partial area of the object so as not to include a hidden area; an integration means for integrating the detection result of the pre-stage detection means and the detection result of the post-stage detection means; It is characterized by having.

  According to the present invention, it is possible to accurately detect a shielded human body and improve the detection speed.

It is a block diagram which shows the function structural example of the human body detection apparatus which concerns on the 1st Embodiment of this invention. 5 is a flowchart illustrating an example of a processing procedure for detecting a shielded human body in the first embodiment of the present invention. It is a figure for demonstrating the position of the detected head and hidden area in a pyramid image. It is a figure for demonstrating the procedure which sets a hidden area. It is a figure for demonstrating the hidden area and application area | region of the whole body detection model in a pyramid image. It is a figure for demonstrating the misdetection in case a part of human body is hidden. It is a figure for demonstrating the outline | summary of the human body detection method in the 1st Embodiment of this invention. It is a block diagram which shows the function structural example of the human body detection apparatus which concerns on the 2nd Embodiment of this invention. It is a figure for demonstrating the position of the hidden area | region of the detected head in each pyramid image and each detection model. It is a figure for demonstrating the position of the final application area | region of the detected head and each detection model in a pyramid image. It is a flowchart which shows an example of the process sequence which detects the shielded human body in the 2nd Embodiment of this invention. It is a block diagram which shows the function structural example of the human body detection apparatus which concerns on the 3rd Embodiment of this invention. It is a figure which shows the detection result of the head of the human body in the 3rd Embodiment of this invention. It is a block diagram which shows the hardware structural example of the human body detection apparatus in embodiment of this invention.

  Hereinafter, the present invention will be described in detail according to preferred embodiments with reference to the accompanying drawings. In addition, this invention is not limited to the following embodiment, It shows only the specific example advantageous for implementation of this invention. What implement | achieves the function used as the characteristic of this invention is also contained in the scope of the present invention.

(First embodiment)
FIG. 14 is a block diagram illustrating a hardware configuration example of the human body detection device 100 according to the present embodiment.
In FIG. 14, the CPU 1401 executes various controls in the human body detection device 100 of the present embodiment. The ROM 1402 stores a boot program and various data that are executed when the human body detection device 100 according to the present embodiment is started up. The RAM 1403 stores a control program for processing by the CPU 1401 and provides a work area when the CPU 1401 executes various controls. A keyboard 1404 and a mouse 1405 provide various input operation environments for the user.

  The external storage device 1406 is composed of a hard disk, flexible disk, optical disk, magnetic disk, magneto-optical disk, magnetic tape, or the like. However, the external storage device 1406 is not necessarily a necessary component as long as all the control programs and various data are stored in the ROM 1402. The display unit 1407 is configured with a display or the like, and displays the result and the like to the user. The network interface 1408 is an interface for connecting to an external device. The video interface 1409 captures a frame image via an imaging unit (not shown) and a coaxial cable. A system bus 1411 is a bus for connecting the above-described components.

FIG. 1 is a block diagram illustrating a functional configuration example of the human body detection device 100 according to the present embodiment.
In FIG. 1, the human body detection device 100 includes an image acquisition unit 101, a feature extraction unit 102, a feature amount storage unit 103, a pre-stage detector 104, a detection result storage unit 105, a hidden area setting unit 106, a scan. And an area setting unit 107. Furthermore, a post-stage detector application determination unit 108, a whole body detection model 109, a post-stage detector 110, and a result integration unit 111 are provided.

In the following, the flow of processing by each configuration shown in FIG. 1 will be described in more detail with reference to the flowchart of FIG. FIG. 2 is a flowchart illustrating an example of a processing procedure for detecting a shielded human body in the present embodiment.
First, in step S201, the image acquisition unit 101 acquires one frame image from an image capturing device or an image storage device (not shown).

  Next, in step S <b> 202, the feature extraction unit 102 extracts a feature amount used by the upstream detector 104 and the downstream detector 110 for the entire acquired frame image. In the present embodiment, the HOG feature value described in Non-Patent Document 2 will be described as an example of the feature value used for the front detector 104 and the rear detector 110. In step S 203, the extracted feature quantity is stored in the feature quantity storage unit 103 that is a part of the external storage device 1406.

  Next, in step S204, the upstream detector 104 sequentially scales the acquired frame image with a predetermined magnification to create a pyramid image. In step S205, the upstream detector 104 controls to repeat the processing from step S206 to step 208 for each layer of the pyramid image of the frame image created in step S204.

  In step S206, the pre-stage detector 104 performs a raster scan of the detection window on each layer of the pyramid image, and controls so that the processing of step S207 described below is repeatedly performed for each detection window.

  In step S207, the upstream detector 104 applies the head Ω detection model that is the upstream detection model to the detection window, and determines whether or not the current detection window is the head. Here, the head Ω detection model is a detection model in which the detection region is an Ω shape formed by the head and shoulder contours, which are part of the upper body of the human body, centered on the head of the human body. 1406 is stored. The pre-stage detector 104 calculates a HOG feature vector from a detection window with a certain layer of the pyramid image according to a head Ω detection model indicating a specific partial region, and calculates the response of the detector. It is determined whether the detection window is the head by comparing the detector response with a threshold. It is assumed that the head Ω detection model is obtained by prior machine learning.

  In step S208, the upstream detector 104 controls to perform the process of step S207 on all detection windows in response to step S206. In step S209, the pre-stage detector 104 performs control so that the processing from step S206 to step S208 is performed for all pyramid layers corresponding to step S205.

  Next, in step S <b> 210, the upstream detector 104 converts the detection results of the respective layers into a frame image coordinate system and stores them in the detection result storage unit 105 which is a part of the external storage device 1406.

  In the present embodiment, a hidden region setting unit 106, a scan region setting unit 107, and a rear detector application determination unit 108 are provided as means for controlling the rear detector 110. The control means of these post-stage detectors 110 reads out the whole body detection model 109 stored in the external storage device 1406 and controls whether to apply it to each scan window according to the hidden area. This control is performed in the following processing.

  First, in step S <b> 211, the latter-stage detector application determination unit 108 reads the whole body detection model 109 from the external storage device 1406. In step S212, the scan region setting unit 107 performs control so that the processes in steps S213 to S219 described below are repeatedly performed for each layer of the pyramid image of the frame image.

  In step S213, the hidden region setting unit 106 converts the detection results of all head Ω shapes (hereinafter referred to as head Ω) stored in the detection result storage unit 105 into the coordinate system of the current layer. Calculate the position coordinates and size of the detection result of the head Ω.

For example, as shown in FIG. 3, in the 0th layer (frame image) of the pyramid image, the upper left point of the detected head Ω is defined as (c f1 , r f1 ) with the upper left point of each layer as the origin. The lower right position is (c f2 , r f2 ), and the scaling factor of the pyramid image is s. Therefore, the position of the head Ω in the j-th layer (the upper left position (c 1 , r 1 ) and the lower right position (c 2 , r 2 )) can be calculated by the following equation (1).
c 1 = c f1 · s j
r 1 = r f1 · s j
c 2 = c f2 · s j
r 2 = r f2 · s j (1)

Further, the size (vertical width H and horizontal width W) of the head Ω in the j-th layer can be calculated from the following equation (2).
H = r 2 −r 1
W = c 2 −c 1 (2)

Next, in step S214, the hidden region setting unit 106 determines from the position and size of the head in the processing target layer and the size of the whole body detection model (vertical width H d2 , horizontal width W d2 ) used for the post-stage detector 110. The hidden region Γ 1 of the whole body detection model is set according to the equation (3). Here, α, β, γ, and δ are positive or negative real numbers, which are coefficients determined by tuning.
x 1 = c 1 + α · W d2
y 1 = r 1 + β · H d2
x 2 = c 2 + γ · W d2
y 2 = r 2 + δ · H d2 (3)

FIG. 4 is a diagram for explaining a mechanism for setting the hidden region Γ 1 of the whole body detection model. When the position of the head Ω is detected from the frame image, the corresponding position and size of the human body are estimated, and the position and size of the human body are converted into a pyramid layer for scanning. As shown in FIG. 4A, when the whole body detection model overlaps with the human body estimated from directly above with a predetermined degree of overlap, the upper position of the hidden region Γ 1 is determined. Further, as shown in FIG. 4B, when the whole body detection model holds the determined upper position and overlaps with the human body estimated from the left with a predetermined overlapping degree, the left position of the hidden region Γ 1 is determined. decide. Similarly, as shown in FIG. 4C, when the whole body detection model holds the determined upper position and overlaps with the human body estimated from the right with a predetermined overlapping degree, the right position of the hidden region Γ 1 To decide. As a result, as shown in FIG. 4 (d), the position above the predetermined height from the position of the detected head Ω in proportion to the height of the whole body detection model is set as the lower position of the hidden region Γ 1. To do. The hidden area Γ 1 determined in this way is indicated by a black area in FIG.

  Also, the black area at the bottom of FIG. 3 is not a hidden area set from the detected position of the head Ω but an area where the lower half of the human body is cut by the boundary of the image. Therefore, only the head Ω detector is applied in this region. The setting of this area is determined from the size of the front detector 104 or the rear detector 110.

Next, in step S215, the hidden region setting unit 106 sets a region other than the black region as shown in FIG. 3 as the application region Λ 1 of the whole body detection model. In step S216, the scan area setting unit 107 performs control so that the processes in steps S217 and S218 described below are repeatedly performed for all detection windows scanned in the current layer.

  First, in step S217, the post-stage detector application determination unit 108 determines whether or not the coordinates of the upper left vertex of the current detection window are in the application region of the whole body detection model. As a result of this determination, if the coordinates of the upper left vertex of the current detection window are in the application area, the process of step S218 is performed. On the other hand, if the coordinates of the upper left vertex of the current detection window are not in the application area, the process proceeds to step S219.

  In step S <b> 218, the post-stage detector application determination unit 108 loads the read whole body detection model 109 to the post-stage detector 110 for the current detection window. Then, the post-stage detector 110 uses the loaded whole body detection model 109 to determine whether or not the current detection window is a human body.

  In step S219, corresponding to step S216, the scan region setting unit 107 performs control so that the processes in steps S217 and S218 are repeated for all detection windows scanned in the current layer. In step S220, the scan region setting unit 107 controls to repeat the processing from step S213 to step S219 for all layers of the pyramid image of the frame image corresponding to step S212.

  Next, in step S221, the result integration unit 111 groups the detection results of the upstream detector 104 and integrates the adjacent detection results as the same head Ω. Similarly, the result integration unit 111 groups the detection results of the post-stage detector 110 and integrates the detection results in the vicinity as the same human body. Further, the integrated head Ω and the integrated human body are output as the final detection result.

  In the present embodiment, the method for setting the hidden area has been described based on the detection result of one head Ω in steps S213 and S214. When there are one or more head Ω detection results, as shown in FIG. 5, for each head Ω detection result, each hidden region is set, and each hidden region is ORed. The whole hidden area. The whole body detector in the subsequent stage is applied to the region other than the entire hidden region.

In this embodiment, in step S214, the hidden region setting unit 106 determines the position and size of the head Ω in the layer to be processed and the size (H d2 , W d2 ) and the hidden area are set. In addition to this, the hidden region setting unit 106 determines the position and size of the head Ω in the current layer and the size (H d1 , A hidden area can also be set from W d1 ). In this case, the hidden area is calculated by the following equation (4), and each parameter is determined by tuning.
x 1 = c 1 + α · W d1
y 1 = r 1 + β · H d1
x 2 = c 2 + γ · W d1
y 2 = r 2 + δ · H d1 (4)

Further, the hidden region setting unit 106 determines the position and size of the head Ω in the layer to be processed, the size of the head Ω detection model (H d1 , W d1 ), and the size of the whole body detection model (H d2 , W It is also possible to set the hidden area from both of d2 ). In this case, the hidden area is calculated by the following equation (5), and each parameter is determined by tuning.
x 1 = c 1 + α 1 · W d1 + α 2 · W d2
y 1 = r 1 + β 1 · H d1 + β 2 · H d2
x 2 = c 2 + γ 1 · W d1 + γ 2 · W d2
y 2 = r 2 + δ 1 · H d1 + δ 2 · H d2 (5)

  Further, in the present embodiment, the feature amounts used by the upstream detector 104 and the downstream detector 110 are all HOG feature amounts, but different feature amounts may be used. In the present embodiment, for the sake of simplicity, the human body region is a rectangular region. Actually, the human body region is not a bounding box, but can be approximated by an inscribed ellipse. The shape of the human body can also be described. The degree of overlap can be calculated according to the detailed shape of the human body, and more detailed hidden areas can be set according to the degree of overlap. In the present embodiment, the front-stage detector 104 is a detector of the Ω-type region of the head and shoulders, which is the main region of the human body, but may be a detector of another part, for example, an upper body detector.

  FIG. 6 is a diagram illustrating an example of a situation in which erroneous detection is likely to occur in a post-stage detector using a conventional whole body detection model. As shown in FIG. 6 (a), the human body C hidden behind the human body B is often undetected because the lower body part does not resemble the human body. In addition, as shown in FIG. 6B, the upper body of the human body C hidden behind the human body B often mixes with the lower body of the human body B, resulting in false detection. Further, the erroneously detected human body C may be integrated with the human body B by the result integration unit 111 and output as an incorrect human body.

  FIG. 7 is a conceptual diagram showing the effect of this embodiment. As shown in FIG. 7, the head Ω of the human body B and the human body C can be detected by the upstream detector 104. Then, the hidden area Γ is set by the hidden area setting unit 106. The post-stage detector application determination unit 108 reads the detection result of the pre-stage detector 104 for the hidden region Γ. On the other hand, the human body A and the human body B can be detected while preventing the human body C from being erroneously detected by reading the whole body detection model 109 and applying the post-stage detector 110 for the region other than the hidden region Γ. Furthermore, when the result integration unit 111 integrates the detection results of the head Ω and the whole body, all three persons can be detected. As shown in FIG. 7, according to the present embodiment, the head Ω of the human body C can be detected and the detection rate can be improved while reducing the false detection of the whole body of the human body C.

(Second Embodiment)
In the first embodiment, the post-stage detector application determination unit 108 reads only the whole body detection model 109 and loads it to the post-stage detector 110. In the present embodiment, as shown in FIG. 8, detection models to be loaded on the subsequent detector 110 include an upper body detection model 802 and a head circular detection model 803 in addition to the whole body detection model 801. In the present embodiment, the human body detection device 800 will be described as having the configuration shown in FIG. In addition, about the structure similar to FIG. 1, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  In this embodiment, each hidden area can be set for each detection model, as in the first embodiment. Only the detection result of the pre-stage detector 104 can be applied to each set hidden area, and each detection model can be applied to areas other than the hidden area.

  FIG. 9 is a diagram for explaining a hidden region setting method. The detection results of the post-stage detector 110 can be integrated with the detection results of the pre-stage detector 104 and output by the result integration unit 111. In this hidden region setting method, detection takes a little time, but the detection rate of the human body can be further increased.

FIG. 10 is a diagram for explaining a method of setting application areas of three detection models. For each detection model, after setting each hidden region by the method shown in FIG. 9, first, the application region Λ 1 of the whole body detection model 801 is set as the application region ψ 1 of the final whole body detection model 801. Further, an AND region of the application region Λ 2 of the upper body detection model 802 and the hidden region Γ 1 of the whole body detection model 801 is set as an application region ψ 2 of the final upper body detection model 802. Furthermore, an AND region of the application region Λ 3 of the head circular detection model 803 and the hidden region Γ 2 of the upper body detection model 802 is set as an application region ψ 3 of the final head circular detection model 803. Further, the rear detector 110 is not applied to the hidden region Γ 3 of the head circular detection model.

FIG. 11 is a flowchart illustrating an example of a processing procedure for setting an application area of a detection model in the present embodiment. Note that the processing from step S201 to step S211 and the processing of step S213 are the same as those in FIG. 2 described in the first embodiment, and thus description thereof is omitted.
In step S1114 to step S1116, the hidden area setting unit 106 sets each hidden area for each detection model. The setting method is the same as in the first embodiment, but the parameter may differ depending on the detection model. This process, hidden area gamma 1 systemic detection model 801, to set a hidden area gamma 3 of the hidden area gamma 2, and head circular detection model 803 of the upper body detection model 802. Thus, as each application area, application area lambda 1 of the whole body detection model 801, sets the application area lambda 3 of application areas lambda 2, and a head circle detection model 803 of the upper body detection model 802.

Next, in step S1117 to step 1119, the hidden area setting unit 106 sets the final application area of each detection model from the set hidden area and application area of each detection model. Specifically, the application area Λ 1 of the whole body detection model 801 is set as the application area Ψ 1 of the final whole body detection model 801. Then, an AND region of the application region Λ 2 of the upper body detection model 802 and the hidden region Γ 1 of the whole body detection model 801 is set as an application region ψ 2 of the final upper body detection model 802. Furthermore, an AND region of the application region Λ 3 of the head circular detection model 803 and the hidden region Γ 2 of the upper body detection model 802 is set as an application region ψ 3 of the final head circular detection model 803.

  Next, in step S1120, the scan region setting unit 107 performs control so that the processing from step S1121 to step S1124 is repeated for all detection windows scanned in the current layer.

First, in step S1121, the post-stage detector application determination unit 108 determines in which application region the current detection window to be scanned is. As a result of this determination, if the current detection window to be scanned is in the application region Ψ 1 of the final whole body detection model 801, the process proceeds to step S1122. If the current scanning detection window is in the application region Ψ 2 of the final upper body detection model 802, the process proceeds to step S1123. If the current scanning detection window is in the application region Ψ 3 of the final head circular detection model 803, the process proceeds to step S1124. On the other hand, if the current detection window to be scanned is in the hidden region Γ 3 of the head circular detection model 803, the process proceeds to step S1125.

  In step S1122, the post-stage detector 110 uses the whole body detection model 801 to determine whether the current detection window to be scanned is a human body. On the other hand, in step S1123, the post-stage detector 110 uses the upper body detection model 802 to determine whether the current detection window to be scanned is a human body. In step S1124, the post-stage detector 110 uses the head circular detection model 803 to determine whether the current detection window to be scanned is a human body.

  In step S1125, the scan region setting unit 107 performs control so as to repeatedly perform the processing in steps S1121 to S1124 for all detection windows scanned in the current layer, corresponding to step S1120.

  Next, in step S1126, the scan region setting unit 107 controls to repeat the processing from step S213 to step S1125 for each layer of the pyramid image of the frame image, corresponding to step S1112.

  Next, in step S1127, the result integration unit 111 groups the detection results of the upstream detector 104, and integrates the detection results in the vicinity as the same head Ω. Similarly, the result integration unit 111 performs grouping for each detection result of the post-stage detector 110 and integrates adjacent detection results as the same human body. Further, the integrated head Ω and the integrated detection result are output as the final detection result.

  In the present embodiment, the post-stage detector 110 is configured by selecting from a plurality of detection models and applied to the detection window. On the other hand, a plurality of subsequent detectors can be prepared, and one can be selected and used by the subsequent detector application determination unit 108. For example, the whole body detector may be an adaboost detector, and the upper body detector may be an SVM detector.

(Third embodiment)
In the first and second embodiments, the hidden region and the application region are set mainly from the single head Ω detection result and the size of the detection model. As shown in FIG. 12, in the present embodiment, a left half body detection model 1204 and a right half body detection model 1205 are further added as detection models compared to the second embodiment. Further, the application areas of these two models are set not from the detection result of a single head Ω but from the distribution of detection results of a plurality of heads Ω.

  FIG. 13 is a diagram for explaining a case where two human bodies overlap and both heads Ω are detected by the upstream detector 104. In the present embodiment, for example, when not only detecting the hidden human body C but also knowing the size of the human body C, the left half body detection model 1204 is applied to the human body C.

The regions of the two heads Ω detected by the upstream detector 104 are (c 11 , r 11 ), (c 21 , c 21 ), (c 12 , r 12 ), (c 22 , r 22 ), and the size of the left half body detection model is (H lf , W lf ). The left body detection model 1204 is applied when the positional relationship between the two head Ω regions satisfies the relationship of Expression (6). Here, α and β are parameters and are determined by tuning.
c 21 > c 12 > c 11
c 12 −c 11 > α · W lf
r 12 > r 11
r 12 −r 11 <β · H lf (6)

  Similarly, the application region of the right body detection model 1205 can be set from the detection result distribution of the upstream detector 104 and the size of the right body detection model 1205. Note that the processing procedure of this embodiment is the same as the flowchart of FIG. 11 described in the second embodiment, and the same processing is performed as the detection model increases. Therefore, detailed description is omitted.

(Other embodiments)
In each of the above-described embodiments, the human body has been described as an example. When detecting general objects such as cars, airplanes, dogs and cats in addition to the human body, first detection is performed using the most important parts detector (such as tire detection in a car) of these objects. To set a hidden area. In this case, by selecting and applying a subsequent detector such as an entire object detector according to the hidden area as the object detecting device, it is possible to suppress erroneous detection of the hidden object and improve the recognition rate.

  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.

DESCRIPTION OF SYMBOLS 101 Image acquisition part 102 Feature extraction part 103 Feature-value memory | storage part 104 Pre-stage detector 105 Detection result memory | storage part 106 Hidden area setting part 107 Scan area | region setting part 108 Post-stage detector application determination part 109 Whole body detection model 110 Post-stage detector 111 Result integration Part

Claims (10)

  1. Pre-detection means for detecting a specific partial region of the object;
    Setting means for setting a hidden region based on the distribution of detection results of the preceding detection means;
    Subsequent detection means for detecting an area different from the specific partial area of the object so as not to include the hidden area set by the setting means;
    An object detection apparatus comprising: an integration unit that integrates the detection result of the upstream detection unit and the detection result of the downstream detection unit.
  2.   The object detection apparatus according to claim 1, wherein the downstream detection unit detects an area including at least a part of the specific partial area with respect to the same object.
  3.   The object detection apparatus according to claim 1, wherein the object is a human body, and the specific partial region is a partial region of the upper half of the human body.
  4.   The object detection apparatus according to claim 3, wherein the rear detection unit detects a region including the whole body of a human body.
  5.   The setting unit sets the hidden region based on the position and size of the specific partial region detected by the front detection unit and the size of the region detected by the front detection unit or the rear detection unit. The object detection apparatus according to any one of claims 1 to 4, wherein
  6.   The object detection apparatus according to claim 1, wherein the latter-stage detection unit detects using a plurality of different detection models or detectors.
  7.   The said setting means sets the said hidden area based on the position and size of the specific partial area detected by the said front | former stage detection means, and these detection models or detectors. The object detection apparatus described.
  8.   The object detection apparatus according to claim 6, wherein the latter-stage detection unit determines an application region corresponding to each of the plurality of detection models or detectors.
  9. A pre-detection step for detecting a specific partial region of the object;
    A setting step for setting a hidden region based on the distribution of detection results of the preceding detection step;
    A subsequent detection step for detecting a region different from the specific partial region of the object so as not to include the hidden region set in the setting step;
    An object detection method comprising: an integration step of integrating the detection result of the upstream detection step and the detection result of the downstream detection step.
  10. A pre-detection step for detecting a specific partial region of the object;
    A setting step for setting a hidden region based on the distribution of detection results of the preceding detection step;
    A subsequent detection step for detecting a region different from the specific partial region of the object so as not to include the hidden region set in the setting step;
    A program for causing a computer to execute an integration step of integrating the detection result of the preceding detection step and the detection result of the subsequent detection step.
JP2013073936A 2013-03-29 2013-03-29 Object detection device, object method of detection, and program Pending JP2014199506A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2013073936A JP2014199506A (en) 2013-03-29 2013-03-29 Object detection device, object method of detection, and program

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP2013073936A JP2014199506A (en) 2013-03-29 2013-03-29 Object detection device, object method of detection, and program

Publications (1)

Publication Number Publication Date
JP2014199506A true JP2014199506A (en) 2014-10-23

Family

ID=52356377

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2013073936A Pending JP2014199506A (en) 2013-03-29 2013-03-29 Object detection device, object method of detection, and program

Country Status (1)

Country Link
JP (1) JP2014199506A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2016162105A (en) * 2015-02-27 2016-09-05 富士通株式会社 Image processing program, image processing apparatus, and image processing method
KR101834279B1 (en) * 2016-10-04 2018-03-06 네이버 주식회사 Image processing method and system for detecting face using lazy feature extract
US10372989B2 (en) 2015-10-30 2019-08-06 Canon Kabushiki Kaisha Control apparatus and control method for determining relation of persons included in an image, and storage medium storing a program therefor

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2016162105A (en) * 2015-02-27 2016-09-05 富士通株式会社 Image processing program, image processing apparatus, and image processing method
US10372989B2 (en) 2015-10-30 2019-08-06 Canon Kabushiki Kaisha Control apparatus and control method for determining relation of persons included in an image, and storage medium storing a program therefor
KR101834279B1 (en) * 2016-10-04 2018-03-06 네이버 주식회사 Image processing method and system for detecting face using lazy feature extract

Similar Documents

Publication Publication Date Title
Qi et al. Unsupervised ship detection based on saliency and S-HOG descriptor from optical satellite images
US9818023B2 (en) Enhanced face detection using depth information
US9208580B2 (en) Hand detection, location, and/or tracking
US8873837B2 (en) Image-based crack detection
Alexe et al. Measuring the objectness of image windows
Hou et al. People counting and human detection in a challenging situation
US10395103B2 (en) Object detection method, object detection apparatus, and program
Stalder et al. Cascaded confidence filtering for improved tracking-by-detection
US9367766B2 (en) Text line detection in images
US8406470B2 (en) Object detection in depth images
US8331619B2 (en) Image processing apparatus and image processing method
US9014467B2 (en) Image processing method and image processing device
JP5775225B2 (en) Text detection using multi-layer connected components with histograms
CN106874894B (en) Human body target detection method based on regional full convolution neural network
Li et al. Automatic pavement crack detection by multi-scale image fusion
JP5712774B2 (en) Object detection method and apparatus
US9025875B2 (en) People counting device, people counting method and people counting program
US10121095B2 (en) Method and device for recognizing subject area of image
JP6660313B2 (en) Detection of nuclear edges using image analysis
JP4410732B2 (en) Face image detection device, face image detection method, and face image detection program
US8923554B2 (en) Information processing device, recognition method thereof and non-transitory computer-readable storage medium
Hu et al. Clothing segmentation using foreground and background estimation based on the constrained Delaunay triangulation
KR20130043222A (en) Gesture recognition system for tv control
US8970696B2 (en) Hand and indicating-point positioning method and hand gesture determining method used in human-computer interaction system
CN105512683B (en) Object localization method and device based on convolutional neural networks