WO2005020152A1 - 人物検出装置および人物検出方法 - Google Patents
人物検出装置および人物検出方法 Download PDFInfo
- Publication number
- WO2005020152A1 WO2005020152A1 PCT/JP2004/011790 JP2004011790W WO2005020152A1 WO 2005020152 A1 WO2005020152 A1 WO 2005020152A1 JP 2004011790 W JP2004011790 W JP 2004011790W WO 2005020152 A1 WO2005020152 A1 WO 2005020152A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- person
- fragment
- image
- spatiotemporal
- temporal
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
- G06V40/25—Recognition of walking or running movements, e.g. gait recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
Definitions
- the present invention relates to a person detection device and a person detection method.
- the present invention relates to a person detecting device and the like for detecting a walking person by image processing, and more particularly to a person detecting device and the like for detecting a position and a walking direction of a person.
- a plurality of slits are arranged in an image, and a moving object image is obtained by using a temporal pixel value change in the slit area.
- a method has also been proposed for forming and counting the moving direction of a moving object and counting (for example, page 7 and FIG. 4 of Patent No. 3183320). According to this technology, it is possible to stably detect a person and determine a moving direction by considering a temporal change of a person passing through the slit.
- the first prior art is limited to the case where the position of the ankle of the person is known and the person is walking left and right with respect to the image. Also, since it is necessary to detect the ankle position in advance, it is assumed that the initial detection of the person has already been performed. Therefore, there is a problem that it is difficult to detect a person walking in various directions in the image.
- the walking direction can be detected by arranging a plurality of slits for detecting a person in an image, but the designer has to arrange the slits in advance. Therefore, there is a problem that an area where a person can be detected in an image is limited. Disclosure of the invention
- An object of the present invention is to solve the above-described problem, and perform a person detection including a person's location and a walking direction without limiting a detection area in an image even when the person's walking direction is not constant. It is an object of the present invention to provide a human detection device and the like that can perform the above-mentioned operations.
- a person detection device is a device for detecting a person included in a moving image, wherein a frame image constituting a moving image of a person is taken along a time axis. From the spatio-temporal volume generation means for generating the arranged three-dimensional spatio-temporal image, and the generated three-dimensional spatio-temporal image, A spatio-temporal fragment extraction means for extracting a real-image spatio-temporal fragment which is an image of a cut plane or a cut piece when the three-dimensional spatio-temporal image is cut, and a person motion model defining the motion characteristics of the person.
- a spatiotemporal fragment output unit for generating and outputting a spatiotemporal fragment of a human body part motion model, which is a spatiotemporal fragment obtained by movement by a human motion model; and a spatiotemporal fragment extracted by the spatiotemporal fragment extraction unit.
- a spatiotemporal fragment matching unit that matches a human body part motion model spatiotemporal fragment output by the spatiotemporal fragment output unit; and a presence / absence of a person in the moving image based on a result of the comparison by the spatiotemporal fragment matching unit.
- the spatiotemporal fragment extraction means determines a fragment extraction line for cutting the frame image, and sets a plane obtained by arranging the determined fragment extraction lines along the time axis as a cut plane.
- the real image spatio-temporal fragment is extracted.
- the real image spatio-temporal fragment is extracted using a fragment extraction line that cuts a leg of a person in the three-dimensional spatio-temporal image.
- the spatiotemporal fragment output means is based on a human motion model that defines the walking characteristics of the human, and a human body part motion model spatiotemporal fragment obtained by a cut surface that cuts a leg of the human motion model during walking.
- the person detection is performed using the walking characteristic peculiar to the motion of the person, it is possible to prevent a moving object such as a car having no walking characteristic from being erroneously detected.
- the person motion model for example, it corresponds to two legs.
- a model is defined in which one end is represented by two connected segments, and each segment alternately rotates up to a maximum angle of 2 ⁇ L at a constant angular velocity ⁇ with the joint point as the center of rotation. be able to.
- the spatiotemporal fragment matching means when the real image spatiotemporal fragment is scanned in the temporal direction with a human body part motion model spatiotemporal fragment for one step output by the spatiotemporal fragment output means.
- the matching is performed by calculating the degree of coincidence between the images. This makes it possible to reliably detect a person by simple image matching by using the fact that the human walking is a cyclic movement.
- the human detection device further includes a movement direction calculation unit that calculates a movement existing in the three-dimensional space-time image from the three-dimensional space-time image generated by the space-time volume generation unit.
- the spatio-temporal fragment extraction means may determine the fragment extraction line according to the movement direction calculated by the movement direction calculation means.
- the moving direction calculating means extracts the moving object for each frame image forming the three-dimensional spatiotemporal image, and obtains a motion vector between the extracted moving objects between the frame images.
- the moving direction of the moving object may be calculated, or each of the frame images constituting the three-dimensional spatiotemporal image may be divided into small areas, and the motion vector between the frame images may be calculated for each small area.
- the moving direction of the moving object may be calculated.
- the fragment extraction line is determined following the moving direction of the person, so that the fragment extracting line that generates the most appropriate spatiotemporal fragment regardless of the moving direction or position of the person, for example, Regardless of the position of the person, a fragment extraction line that always cuts the leg of the person is automatically determined.
- the fragment extraction line is, for example, a straight line or a curve.
- the attribute output means may include the real image by the spatiotemporal fragment matching means.
- a parameter for specifying the cut surface or the cut piece and a parameter for specifying the human body motion model are obtained.
- the person detecting device further includes a display unit for displaying a person attribute including the position and the moving direction of the person output from the attribute output unit. May be provided.
- the person detecting device further includes a periodicity analysis unit that analyzes whether the real image spatiotemporal fragment extracted by the spatiotemporal fragment extraction unit is an image based on a periodic motion peculiar to walking of a person.
- the spatiotemporal fragment extracting means may change a fragment extraction line based on an analysis result by the periodicity analyzing means, and extract a real image spatiotemporal fragment again using the changed fragment extraction line.
- the periodicity analysis means generates time-series data of a correlation length by obtaining a self-correlation function with respect to one-dimensional data indicating an image at each time constituting the real image spatiotemporal fragment.
- the real image spatio-temporal fragment is analyzed to be an image based on a periodic motion peculiar to human walking, and the correlation length time-series data
- a graph showing the change of the autocorrelation function value with respect to the correlation length is obtained by calculating the autocorrelation function for the correlation length. If there is a peak in the graph, it is determined that the time series data of the correlation length has periodicity. May be.
- the parameters of the fragment extraction line are changed and determined so that the periodicity based on the walking characteristic peculiar to the motion of the human is detected in the spatiotemporal fragment of the real image. People can be reliably detected without relying on the information.
- the human detection device may further change the parameter for specifying the cut plane or the cut piece by the spatiotemporal fragment extraction means based on the result of the comparison by the spatiotemporal fragment collation means, and then return the real image Extracting at least one spatial fragment, and causing the spatiotemporal fragment output means to change at least one of the parameters for identifying the human body motion model and then again outputting the human body part motion model spatiotemporal fragment.
- a solution search means for searching for an optimal parameter for specifying the cut plane or the cut piece and an optimal parameter for specifying the human body motion model may be provided.
- the solution search means searches for an optimal parameter using, for example, a genetic algorithm.
- the spatiotemporal volume generation means generates the three-dimensional spatiotemporal image by superimposing at least one or more images obtained by binarizing the frame image after subtracting the background difference or the interframe difference, and then superimposing at least one image. You may. As a result, a three-dimensional spatiotemporal image is generated only from a moving object, so that the matching speed is increased and the matching accuracy is improved.
- the spatiotemporal fragment output means generates and outputs a spatiotemporal fragment of a human body part movement model corresponding to a human motion model selected from a plurality of different types of human motion models stored in advance.
- the fragment matching means when the result of the matching does not satisfy a certain criterion, causes the spatiotemporal fragment output means to generate and output a human body part motion model spatiotemporal fragment corresponding to a new human motion model. In this case, the matching may be repeated.
- the plurality of person motion models stored in advance by the spatiotemporal fragment output means the gender, age, the state of the road on which the person walks, and the walking At least one congestion at the location is different It may be made to be.
- the present invention can be realized not only as a person detection device, but also for comparing an image of a person included in a moving image with a previously stored image of a person.
- a collation camera having at least one function of pan, tilt, and zoom; and a pan, tilt, and zoom by the collation camera based on the position or moving direction of the person detected by the person detection device.
- the present invention is realized as a person matching device including a force camera control unit that controls at least one person, and a person matching unit that matches a person image captured by the matching camera with a previously stored person image. You can also.
- the present invention can be applied to a person monitoring device, a person authentication device, or the like that searches for a specific person or authenticates a person.
- the present invention is an apparatus for fitting a person motion model defining the motion characteristics of a person to the motion of the person on the image, wherein a frame image constituting the moving image of the person is taken along a time axis.
- a spatio-temporal volume generating means for generating an arranged three-dimensional spatio-temporal image, and an image on a cutting plane or a slice when the three-dimensional spatio-temporal image is cut from the generated three-dimensional spatio-temporal image.
- Spatio-temporal fragment extraction means for extracting a real image spatio-temporal fragment, spatio-temporal fragment output means for generating and outputting a human body part motion model spatio-temporal fragment which is a spatio-temporal fragment obtained by a motion based on the human motion model, Spatio-temporal fragment matching means for comparing the real-image spatio-temporal fragment extracted by the spatio-temporal fragment extraction means with the human body part motion model spatio-temporal fragment output by the spatio-temporal fragment output means; Based on the comparison result by the spatiotemporal fragment verification unit, so that the person motion model indicating the motion of a person in the moving image, the model Fi Tsu Te to determine the value of the parameter that identifies the pre-Symbol human movement model It can also be realized as a person model fitting device provided with a singing means.
- the present invention is an apparatus for generating an image or the like used for detecting a person included in a moving image, the apparatus being configured to generate a three-dimensional spatiotemporal image in which frame images constituting the moving image are arranged along a time axis.
- This is a spatio-temporal fragment obtained by a motion based on a person motion model based on a person motion model that defines the motion characteristics of a person when an image on a cut plane or a cut piece at the time of cutting is defined as a spatio-temporal fragment.
- the present invention can also be realized as an image generation device including a spatiotemporal fragment output unit that generates and outputs a spatiotemporal fragment of a human body part motion model.
- a spatiotemporal fragment representing the motion of a characteristic human body part is generated from the human motion model, so that, for example, reference data corresponding to various human models to be collated in the human detection device is generated. It can be used as a dedicated device.
- the present invention can be realized not only as such a person detecting device, a person matching device, a person model fitting device, and an image generating device, but also as a person detecting method, a person matching method, and a person model fitting device.
- the present invention can also be realized as a programming method and an image generation method, as a program for causing a computer to execute such a method, or as a computer-readable recording medium on which the program is recorded.
- FIG. 1 shows a configuration of a person detection device according to the first embodiment of the present invention.
- FIG. 3 is a functional block diagram.
- FIG. 2 is a flowchart showing the operation of the person detecting device according to the first embodiment of the present invention.
- FIG. 3 is a diagram illustrating extraction of a spatiotemporal fragment according to the first embodiment of the present invention.
- FIG. 4 is a functional block diagram showing the configuration of the spatiotemporal fragment extraction unit according to the first embodiment of the present invention.
- FIG. 5 is a diagram showing fragment extraction lines on world coordinates according to the first embodiment of the present invention.
- FIG. 6 is a functional block diagram showing a configuration of a human body part movement spatiotemporal fragment output unit according to the first embodiment of the present invention.
- FIG. 7 is a diagram illustrating a person motion model according to the first embodiment of the present invention.
- FIG. 8 is a functional block diagram showing the configuration of the spatiotemporal fragment matching unit according to the first embodiment of the present invention.
- FIG. 9 (a) is a diagram showing a spatiotemporal fragment according to the first embodiment of the present invention
- FIG. 9 (b) is a human body part motion model spatiotemporal fragment according to the first embodiment of the present invention
- FIG. 9 (c) is a diagram showing matching in the first embodiment of the present invention.
- FIG. 10 is a functional block diagram showing the configuration of the person detecting device according to the second embodiment of the present invention.
- FIG. 11 is a functional block diagram illustrating a configuration of a solution search unit according to the second embodiment of the present invention.
- FIG. 12 is a functional block diagram showing the configuration of the person detecting device according to the third embodiment of the present invention.
- FIG. 13 is a diagram showing a display on a display unit according to the third embodiment of the present invention. It is.
- FIG. 14 is a functional block diagram showing the configuration of the person detecting device according to the fourth embodiment of the present invention.
- FIG. 15 is a diagram illustrating an example of a moving direction calculation method according to the fourth embodiment of the present invention.
- FIG. 16 is a diagram illustrating an example of a moving direction calculation method according to the fourth embodiment of the present invention.
- FIG. 17 is a functional block diagram showing a configuration of a spatiotemporal fragment extraction unit according to the fourth embodiment of the present invention. .
- FIG. 18 is a functional block diagram showing the configuration of the person detecting device according to the fifth embodiment of the present invention.
- FIG. 19 is a functional block diagram showing the configuration of the periodicity analysis unit according to the fifth embodiment of the present invention.
- FIG. 20 is a diagram illustrating an example of autocorrelation function calculation according to the fifth embodiment of the present invention.
- FIG. 21 is a diagram illustrating an example of calculation of the autocorrelation function of the correlation length according to the fifth embodiment of the present invention.
- FIG. 22 is a functional block diagram showing the configuration of the person model fitting device according to the sixth embodiment of the present invention.
- FIG. 23 is a functional block diagram showing a configuration of a person model fitting unit according to the sixth embodiment of the present invention.
- FIG. 24 is a diagram illustrating an example of a person motion model template for each type.
- FIG. 25 is a diagram illustrating an example of a plurality of human motion model templates that are different for each situation of a walking road surface.
- FIG. 4 is a diagram illustrating an example of a deltemplate. BEST MODE FOR CARRYING OUT THE INVENTION
- FIG. 1 is a functional block diagram showing the configuration of the person detecting device according to the present embodiment.
- This person detection device is a device that detects a person present in a video taken of a street, a parking lot, a store, or the like.
- Camera 10, video processing unit 11, continuous image processing unit 12, spatiotemporal It comprises a volume generation unit 13, a spatiotemporal fragment extraction unit 14, a human body part motion model spatiotemporal fragment output unit 15, a spatiotemporal fragment collation unit 16, an attribute output unit 17, and a display unit 18.
- a spatiotemporal fragment is extracted as an image will be described.
- the spatiotemporal fragment does not necessarily need to be an image, and may be obtained by fitting feature amounts extracted from the image, ellipses, straight lines, and the like. It may be an extracted parameter.
- a camera 10 is an imaging device that captures a video and outputs the captured video to a video processing unit 11 for each frame.
- the video processing unit 11 is a processing unit that performs an inter-frame difference process or a background difference process for each frame of the input video, and binarizes the video based on a threshold.
- the continuous image processing unit 12 accumulates the input image in chronological order of the frame.
- the input image This is a processing unit that performs feature extraction and fits ellipses, straight lines, etc. using, and accumulates the results as parameters in chronological order.
- the spatiotemporal volume generation unit 13 is a processing unit that generates a spatiotemporal volume using images or parameters for a predetermined number of N frames.
- the spatiotemporal fragment extraction unit 14 is a processing unit that extracts spatiotemporal fragments using the spatiotemporal volume generated by the spatiotemporal volume generation unit 13.
- the spatiotemporal fragment is an image on a cut plane or a slice when the three-dimensional spatial image indicated by the spatiotemporal volume is cut along the time axis.
- this is an image on a cut plane when a 3D spatial image is cut along a plane parallel to the horizontal axis of the frame and parallel to the time axis.
- the human body part motion model spatiotemporal fragment output unit 15 is a processing unit that outputs a human body part motion model spatiotemporal fragment according to the human motion model.
- the spatiotemporal fragment collating unit 16 includes a spatiotemporal fragment from the real image generated by the spatiotemporal fragment extraction unit 14 and a human body part motion model. This is a processing unit that performs matching with the motion model spatiotemporal fragment.
- the attribute output unit 17 uses the result of the spatiotemporal fragment matching unit 16 to calculate the position, walking direction, angular velocity, etc. of the detected person, or predict the position of the person after t seconds. It is a processing unit.
- the display unit 18 is a processing unit, a display, and the like for arranging and displaying the location and the walking direction of the person detected by the spatiotemporal fragment collation unit 16 in an overhead view.
- the camera 10 captures an image of an object such as a moving object.
- the camera 10 may be constituted by a plurality of cameras.
- the video processing unit 11 receives the input from the camera 10.
- the video is subjected to inter-frame difference processing or background difference processing for each frame, and the video is binarized using a threshold.
- background subtraction processing an image serving as a background where no person exists is prepared in advance.
- the continuous image processing unit 12 stores the binarized image as a buffer in the built-in hard disk or the like as a buffer in order to generate a spatiotemporal volume.
- an image is transmitted to the spatiotemporal volume generation unit 13 for each frame, or when a spatiotemporal volume is generated from a source other than the image, feature extraction, ellipse, straight line fitting, etc. are performed using the input image. The result is output as a parameter.
- the spatio-temporal volume generation unit 13 uses the predetermined number of frames of N images or the time N parameters to generate the spatio-temporal volume. Generate Here, if the N + 1st image is newly transmitted from the continuous image processing unit 12, the first frame of the spatiotemporal volume is discarded in S 206, so that N frames are always Generate a spatiotemporal volume holding images of minutes.
- the spatiotemporal fragment extraction unit 14 determines the parameters of the fragment extraction line, and the spatiotemporal volume generation unit 13 uses the determined fragment extraction line to generate the parameter. Extract spatiotemporal fragments from spatiotemporal volumes.
- FIG. 3A shows the spatiotemporal volume 21
- FIG. 3B shows the spatiotemporal fragment 22 determined by the fragment extraction line 23.
- the spatiotemporal fragment 22 shown in Fig. 3 (b) is a one-dimensional image of the fragment extraction line 23 in each frame constituting the spatiotemporal volume 21 arranged on the time axis. This corresponds to the image on the cut plane when the three-dimensional spatiotemporal image shown in 21 is cut along the time axis.
- the fragment extraction line 23 for extracting the spatiotemporal fragment may be a curve as well as a straight line as shown in FIG. 3 (a). Furthermore, the line may have a certain thickness (thickness). The method for determining the fragment extraction line will be described later.
- the human body part motion model spatiotemporal fragment output unit 15 determines the parameters of the human motion model based on the parameters of the fragment extraction line determined in S207. Then, a human body part motion model spatiotemporal fragment is generated from the human motion model generated based on the determined parameters of the human motion model. The method for determining the parameters of the human motion model and the method for generating the human body part motion model spatiotemporal fragment will be described later.
- the spatiotemporal fragment matching unit 16 calculates the spatiotemporal fragment extracted by the spatiotemporal fragment extraction unit 14 and the human body part motion generated by the human body part motion model spatiotemporal fragment output unit 15. Perform matching with model spatiotemporal fragments.
- the spatiotemporal fragment collating unit 16 collates the spatiotemporal fragment with the spatiotemporal fragment of the human body part motion model, and if the collation result does not match (NO in S210), If the process has proceeded to S211 and all the parameters of the human motion model have not been tried (NO in S211), the process proceeds to S208 and the human body part motion model spatiotemporal fragment output unit When 15 generates a new spatiotemporal fragment of the human body part motion model, and if all the parameters of the human motion model have been tried (YES in S211), the process proceeds to S207 and again.
- the spatiotemporal fragment extraction unit 14 determines a fragment extraction line 23.
- the spatiotemporal fragment and the human body part motion model spatiotemporal fragment match or are equal to or larger than the threshold in the matching result (YES in S210)
- the spatiotemporal fragment The collating unit 16 calculates the fragment extraction line at that time.
- the parameters and the parameters of the person motion model are output to the attribute output unit 17.
- the attribute output unit 17 calculates the presence position and the walking direction of the person, and outputs the calculated position to the display unit 18.
- the display unit 18 displays the presence or absence of the person, the presence position, and the walking direction on the screen.
- the spatiotemporal fragment extraction unit 14 performs fragment extraction that cuts the spatiotemporal volume in a field coordinate system, which is a coordinate axis that represents the position and movement direction of a person in real space in real space.
- a fragment extraction line generation unit 30 that determines the line 23
- a coordinate conversion unit 31 that converts from the world coordinate system to a pixel coordinate system that expresses the image plane using parameters related to the installation of the camera 10, and a spatiotemporal volume
- a spatiotemporal volume cutting unit 32 for extracting spatiotemporal fragments 22 from 21 is provided.
- the fragment extraction line generation unit 30 defines a straight line and a curve on the world coordinates.
- the fragment extraction line generation unit 30 uses a world coordinate system that expresses coordinate axes in the real space in order to generate straight lines and curves based on the position and the moving direction of the person in the real space.
- the world coordinates are indicated by (Xw, YwZw). The details of world coordinates are described in Xu and Tsuji, "3D Vision", page 9, Kyoritsu Shuppan, published in 1998.
- ⁇ w is a parameter related to the walking direction of a person existing on the world coordinates representing the real space. If the intercept b w can be obtained, the walking of the person on the world coordinates is on the straight line represented by the equation (1).
- a curve is also possible to use a curve as the fragment extraction line 23. For example, it is also effective to use a curve according to the fluctuation of walking. In this case, by assuming a walking cycle, a fragment extraction line is determined by a sin curve or the like, and fragment extraction is performed while shifting the phase, thereby performing fragment extraction that matches the periodic vertical movement in walking. it can.
- the above 0 W and bw are determined in a sequential manner based on the result of the later-described comparison performed by the spatiotemporal fragment collation unit 16, assuming a combination that covers the inside of the monitoring area.
- the above two parameters may be determined based on the detection result, and are not necessarily exhaustive.
- the coordinate transformation unit 3 1 a fragment extraction line 2 3 produced by the parameter of the 0 W and b w, the installation position of the camera 1 0, focal length, using pre-known parameters such as scale factor Transforms from the world coordinate system to the pixel coordinate system representing the image plane.
- the fragment extraction line becomes a line on the two-dimensional image.
- the spatiotemporal volume cutting unit 32 extracts spatiotemporal fragments.
- This spatio-temporal fragment was extracted by the spatio-temporal volume generation unit 13 using the fragment extraction line in the pixel coordinate system generated by the coordinate transformation unit 31. This is done by cutting the spatiotemporal volume.
- the spatiotemporal fragment matching unit 16 compares the spatiotemporal fragment extracted by the spatiotemporal volume cutting unit 32 with the human body part motion model spatiotemporal fragment output from the human body part motion model spatiotemporal fragment output unit 15. Then, a fragment extraction line parameter change signal, which is a signal indicating a parameter change of the fragment extraction line, is output to the fragment extraction line generation unit 30 based on the comparison result.
- the spatiotemporal fragment extraction unit 14 creates a fragment extraction line and creates a spatiotemporal fragment until the input of the fragment extraction line parameter change signal is completed.
- the human body part motion model spatiotemporal fragment output unit 15 includes a human motion model generation unit 50 that models the walking of a person on world coordinates using fragment extraction lines 23, and world coordinates.
- a coordinate transformation unit 51 for converting a system into a pixel coordinate system, and a spatiotemporal fragment output unit 52 for generating a spatiotemporal fragment according to a human motion model are provided.
- a person's walking is modeled using two line segments representing legs.
- a more detailed human motion model may be used in consideration of the knee joint, ankle, and the like.
- the motion of the person may be modeled using an ellipse or the like, instead of the line segment.
- human motion model generating unit 5 0 a one step walking in the world coordinate system as the basic unit, as shown in FIG. 7, the position on the Waal de coordinates human movement model walks (x start , Y start) and angular velocity ⁇ .
- start% y start is one point on the fragment extraction line 23 on the world coordinates determined by the spatiotemporal fragment extraction unit 14.
- FIG. 7 shows a case where the legs are opened to the maximum in the human motion model.
- the next step starts from (X 2nd, V 2nd) and moves to angular velocity and angle 2.
- the parameters 0 w and b w of the fragment extraction line determined by the fragment extraction line generator 30 represent a straight line on world coordinates.
- the human motion model is a model of the motion of a human leg. Two straight lines are used as legs, and the position (x start , y start) in world coordinates is determined. By moving the legs of the human motion model at the angular velocity ⁇ as the starting point, a one-step walk is generated. If the crotch angle of the human motion model is 20 L, then 20 L Z o) is the time or the number of frames required to walk one step.
- ⁇ is a comprehensive trial of the possible angular velocity ⁇ in consideration of the walking speed of the person, so that the combination of the angular velocity ⁇ and the position ( Xstart , ystart) in world coordinates is used.
- the case where the length L of the leg and the angle e L between the crotch are determined by the designer in advance will be described.
- the walking generated by the human motion model is limited to one step. Instead, the basic unit can be several steps.
- the coordinate conversion unit 51 uses the known coordinates such as the installation position of the camera 10, the focal length, and the scale factor to generate the world coordinates generated by the human motion model.
- the spatiotemporal polygram is generated from the human motion model by transforming the human motion model for one step above into walking in the pixel coordinate system.
- the spatio-temporal fragment output unit 52 outputs the same 0 W and b w parameters as the spatio-temporal fragment extraction unit 14 to the spatio-temporal volume in the pixel coordinate system generated by the coordinate transformation unit 51. To generate a spatiotemporal fragment.
- a spatiotemporal volume is generated by the coordinate transformation unit 51 using the human motion model generated by the human motion model generation unit 50, and a spatiotemporal fragment of the human motion model is generated from the spatiotemporal volume.
- the human motion model is modeled as leg motion, but the human body part motion model spatiotemporal fragment is generated as a fragment that focuses on one leg, such as an ankle position. Become.
- a spatiotemporal fragment of the human body part motion model can be generated, and the amount of calculation can be reduced.
- the spatiotemporal fragment matching unit 16 compares the spatiotemporal fragment extracted by the spatiotemporal fragment extraction unit 14 with the human body part motion model spatiotemporal fragment output from the human body part motion model spatiotemporal fragment output unit 15. Then, a human body part motion model spatiotemporal fragment parameter change signal, which is a signal indicating a parameter change of the human body part motion model spatiotemporal fragment, is output to the human motion model generation unit 50 based on the collation result.
- the human body part motion model spatiotemporal fragment output unit 15 creates the human body part motion model spatiotemporal fragment from the human motion model until the input of the human body part motion model spatiotemporal fragment parameter change signal is completed. I do.
- the spatiotemporal fragment matching unit 16 is composed of the spatiotemporal fragment extracted by the spatiotemporal fragment extraction unit 14 and the human body part motion model spatiotemporal generated from the human motion model by the human body part motion model spatiotemporal fragment output unit 15. It comprises a matching processing unit 150 for performing matching with a fragment, and a comparing unit 152 for holding a matching result, comparing the result with a predetermined threshold value, and outputting a matching result or a parameter change request.
- the matching processing unit 150 matches the spatiotemporal fragment extracted from the real image by performing matching while scanning the spatiotemporal fragment of the human body part motion model generated from the human motion model in the time direction.
- a collation example for a binarized image will be described.
- Fig. 9 (a) is a spatiotemporal fragment 70 focusing on the motion of the human leg extracted from the real image
- Fig. 9 (b) is a human body part motion model spatiotemporal fragment 7 1 generated from the human motion model.
- Figure 9 (c) shows the matching score, which is the matching score of both.
- the human body part motion model spatiotemporal fragment 71 generated from the human motion model is scanned from top to bottom for each pixel, and the matching score is calculated. .
- a step of calculating a matching score is referred to as a step.
- the matching score is calculated based on the spatio-temporal fragment 70 and the human body part motion model. If the Dell spatiotemporal fragment 1 is binarized to "0" and “1", the pixel whose pixel value is “1” is set to the ON pixel, and the pixel value is "0" Is the OFF pixel, and the ON and OFF pixels of the spatiotemporal fragment 71 of the human body motion model are compared with the ON and OFF pixels of the spatiotemporal fragment 70.
- a human body part motion model spatiotemporal fragment 7 1 is superimposed on an arbitrary position of the spatiotemporal fragment 70.
- the value obtained by normalizing the number of ON pixels to be compared with the number of ON pixels of the human body part motion model spatio-temporal fragment 71 and the number of OFF pixels to be collated are normalized by the number of OFF pixels of the human body part motion model temporal space fragment 71. By adding the converted values, the matching score is obtained.
- the matching score is calculated while scanning the human body part motion model spatiotemporal fragment 71 for each step, and is output to the comparing section 151.
- the comparing unit 151 compares the score indicating the largest matching score in the scanning process with a predetermined threshold value, and matches the matching score exceeding the threshold value, the number of steps thereof, and the parameters of the human motion model. The result is output to the attribute output unit 17.
- the largest matching Only the score indicating the core is compared with the threshold value, but may be compared with the threshold value at each step.
- the spatiotemporal fragment matching unit 16 changes the parameters of the human body part motion model spatiotemporal fragment.
- the required body part motion model spatiotemporal fragment parameter change signal is output to the human body part motion model spatiotemporal fragment output unit 15. If all the human motion model parameters have been examined, fragment extraction is performed.
- a fragment extraction line parameter change signal requesting a change in line parameters is output to the spatiotemporal fragment extraction unit 14.
- the spatiotemporal fragment parameter change signal and the fragment extraction line parameter change signal of the human body part motion model include 0 W and b w which are the parameters of the fragment extraction line 23 and the parameters of the human motion model (X st art yst art / ⁇ and ⁇ are included.
- the time in FIG. 9 (a) is the number N of frames determined in advance by the spatiotemporal volume generation unit 13 and the time in FIG. 9 (b) is generated by the human motion model generation unit 50. If the crotch angle of the obtained human motion model is 2 ⁇ L and the angular velocity is ⁇ , it is 20 ⁇ , and the steps in FIG. 9 (c) are the number of steps of the scan processing performed by the matching processing unit 150. .
- the parameters of the human motion model of the human body part motion model spatiotemporal fragment output unit 15 are changed under the constraint of the fragment extraction line parameters, and the combination of the parameters relating to the human motion model is tried.
- the detection result with the highest accuracy all combinations of parameters covering the monitoring area are tried, but if the detection result with sub-optimal accuracy is acceptable, detection is performed using the threshold value in the matching processing unit 150.
- the result can also be.
- the human candidate area can be calculated using other initial detection methods Does not necessarily try all parameter combinations that cover the monitored area.
- the parameter of the fragment extraction line of the spatiotemporal fragment extraction unit 14 is changed again, and the combination of the parameters related to the human motion model is repeatedly tried again under the constraint of the fragment extraction line.
- the matching processing unit 150 It is also possible to use the threshold value as a detection result.
- a person candidate area can be calculated using another initial detection method, it is not always necessary to try combinations of all parameters covering the monitoring area.
- the human body part movement model spatiotemporal fragment is also a fragment having a similar thickness.
- the spatiotemporal fragment collation unit may collate fragments having a large thickness, or a spatiotemporal fragment extracted from the spatiotemporal volume and a human body part motion model spatiotemporal fragment generated from the human motion model. May be compared to each other by compressing the image into a single image.
- the attribute output unit 17 uses the combination of the parameters output from the spatiotemporal fragment matching unit 16 to perform the same walking after t seconds based on the detected position, walking direction, and angular velocity of the person. Direction and constant angular velocity as in detection Calculate the position of the person t seconds later, assuming that the person is moving.
- the fragment extraction line parameter 0 w corresponds to the walking direction on the world coordinate
- the parameters of the human motion model ( X3art , ystart) Power ⁇ corresponds to the position of the person.
- the detection time is calculated by the number of steps of the scan processing performed by the matching processing unit 150 attached to the combination of parameters, and specifically, can be calculated from the number of steps when the maximum matching score is shown.
- the stride length can be calculated, and after detecting from the stride length, the angular velocity and the walking direction of the human motion model, t It predicts the location after a lapse of seconds. Note that this predicted value is the position of the person on world coordinates.
- the display unit 18 arranges and displays the position and the walking direction of the person detected by the spatiotemporal fragment matching unit 16 on the overhead view.
- FIG. 10 is a functional block diagram showing the configuration of the person detecting device according to the present embodiment.
- This person detection device is a device that detects a person present in a video taken of a street, a parking lot, a store, or the like, as in the first embodiment. However, compared to the first embodiment, it has a feature that it is possible to match spatiotemporal fragments at a higher speed.
- spatio-temporal volume generation unit 13 spatio-temporal fragment extraction unit 14
- spontaneous body motion model spatio-temporal fragment output unit 15, spatio-temporal fragment collation unit 16, attribute output unit 17, display unit 18
- a solution search unit 110 This configuration corresponds to a configuration in which a solution search unit 110 is added to the configuration of the person detection device according to the first embodiment.
- the points different from the first embodiment will be mainly described.
- matching of spatiotemporal fragments is performed by performing a full search for the parameters of the fragment extraction line and the parameters of the human motion model, or by searching for a parameter whose matching result is equal to or greater than a threshold value.
- the provision of the solution search unit 110 for determining the above parameters enables high-speed spatio-temporal fragment matching.
- the camera 10 the video processing unit 11, the continuous image processing unit 12, the spatiotemporal movie generator 13, the spatiotemporal fragment extraction unit 14, the human body part motion model spatiotemporal fragment output unit 15,
- the operation of the spatiotemporal fragment collating unit 16 is the same as that of the first embodiment, and a description thereof will be omitted.
- the spatiotemporal fragment matching unit 16 performs matching between the spatiotemporal fragment 70 and the human body part motion model spatiotemporal fragment file 1 as in the first embodiment, and after matching, a matching score, The number of steps, the parameters of the fragment extraction line, and the parameters of the human body part motion model spatiotemporal fragment are output to the solution search unit 110. Note that the matching method is the same as in the first embodiment.
- the solution search unit 110 uses the genetic algorithm described in Kitano, “Genetic Algorithms”, pp. 1-41, Sangyo Tosho, published in 1993, to create a suboptimal solution. explore the parameter set, a parameter of the fragment extraction line 2 3 0 W and b w fragment extraction line generating section 3 0, person movement model By outputting the parameters (X start, V start) and ⁇ , high-speed human detection is realized.
- FIG. 11 is a functional block diagram showing a detailed configuration of the solution search unit 110.
- the solution search unit 110 includes a binarization unit 121 that converts a parameter such as a matching score input from the spatiotemporal fragment matching unit 16 into a bit sequence, and a plurality of parameters converted into a bit sequence.
- a gene selector 122 selects a bit string to perform genetic operations, and a set of selected bit strings is used to perform mutation, crossover, etc.
- a genetic operation unit 123 for generating a new bit string by performing a genetic method and a real number conversion unit 124 for converting the generated new bit string into a real value again are provided.
- the binarization unit 1 2 1 is the parameters of the segment extraction line 23 0 W and b w input from the spatiotemporal fragment matching unit 16, and the parameters of the human motion model (start% V start) and ⁇ are converted into bit strings, respectively, and joined to generate one bit string.
- the conversion to a bit string is a conversion from a 10-ary system to a binary system.
- the gene selection unit 122 can determine the initial value of the bit string by randomly selecting the initial value of the parameter input from the spatiotemporal fragment matching unit 16 The bit strings are sorted in order with the higher matching score.
- the genetic operation unit 123 considers a bit string obtained by connecting the parameters as a gene, and selects a bit string having a higher matching score value as a parent with a higher probability.
- a new parameter bit sequence is obtained.
- the crossover is performed, for example, by determining a crossing bit position, called a crossing point, of the selected two bit strings forming a pair with a random number, and before and after the crossing point as a boundary.
- Mutation is to create a new bit sequence by alternately replacing the bits.Mutation, for example, randomly determines the position of the bit to be mutated with a certain probability and inverts the bit sequence.
- bit string is changed.
- bit string representing a is a parameter of a person movement model (X start, y start) uses a constraint that the straight line of which is a parameter of fragment extraction line 2 3 e w and b w You.
- the output result of the genetic operation unit 123 is output to the gene selection unit 122, and the genetic operation is repeated, so that a solution can be efficiently searched.
- the real number conversion unit 124 converts the bit string newly created by the genetic operation unit 123 into real-valued parameters, This is output as a fragment extraction line parameter change signal including parameters, and is output to the human body part motion model spatiotemporal fragment output unit 15 as a human body part motion model spatiotemporal fragment parameter change signal including various parameters.
- the spatiotemporal fragment extraction unit 14 determines the fragment extraction line 23 based on the fragment extraction line parameter change signal including the parameters input from the solution search unit 110, and extracts the spatiotemporal fragment. Then, the human body part motion model spatiotemporal fragment output unit 15 generates a human motion model based on the human body part motion model spatiotemporal fragment parameter change signal including the parameter input from the solution search unit 110, The human motion model spatiotemporal fragments are generated and output to the spatiotemporal fragment collating unit 16, and the spatiotemporal fragment collating unit 16 compares them.
- the parameters of the fragment extraction line 23 and the parameters of the human motion model are searched for by the genetic algorithm, so that the spatiotemporal fragment Faster human detection is possible than when matching is performed.
- FIG. 12 is a functional block diagram showing a configuration of the person verification device according to the present embodiment.
- This person verification device is a device for verifying a person by controlling a camera using the person detection device according to the first embodiment.
- the display unit 18 includes a display unit 18, a camera control unit 100, a person collation unit 103, and a collation camera 9441 "! To n.
- This configuration is included in the person detection device according to the first embodiment.
- the camera control unit 100, the person collation unit 103, and the collation camera 944-1 "! To n are added.
- the points different from the first embodiment will be mainly described.
- the camera control unit 100 uses the result of the attribute output unit 17 to track the corresponding person.
- the collation cameras 944-1 "! To n are pan, tilt, and zoom. It has a function, and the position on the set world coordinates and movable The range and the photographable range are stored.
- the person collation unit 103 performs collation of persons using the images captured by the collation cameras 944-1 "! -N.
- the display unit 18 is a processing unit that displays an overhead view in which the detected persons are arranged, a display, and the like.
- the camera control unit 1000 calculates the matching power cameras 941 to n that are closest to the positions on the world coordinates at which the person 9 1 to 1 to n can be photographed, and performs pan, tilt, and zoom. This is a control unit that outputs the control signal to be performed to the collation camera 941-1-n.
- the collation cameras 941-1 to n are imaging devices that perform panning, tilting, and zooming based on the control signal and photograph the corresponding person.
- the person matching unit 103 is a processing unit that compares a photographed person with a previously stored video to perform person matching.
- FIG. 13 shows a bird's-eye view 90 shown on the display unit 18 and a state of person verification.
- the bird's-eye view 90 is an area in which a person can be monitored, and the persons 911 "! To n indicate the position and walking direction of the person.
- the position of the detected person is displayed by creating in advance the correspondence with the position on the world coordinates representing the space.
- the person to be compared may be selected by the force camera control unit 100 sequentially or may be arbitrarily selected by the observer.
- the attribute output unit 17 outputs the predicted person position and the person information in the moving direction detected by the spatiotemporal fragment matching unit 16 to the display unit 18 so that the person 91-1 is displayed on the monitor.
- person information indicating the position and moving direction of the person 91-1 is output to the camera control unit 100.
- the camera control unit 1000 selects the optimal matching power camera 9411 from the personal information output from the attribute output unit 17 and uses the personal information to determine the optimal matching power.
- the camera control signal which determines the control amount of the zoom, tilt, and zoom, is output to the verification camera 941-1.
- the collation camera 9411 captures the person to be tracked by operating according to the camera control signal, and highlights the display of the person 911 in the bird's-eye view 90 of the display unit 18. .
- the matching camera 9411 captures the person 911 based on the camera control signal, and outputs the captured image to the person matching unit 103.
- the person matching unit 103 compares the captured image 95 with a previously stored image to be compared 96, and performs matching. . If the matching result 97 matches, the matching process ends, and the matching result 97 does not match. If the matching result 97 does not match, the attribute output unit 17 determines that the next matching person 9 1 — 2 is selected and the person information is output to the camera control unit 100.Then, the same processing is performed until the matching result 97 becomes negative or until all the persons to be matched are selected. repeat.
- the number of collation cameras 94 selected from the camera control unit 100 may be plural, and the images used for collation are captured by the plurality of collation cameras 94. By selecting from images, the matching accuracy can be improved.
- the force camera control unit 100 uses the detected walking direction to determine a force camera capable of photographing a person from the position of the person, and By controlling the camera in the direction most directly opposite to the detected walking direction in consideration of the movable range of the camera, it is possible to capture a frontal face image. This makes it possible to provide a more detailed face image, and also to improve the reliability of face matching when performing automatic face matching.
- the position of a person and the direction of walking are detected, and the camera is controlled based on the detection result.
- An image of an object can be taken.
- FIG. 14 is a functional block diagram showing the configuration of the person detecting device according to the present embodiment.
- This person detection device is a device for detecting a person present in a video taken of a street, a parking lot, a store, or the like as in the first to third embodiments, but calculates a moving direction of a moving object. It is characterized in that the fragment extraction line is determined according to the calculated moving direction, and the camera 10, video processing unit 11, continuous image processing unit 12, spatiotemporal volume generation unit 13, spatiotemporal It has a fragment extraction unit 14, a human body part motion model spatiotemporal fragment output unit 15, a spatiotemporal fragment collation unit 16, an attribute output unit 17, a display unit 18, and a movement direction calculation unit 140.
- This configuration corresponds to a configuration in which the moving direction calculation unit 140 is added to the configuration of the person detection device according to the first embodiment. The following description focuses on the differences from the first embodiment.
- spatiotemporal fragment matching is performed by performing a full search for the parameters of the fragment extraction line and the parameters of the human motion model, or by searching for a parameter whose matching result is equal to or greater than a threshold value.
- the moving direction calculation unit 140 for calculating the moving direction of the moving object, high-speed spatio-temporal fragment matching becomes possible, and more accurate person detection becomes possible.
- the operation of the fragment matching unit 16 is the same as that of the first embodiment, and a description thereof will be omitted.
- the movement direction calculation unit 140 that calculates the movement direction of an object on an image from the spatiotemporal volume generated by the spatiotemporal volume generation unit 13 will be described.
- the method of calculating the moving direction includes a method of calculating the moving direction by detecting a candidate for the target object, and a method of calculating the moving direction without detecting the target object.
- the isolated object is regarded as a candidate for one detection target object by using a labeling algorithm for extracting the isolated object from one difference image.
- Object extraction by the labeling algorithm is performed by connecting pixels with a pixel value of 1 and attaching different labels to different connected parts.
- the procedure was as follows.
- a pixel P having a pixel value of 1 and an unlabeled pixel P is detected and labeled.
- the same label L is assigned to all pixels connected to pixel P.
- Object extraction is performed by continuing this operation until there are no unlabeled pixels.
- other labeling methods may be used as long as isolated objects can be extracted.
- the small region is perturbed to obtain the difference at time t + 1.
- Matching with the binarized difference image is performed.
- the motion vector up to the position with the maximum matching value is calculated.
- the motion vector is calculated in each small area. This motion vector calculation processing is performed at a fixed time, and the average motion vector for each small area is obtained. Then, the average motion vector calculated for each small area is voted, and if a vote equal to or more than a certain value is obtained, the movement direction of the target object candidate is determined.
- the moving direction calculating method as described above has been described.
- another object detecting method may be used as long as the moving vector can be calculated.
- the spatiotemporal fragment extraction unit 14 includes a fragment extraction line generation unit 171 that generates fragment extraction lines and a spatiotemporal volume cutting unit 172 that extracts spatiotemporal fragments. And a coordinate conversion unit 173 for converting a straight line and a curve parameter on the image into a straight line and a curve on the world coordinates.
- the fragment extraction line generation unit 171 will be described. First, define a straight line and a curve on the image. Here, a case will be described in which a fragment extraction line 23 is obtained by drawing a straight line on an image.
- the fragment extraction line 23 can be defined by Equation 2 below.
- Y i a X i + b-(Equation 2)
- the inclination a is a parameter relating to the moving direction of the moving object on the image
- b is the intercept on the image.
- the inclination a can be obtained from the moving direction of the target object candidate calculated by the moving direction calculation unit 140.
- a fragment extraction line 23 is generated, and the spatiotemporal volume cutting unit 172 can extract the spatiotemporal fragment 22.
- the extracted spatiotemporal fragment 22 is sent to the spatiotemporal fragment collating unit 16.
- the coordinate conversion unit 173 converts the straight line on the image generated by the fragment extraction line generation unit 171 into a straight line in world coordinates. Since the human body part motion model spatiotemporal fragment output unit 15 generates a human motion model according to the straight line in the world coordinate system converted here, the following operation is the same as that of the first embodiment.
- the spatiotemporal fragment matching unit 16 compares the spatiotemporal fragment 70 with the human body part motion model spatiotemporal fragment 71, and extracts a fragment extraction line from the matching result. Then, a fragment extraction line parameter change signal, which is a signal indicating the parameter change, is output to the fragment extraction line generation unit 17 1.
- the spatiotemporal fragment extraction unit 14 creates spatiotemporal fragments according to the parameters from the spatiotemporal data until the input of the fragment extraction line parameter change signal is completed.
- the fragment extraction line parameter change signal may be changed for all parameter candidates based on the moving direction of the moving object candidate calculated by the moving direction calculation unit, or the attribute output unit “! 7”. This may be until the moving object is detected.
- the fragment extraction line parameter change signal is a and b which are the parameters of the fragment extraction line.
- the parameter of the fragment extraction line 23 and the parameter of the human motion model can be reduced by calculating the moving direction of the moving object. Compared with the case of performing spatial fragment matching, faster human detection is possible.
- FIG. 18 is a functional block diagram illustrating a configuration of the person detecting device according to the present embodiment.
- This person detection device is a device that detects a person present in a video shot of a street, a parking lot, a store, etc., as in the first to fourth embodiments, but calculates the moving direction of the moving object.
- the method has the characteristic that the fragment extraction line is determined while also verifying the periodic motion peculiar to walking, and it is characterized by the following: power camera 10, video processing unit 11, continuous image processing unit 12, Component 13; spatiotemporal fragment extraction unit 14; human body part motion model spatiotemporal fragment output unit 15; spatiotemporal fragment collation unit 16; attribute output unit 17; display unit 18; movement direction calculation unit 1 40 and a periodicity analysis unit 190 are provided.
- This configuration is obtained by adding a moving direction calculation unit 140 and a periodicity analysis unit 190 to the configuration of the person detection device according to the first embodiment, that is, the configuration according to the fourth embodiment. This corresponds to the addition of the periodicity analysis unit 190.
- the points different from the first and fourth embodiments will be mainly described.
- spatio-temporal fragment matching is performed by performing a full search for the parameters of the fragment extraction line and the parameters of the human motion model, or by searching for a parameter whose matching result is equal to or greater than a threshold value.
- a moving direction calculating unit 140 for calculating the moving direction of a moving object is provided, and a periodicity analyzing unit 190 for verifying a periodic motion peculiar to walking is provided.
- Camera 10 video processing unit 11, continuous image processing unit 12, spatiotemporal volume generation unit 13, human body part motion model spatiotemporal fragment output unit 15, spatiotemporal fragment collation unit 16, moving direction
- the operation of the calculation unit 140 is the same as that of the first and fourth embodiments, and therefore the description is omitted.
- the spatiotemporal fragment extraction unit 14 defines straight lines and curved lines on an image based on the moving direction of the moving object calculated by the moving direction calculation unit 140.
- a fragment extraction line 23 is obtained by drawing a straight line on an image.
- the fragment extraction line 23 can be defined by Equation 3 below.
- the inclination a is a parameter relating to the moving direction of the moving object on the image
- b is the intercept on the image.
- the inclination a can be obtained from the moving direction of the target object candidate calculated by the moving direction calculation unit 140.
- the fragment extraction line 23 is generated, and the spatiotemporal fragment 22 can be extracted.
- the extracted parameters of the spatiotemporal fragment 22 and the fragment extraction line 23 are sent to the periodicity analyzer 190.
- the periodicity analysis unit 190 calculates an autocorrelation function for each spatio-temporal fragment at each time t, and calculates a correlation length for each of the correlation length calculation units 191, Correlation length autocorrelation calculation that calculates the autocorrelation function again for the time series of correlation lengths in which the calculated correlation lengths are arranged in the time direction
- the peak position is detected from the autocorrelation function of the correlation length input from the correlation length autocorrelation calculation unit and the correlation length autocorrelation calculation unit, and it is determined whether the detected peak position matches the moving period of the moving object.
- a peak detector 193 that determines whether or not the object is a moving object by verifying it.
- a coordinate converter 1 that converts a straight line on an image when a peak is detected by the peak detector 193 into a straight line in world coordinates. 9 and 4.
- the periodicity analyzer 190 analyzes the periodicity of the spatiotemporal fragment 22, and when the periodicity analysis is completed, a fragment extraction line parameter change signal, which is a signal indicating a parameter change of the fragment extraction line 23, is sent. Output to spatiotemporal fragment extraction unit 14.
- the spatiotemporal fragment extraction unit 14 creates a fragment extraction line 23 and creates a spatiotemporal fragment 22 until the input of the fragment extraction line parameter change signal is completed.
- the correlation length calculation unit 19 1 divides the spatiotemporal fragment 201 shown in FIG. 20 (a) by one-dimensional at every time t as shown in the example shown in FIG. 20 (b).
- the data 202 is created, and the autocorrelation function 203 is calculated for each one-dimensional data 202.
- the time in FIG. 20 (a) is the number N of frames determined in advance by the spatiotemporal volume generation unit 13.
- the length of FIG. 20 (b) is the width X of the spatiotemporal fragment 201 shown in FIG. 20 (a).
- the calculation of the autocorrelation function 203 can be defined by Equation 4 below.
- f (X) is one-dimensional data 202
- C (r) is a self-correlation function 203.
- the autocorrelation function C ( ⁇ ) is When f (x) is shifted by an interval r (f (X + te)), it is a measure of how similar it is to the original one-dimensional data f (X).
- Figure 20 (c) shows the relationship as the autocorrelation function c (r).
- the autocorrelation function c (0) takes the maximum value because it shows the correlation with itself.
- the autocorrelation function C (rp) is p at the position where the autocorrelation function C (r) peaks, and the interval between ON pixels having high correlation in one-dimensional data corresponds to p.
- the position p where the peak exists in the autocorrelation function C (r) indicates the stride when focusing on the leg of the moving object.
- the temporal change of the autocorrelation function C (te) indicates the temporal change of the stride in movement, and can be expected to be periodic.
- Figures 20 (d), (e), and (f) are obtained when the characteristic of walking (periodicity) does not exist, for example, when the fragment extraction line is set at a position that crosses the human body.
- the spatio-temporal fragment data examples the spatio-temporal fragment, an example of one-dimensional data at a certain time of the spatio-temporal fragment, and a graph showing the relationship between the periodicity and the autocorrelation function C ( ⁇ "), respectively.
- L is referred to as a correlation length.
- the correlation length r L is calculated for each time, and the correlation length ⁇ L calculated for each time is arranged in chronological order to obtain time-series data 210 of the correlation length.
- Figure 21 (a) shows the time-series data 210 of the correlation length ITL.
- the time-series data 210 having a correlation length L is equivalent to a temporal change of a stride when an ideal spatiotemporal fragment is input, and periodically fluctuates with time.
- the correlation length calculation unit 1911 outputs the calculated time series data 210 of the correlation length L to the correlation length autocorrelation calculation unit 1992.
- the correlation length autocorrelation calculation section 1992 calculates an autocorrelation function 2 11 1 with respect to the time series data 210 of the correlation length I ′′ L as shown in FIG. 21 (a). Is the same as Equation 4.
- the calculation result is shown in Fig. 21 (b).
- the autocorrelation function 2 1 1 is obtained for the time series data 210 with a correlation length L.
- C t ( ⁇ ) which is the result of the calculation, is output to the peak detector 193.
- the fragment extraction line parameter change signal is b, which is the parameter of the fragment extraction line.
- the peak detector 193 detects the peak position w from the autocorrelation function C t (h) of the correlation length input from the correlation length autocorrelation calculator, and the detected peak position a w is the moving period of the moving object. It is determined whether or not the object is a moving object by verifying whether or not the moving object is matched. If it is determined that the object is a moving object, the position and the moving direction of the moving object are specified and output to the display unit 18.
- the correlation length of the autocorrelation function C t () is a fresh shifted by a certain interval of L (t) Te L (t + ⁇ ), 7 : measure of your L and (t) are similar much It becomes.
- the peak detection unit 1993 determines that the object is a moving object.
- the number of frames required for one step of movement depends on how many frames can be acquired per second by the video processing unit 11, but in the present embodiment, when 30 frames are input per second, Let 20 to 30 frames be the number of frames required for one step movement. If the peak position is between 20 and 30 frames, and the peak of the autocorrelation function C t () of the correlation length is detected, Judge as a moving object. This number of frames can be freely determined by the designer. When performing a more accurate moving object determination, it is also possible to detect a moving cycle in which two or more steps are defined as one unit. In that case, the number of steps determined by the designer can be increased by the number of steps.
- the periodicity can be similarly determined.
- the peak position a w of the peak detecting section 1993 may be calculated from the time required for the moving object to move one step.
- the peak detecting section 1993 outputs a spatio-temporal fragment extraction signal which is a signal indicating a parameter change for extracting spatio-temporal fragments. Output to output unit 14.
- the spatiotemporal fragment extraction unit 14 follows the parameters from the spatiotemporal data under the constraint of the parameter obtained from the moving direction calculation unit until the input of the fragment extraction line parameter change signal ends.
- the fragment extraction line parameter change signal may be changed for all parameter candidates based on the moving direction of the moving object candidate calculated by the moving direction calculator, or may be moved by the attribute output unit 17. It may be until object detection is performed.
- the peak detection unit 1993 transmits the fragment extraction line parameter to the coordinate conversion unit 1994.
- the coordinate converter 194 converts a straight line on the image when the peak is detected by the peak detector 193 into a straight line in world coordinates.
- the human body part motion model spatiotemporal fragment output unit 15 generates a human motion model according to the straight line in the world coordinate system converted here. The following operation is the same as in the first embodiment.
- the attribute output unit 17 specifies the existence position and the moving direction of the moving object, and outputs them to the display unit 18.
- the spatiotemporal fragment matching unit 16 includes a spatiotemporal fragment 70 when a peak is detected by the periodicity analysis unit 190 and a human body part motion model spatiotemporal fragment. Then, it performs a comparison with 7 1, and outputs a fragment extraction line parameter change signal, which is a signal indicating a parameter change of the fragment extraction line, to the spatiotemporal fragment extraction unit 14 based on the comparison result.
- the spatiotemporal fragment extraction unit 14 creates spatiotemporal fragments according to the parameters from the spatiotemporal data until the input of the fragment extraction line parameter change signal is completed.
- the fragment extraction line parameter change signal is The parameter may be changed for all parameter candidates based on the moving direction of the moving object candidate calculated by the direction calculation unit, or until the moving object detection is performed by the attribute output unit 17.
- the fragment extraction line parameter change signal is a and b which are the parameters of the fragment extraction line.
- the parameters of the fragment extraction line 23 and the parameters of the human motion model are reduced by calculating the moving direction of the moving object and analyzing the periodicity of the moving object. By doing so, it is possible to detect humans faster than in the case of performing spatiotemporal fragment matching while performing full search.
- FIG. 22 is a functional block diagram showing the configuration of the person model fitting device in the present embodiment.
- This person model fitting device is a device for fitting a model to a person present in an image without requiring a special device such as a marker to be attached to the subject, and includes a camera 10 and a video processing unit 11. , Continuous image processing unit 12, spatiotemporal volume generation unit 13, spatiotemporal fragment extraction unit 14, spontaneous body motion model spatiotemporal fragment output unit 15, spatiotemporal fragment matching unit 16, display unit 18, and model A fitting section 220 is provided.
- This configuration corresponds to a configuration in which the attribute output unit 17 is replaced with a model fitting unit 220 among the components included in the human detection device according to the first embodiment.
- the fitting of the person model refers to fitting the person motion model to a person existing in the image.
- the camera 10 video processing unit 11, continuous image processing unit 12, spatiotemporal volume generation unit 13, spatiotemporal fragment extraction unit 14, human body part motion model spatiotemporal fragment output unit 15, and spatiotemporal
- the operation of the fragment matching unit 16 is the same as that of the first embodiment, and a description thereof will be omitted.
- the model fitting unit 220 includes a model generating unit 231 and a coordinate transforming unit 232.
- the model generation unit 231 re-generates the human motion model shown in FIG. 7 from the combination result of the parameter indicating the position of the person on the first coordinate output from the spatiotemporal fragment matching unit 16. I do. Since the leg length L and the crotch angle 0 L of the human motion model are known, the human motion model shown in FIG. 7 can be generated from these parameters.
- model generation unit 2 31 has the same function as the human motion model generation unit 50 included in the human body part motion model spatiotemporal fragment output unit 15.
- the coordinate conversion unit 232 performs coordinate conversion to fit the human motion model on the image. I do. Specifically, among the combinations of the obtained parameters a high matching score, 0 w fragment extraction line parameter corresponds to the walking direction in world coordinates, parameter Ichita (X St art person movement model, y start ) Corresponds to the position of the person.
- the detection time is the number of steps of the scanning process performed by the matching processing unit 150 of the spatiotemporal fragment matching unit 16 attached to the parameter combination.
- the fitting is performed by superimposing the model obtained by performing coordinate transformation on the image at that time.
- the display unit 18 displays the image fitted by the model fitting unit 220.
- a special device such as a marker
- a model can be fitted to a person present in an image.
- present embodiment can be configured to reduce the calculation time as in the second, fourth, and fifth embodiments.
- the person detection device, the person verification device, and the person model fitting device according to the present invention have been described based on the six embodiments, but the present invention is not limited to these embodiments.
- the human motion model generation unit 50 of the human body part motion model spatiotemporal fragment output unit 15 uses the parameter of the fragment extraction line output from the spatiotemporal fragment extraction unit 14
- a human model having specific parameters is generated by using the above, a plurality of human motion model templates may be held in advance, and an optimal template may be selected and output from the human motion model templates.
- the human body part motion model spatiotemporal fragment output unit 15 is, as shown in Fig. 24, a human motion model template for each type of adult (male), adult (female), child and elderly.
- the spatiotemporal fragment collating unit 16 compares the spatiotemporal fragment output from the spatiotemporal fragment extraction unit 14 and the human body part motion model spatiotemporal fragment output from the human body part motion model spatiotemporal fragment output unit 15. Collate.
- the human body part motion model spatiotemporal fragment output unit 15 outputs A candidate human motion model template is selected, a human body part motion model spatiotemporal fragment of the human motion model template is output, and the spatiotemporal fragment collating unit 16 repeats the collation.
- the person motion model template having the highest matching degree is determined, and the type of the person in the target image is determined.
- each parameter was set to one value, but it was set within a certain range (for example, 300 as a child parameter). ⁇ 700 mm) may be set.
- the spatiotemporal fragment matching unit 16 repeatedly performs the matching calculation within the set range, so that the optimal value and the optimal type of human motion model template may be searched.
- such a person movement model template is not limited to each type of person, and as shown in Fig. 25, a plurality of templates are prepared for each situation of a walking road surface. In addition, as shown in Fig. 26, a plurality of templates may be prepared according to the degree of congestion in the walking place.
- the present invention relates to a person detection device that detects the presence, position, walking direction, and the like of a person, such as a monitoring device that is installed on a street or a facility, or an authentication device that is used when entering or leaving a building.
- a person detection device that detects the presence, position, walking direction, and the like of a person, such as a monitoring device that is installed on a street or a facility, or an authentication device that is used when entering or leaving a building.
- a person detection device that detects the presence, position, walking direction, and the like of a person, such as a monitoring device that is installed on a street or a facility, or an authentication device that is used when entering or leaving a building.
- the person ’s current position and walking direction are known, Even in the case where there is no detection area, it can be used as a person detection device, a person verification device, or the like that can perform person detection without limiting the detection area in the image.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2005513284A JP3775683B2 (ja) | 2003-08-21 | 2004-08-11 | 人物検出装置および人物検出方法 |
US11/320,416 US7613325B2 (en) | 2003-08-21 | 2005-12-29 | Human detection device and human detection method |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2003297118 | 2003-08-21 | ||
JP2003-297118 | 2003-08-21 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US11/320,416 Continuation US7613325B2 (en) | 2003-08-21 | 2005-12-29 | Human detection device and human detection method |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2005020152A1 true WO2005020152A1 (ja) | 2005-03-03 |
Family
ID=34213634
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2004/011790 WO2005020152A1 (ja) | 2003-08-21 | 2004-08-11 | 人物検出装置および人物検出方法 |
Country Status (4)
Country | Link |
---|---|
US (1) | US7613325B2 (ja) |
JP (1) | JP3775683B2 (ja) |
CN (1) | CN1839409A (ja) |
WO (1) | WO2005020152A1 (ja) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2008040781A (ja) * | 2006-08-07 | 2008-02-21 | Matsushita Electric Ind Co Ltd | 被写体照合装置および被写体照合方法 |
JP2009140009A (ja) * | 2007-12-03 | 2009-06-25 | Sony Corp | 情報処理装置および情報処理方法、プログラム、並びに記録媒体 |
JP2010079910A (ja) * | 2008-09-26 | 2010-04-08 | Axis Ab | ビデオモーション検出方法 |
WO2011161924A1 (ja) * | 2010-06-23 | 2011-12-29 | 国立大学法人大阪大学 | 移動物体検出装置 |
WO2018043437A1 (ja) * | 2016-09-01 | 2018-03-08 | 公立大学法人会津大学 | 画像距離算出装置および画像距離算出用プログラムを記録したコンピュータ読み取り可能な非一過性の記録媒体 |
JP2018040789A (ja) * | 2016-09-01 | 2018-03-15 | 公立大学法人会津大学 | 画像距離算出装置、画像距離算出方法および画像距離算出用プログラム |
Families Citing this family (30)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP4432929B2 (ja) * | 2006-04-25 | 2010-03-17 | トヨタ自動車株式会社 | 駐車支援装置及び駐車支援方法 |
EP1870857A1 (en) * | 2006-06-19 | 2007-12-26 | Koninklijke Philips Electronics N.V. | Global motion estimation |
JP4386083B2 (ja) * | 2007-02-27 | 2009-12-16 | トヨタ自動車株式会社 | 駐車支援装置 |
JP4315991B2 (ja) * | 2007-04-20 | 2009-08-19 | 本田技研工業株式会社 | 車両周辺監視装置、車両周辺監視方法、車両周辺監視プログラム |
CN101515374B (zh) * | 2008-02-20 | 2010-12-01 | 中国科学院自动化研究所 | 基于图像的个性化真实感虚拟人物造型方法 |
US8170278B2 (en) | 2008-08-06 | 2012-05-01 | Sri International | System and method for detecting and tracking an object of interest in spatio-temporal space |
WO2010098024A1 (ja) * | 2009-02-24 | 2010-09-02 | 三菱電機株式会社 | 人物追跡装置及び人物追跡プログラム |
US8744121B2 (en) | 2009-05-29 | 2014-06-03 | Microsoft Corporation | Device for identifying and tracking multiple humans over time |
JP2011114823A (ja) * | 2009-11-30 | 2011-06-09 | Sanyo Electric Co Ltd | 画像処理装置及び撮像装置 |
WO2012077287A1 (ja) | 2010-12-09 | 2012-06-14 | パナソニック株式会社 | 姿勢状態推定装置および姿勢状態推定方法 |
US8913129B2 (en) * | 2011-01-27 | 2014-12-16 | Thermal Matrix USA, Inc. | Method and system of progressive analysis for assessment of occluded data and redundant analysis for confidence efficacy of non-occluded data |
EP2693373A4 (en) * | 2011-03-31 | 2015-02-18 | Panasonic Corp | DEVICE FOR MEASURING THE NUMBER OF PEOPLE |
CN102609680B (zh) * | 2011-12-22 | 2013-12-04 | 中国科学院自动化研究所 | 一种基于三维深度图像信息的并行统计学习人体部位检测方法 |
WO2013114862A1 (ja) * | 2012-01-30 | 2013-08-08 | パナソニック株式会社 | 最適カメラ設定装置及び最適カメラ設定方法 |
US8872850B2 (en) * | 2012-03-05 | 2014-10-28 | Microsoft Corporation | Juxtaposing still and dynamic imagery for cliplet creation |
US10009579B2 (en) | 2012-11-21 | 2018-06-26 | Pelco, Inc. | Method and system for counting people using depth sensor |
US9367733B2 (en) * | 2012-11-21 | 2016-06-14 | Pelco, Inc. | Method and apparatus for detecting people by a surveillance system |
JP5961123B2 (ja) * | 2013-01-30 | 2016-08-02 | パナソニック株式会社 | 人物検出追跡装置、人物検出追跡方法、および人物検出追跡プログラム |
US9639747B2 (en) | 2013-03-15 | 2017-05-02 | Pelco, Inc. | Online learning method for people detection and counting for retail stores |
US10347100B2 (en) | 2013-04-26 | 2019-07-09 | Nec Corporation | Information processing system, method and computer readable medium for determining whether moving bodies appearing in first and second videos are the same or not |
US9256348B2 (en) * | 2013-12-18 | 2016-02-09 | Dassault Systemes Americas Corp. | Posture creation with tool pickup |
KR101679741B1 (ko) * | 2015-05-06 | 2016-11-28 | 고려대학교 산학협력단 | 외곽 공간 특징 정보 추출 방법 |
US9648211B2 (en) * | 2015-05-14 | 2017-05-09 | Xerox Corporation | Automatic video synchronization via analysis in the spatiotemporal domain |
CN105892538B (zh) * | 2016-04-22 | 2018-08-31 | 郑贵林 | 用于构建主动消防体系及智慧建筑的智能终端和智能终端系统 |
US10402643B2 (en) * | 2016-06-15 | 2019-09-03 | Google Llc | Object rejection system and method |
JP6701018B2 (ja) * | 2016-07-19 | 2020-05-27 | キヤノン株式会社 | 情報処理装置、情報処理方法及びプログラム |
CN106652002A (zh) * | 2016-09-20 | 2017-05-10 | 努比亚技术有限公司 | 一种轨迹动画影像的生成方法及生成系统 |
EP3407232B1 (en) * | 2017-05-23 | 2021-07-28 | Ordnance Survey Limited | Spatiotemporal authentication |
US11238046B2 (en) * | 2018-02-19 | 2022-02-01 | Nippon Telegraph And Telephone Corporation | Information management device, information management method, and information management program |
US11315363B2 (en) * | 2020-01-22 | 2022-04-26 | Board Of Trustees Of Michigan State University | Systems and methods for gait recognition via disentangled representation learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH07121499A (ja) * | 1993-10-20 | 1995-05-12 | Nippon Telegr & Teleph Corp <Ntt> | 動きパラメータ推定方法 |
JP2001109891A (ja) * | 1999-10-08 | 2001-04-20 | Toshiba Corp | 物体領域抽出方法 |
JP2001202379A (ja) * | 2000-01-21 | 2001-07-27 | Kobe University | 仮想空間における仮想人物動作データブラウジング装置 |
JP2001266131A (ja) * | 2000-03-23 | 2001-09-28 | Minolta Co Ltd | 監視システム |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3183320B2 (ja) | 1994-08-31 | 2001-07-09 | 日本電信電話株式会社 | 移動物体方向別計数方法および装置 |
EP0700017B1 (en) * | 1994-08-31 | 2001-11-07 | Nippon Telegraph And Telephone Corporation | Method and apparatus for directional counting of moving objects |
JP3603737B2 (ja) * | 2000-03-30 | 2004-12-22 | 日本電気株式会社 | 移動体追尾方法及びその装置 |
-
2004
- 2004-08-11 CN CNA2004800240071A patent/CN1839409A/zh not_active Withdrawn
- 2004-08-11 WO PCT/JP2004/011790 patent/WO2005020152A1/ja active Application Filing
- 2004-08-11 JP JP2005513284A patent/JP3775683B2/ja not_active Expired - Fee Related
-
2005
- 2005-12-29 US US11/320,416 patent/US7613325B2/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH07121499A (ja) * | 1993-10-20 | 1995-05-12 | Nippon Telegr & Teleph Corp <Ntt> | 動きパラメータ推定方法 |
JP2001109891A (ja) * | 1999-10-08 | 2001-04-20 | Toshiba Corp | 物体領域抽出方法 |
JP2001202379A (ja) * | 2000-01-21 | 2001-07-27 | Kobe University | 仮想空間における仮想人物動作データブラウジング装置 |
JP2001266131A (ja) * | 2000-03-23 | 2001-09-28 | Minolta Co Ltd | 監視システム |
Non-Patent Citations (1)
Title |
---|
NIYOGI A. ET AL.: "Analyzing and recognizing walking figures in XYT", M.I.T. MEDIA AND LAB VISION AND MODELING GROUP TECHNICAL REPORT, no. 223, 1994, XP000515879 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2008040781A (ja) * | 2006-08-07 | 2008-02-21 | Matsushita Electric Ind Co Ltd | 被写体照合装置および被写体照合方法 |
US8208028B2 (en) | 2006-08-07 | 2012-06-26 | Panasonic Corporation | Object verification device and object verification method |
JP2009140009A (ja) * | 2007-12-03 | 2009-06-25 | Sony Corp | 情報処理装置および情報処理方法、プログラム、並びに記録媒体 |
JP2010079910A (ja) * | 2008-09-26 | 2010-04-08 | Axis Ab | ビデオモーション検出方法 |
US8121424B2 (en) | 2008-09-26 | 2012-02-21 | Axis Ab | System, computer program product and associated methodology for video motion detection using spatio-temporal slice processing |
WO2011161924A1 (ja) * | 2010-06-23 | 2011-12-29 | 国立大学法人大阪大学 | 移動物体検出装置 |
JP5467300B2 (ja) * | 2010-06-23 | 2014-04-09 | 国立大学法人大阪大学 | 移動物体検出装置 |
US8958641B2 (en) | 2010-06-23 | 2015-02-17 | Osaka University | Moving object detection device |
WO2018043437A1 (ja) * | 2016-09-01 | 2018-03-08 | 公立大学法人会津大学 | 画像距離算出装置および画像距離算出用プログラムを記録したコンピュータ読み取り可能な非一過性の記録媒体 |
JP2018040789A (ja) * | 2016-09-01 | 2018-03-15 | 公立大学法人会津大学 | 画像距離算出装置、画像距離算出方法および画像距離算出用プログラム |
US10803609B2 (en) | 2016-09-01 | 2020-10-13 | The Public University Corporation, The University Aizu | Image distance calculator and computer-readable, non-transitory storage medium storing image distance calculation program |
Also Published As
Publication number | Publication date |
---|---|
JP3775683B2 (ja) | 2006-05-17 |
JPWO2005020152A1 (ja) | 2006-10-19 |
US20060115116A1 (en) | 2006-06-01 |
CN1839409A (zh) | 2006-09-27 |
US7613325B2 (en) | 2009-11-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2005020152A1 (ja) | 人物検出装置および人物検出方法 | |
US7330566B2 (en) | Video-based gait recognition | |
Shen et al. | Multiobject tracking by submodular optimization | |
US7702019B2 (en) | Moving object detection device and moving object detection method | |
JP5820366B2 (ja) | 姿勢推定装置及び姿勢推定方法 | |
RU2546327C1 (ru) | Устройство для отслеживания человека, способ отслеживания человека и невременный машиночитаемый носитель, хранящий программу для отслеживания человека | |
JP5424852B2 (ja) | 映像情報処理方法及びその装置 | |
US20060269145A1 (en) | Method and system for determining object pose from images | |
Świtoński et al. | Human identification based on gait paths | |
CN105531995A (zh) | 用于使用多个摄像机进行对象和事件识别的系统和方法 | |
KR20140114741A (ko) | 인체 포즈 추정 장치 및 방법 | |
JP2008219570A (ja) | カメラ間連結関係情報生成装置 | |
JP3655618B2 (ja) | 歩行者年齢判定装置、歩行状態・歩行者年齢判定方法およびプログラム | |
CN109583373A (zh) | 一种行人重识别实现方法 | |
KR102475752B1 (ko) | 이동 경로 예측 방법 및 장치 | |
JP2007028680A (ja) | 監視装置 | |
JP2020052822A (ja) | 情報処理装置、認証システムおよびそれらの制御方法、プログラム | |
JP2010015472A (ja) | 歩行者性別判定装置、歩行者性別判定方法、歩行者性別判定プログラムおよびその記録媒体 | |
Ramasso et al. | Human shape-motion analysis in athletics videos for coarse to fine action/activity recognition using transferable belief model | |
Albu et al. | Generic temporal segmentation of cyclic human motion | |
Swain et al. | Human Pose Estimation Using GNN | |
JP6893812B2 (ja) | 物体検出装置 | |
Panagiotakis et al. | Shape-motion based athlete tracking for multilevel action recognition | |
Serrano et al. | Automated feet detection for clinical gait assessment | |
Hild | Estimation of 3d motion trajectory and velocity from monocular image sequences in the context of human gait recognition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 200480024007.1 Country of ref document: CN |
|
AK | Designated states |
Kind code of ref document: A1 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BW BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE EG ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NA NI NO NZ OM PG PH PL PT RO RU SC SD SE SG SK SL SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW |
|
AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): BW GH GM KE LS MW MZ NA SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IT LU MC NL PL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG |
|
DPEN | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed from 20040101) | ||
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
WWE | Wipo information: entry into national phase |
Ref document number: 2005513284 Country of ref document: JP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 11320416 Country of ref document: US |
|
WWP | Wipo information: published in national office |
Ref document number: 11320416 Country of ref document: US |
|
122 | Ep: pct application non-entry in european phase |