WO2009113231A1 - 画像処理装置および画像処理方法 - Google Patents
画像処理装置および画像処理方法 Download PDFInfo
- Publication number
- WO2009113231A1 WO2009113231A1 PCT/JP2009/000317 JP2009000317W WO2009113231A1 WO 2009113231 A1 WO2009113231 A1 WO 2009113231A1 JP 2009000317 W JP2009000317 W JP 2009000317W WO 2009113231 A1 WO2009113231 A1 WO 2009113231A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- contour
- tracking
- unit
- image frame
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/174—Segmentation; Edge detection involving the use of two or more images
-
- 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
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/167—Detection; Localisation; Normalisation using comparisons between temporally consecutive images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
Definitions
- the present invention relates to information processing technology, and more particularly, to an image processing apparatus that analyzes an image and extracts features and an image processing method that is executed there.
- the technique for extracting the contour of an object is a key technique in a wide range of fields such as visual tracking, computer vision, medical image analysis, and retouching. Since the contour line of the object in the image can be grasped as a part of the edge, an edge extraction filter is often used to extract the contour line.
- the existence probability distribution of the tracking target is expressed using a finite number of particles, and the candidate contour line having the same shape as the tracking target, which is determined by one particle, By matching with the edge image, the likelihood of each particle is observed, and the next existence probability distribution is estimated (for example, see Non-Patent Documents 1 to 3).
- a contour model of an object is expressed by a closed curve, and the contour of the object is estimated by deforming the closed curve so that a predefined energy function is minimized.
- An active contour model (snakes) has also been proposed (see Patent Document 1 or 2). Contour tracking by stochastic propagation of conditional density, Michael Isard and Andrew Blake, Proc.European Conf. On Computer Vision, vol. 1, pp.343-356, Cambridge UK (1996) CONDENSATION-conditional density propagation for visual tracking, Michael Isard and Andrew Blake, Int. J.
- edge extraction filters In general edge extraction filters, it is often the case that detailed shadows and patterns other than the contour are extracted or the contour line is interrupted or not extracted depending on the shooting conditions and setting parameters. This is because, in the edge extraction filter, the edge region is determined by the threshold value for the intermediate value obtained by filtering, so the edge extraction frequency of the entire screen changes depending on the threshold value setting, and only the contour line is extracted. This is due to the fact that it may be difficult.
- the active contour model there is a problem that the initial setting of the contour model is required for each target object, or the final result is influenced by the initial setting. Further, since the amount of calculation is large, there is a problem that, when the contour of the object in the moving image is sequentially obtained, such as the above-described visual tracking technique, it is impossible to follow the change in the shape of the object.
- the present invention has been made in view of such problems, and an object of the present invention is to provide a technique for extracting the contour of an object at high speed regardless of the content of the image. It is another object of the present invention to provide a technique capable of performing visual tracking with high accuracy even when the photographing environment changes.
- An aspect of the present invention relates to an object tracking device.
- This object tracking device is based on the estimated existence probability distribution of the object in the first image frame out of the first image frame and the second image frame constituting the moving image obtained by capturing the object to be tracked. Determining a candidate contour of the object in the second image frame, matching the edge image of the second image frame, observing the likelihood of the candidate contour, and estimating the existence probability distribution of the object in the second image frame
- a tracking processing unit that performs tracking processing of the object, a tracking target region detecting unit that detects a region of the target in the first image frame by a predetermined analysis method, and acquires a predetermined feature amount representing the region;
- the tracking processing unit temporarily sets at least one of the parameters used for the tracking processing to obtain an estimated existence probability distribution of the object in the first image frame, and based on this
- the initial contour estimation unit that estimates the contour of the target object, the feature amount of the target region based on the contour estimated by the initial contour estimation unit, and the feature amount acquired by the
- first image frame and the “second image frame” may be adjacent image frames in the image stream or may be image frames positioned apart from each other.
- first image frame is an image frame temporally prior to the “second image frame”. Is not limited to this.
- the “existence probability distribution” may be an existence probability distribution with respect to the position coordinates in the image frame of the object, and a parameter representing any of the attributes of the object such as shape, color, size, or a combination thereof. It may be an existence probability distribution for the extended space.
- a “candidate contour” is a graphic representing a candidate for a contour line of a part or the whole of an object.
- the “likelihood” is a degree representing how close the candidate contour is to the object, for example, a numerical value indicating the degree of overlap with the object, the distance from the object, and the like.
- Another aspect of the present invention relates to an object tracking method.
- the computer calculates the estimated existence probability distribution of the object in the first image frame out of the first image frame and the second image frame constituting the moving image obtained by capturing the object to be tracked. Based on this, the candidate contour of the target object in the second image frame is determined, the likelihood of the candidate contour is observed by matching with the edge image of the second image frame, and the existence probability distribution of the target object in the second image frame is determined.
- a method of performing tracking processing of an object by estimating, reading a first image frame from a memory storing a moving image, detecting a region of the object in the image frame by a predetermined analysis method, And acquiring a predetermined feature amount representing at least one of parameters used for the tracking process and temporarily setting the value in the first image frame
- An estimated existence probability distribution of the target object and based on the estimated probability distribution of the target object, the feature amount acquired in the step of acquiring the feature amount, and the target object based on the contour estimated in the step of estimating the contour Comparing the feature quantity of the region, and starting a tracking process by applying a temporarily set parameter value when the comparison result satisfies a predetermined condition.
- Still another embodiment of the present invention relates to an image processing apparatus.
- This image processing apparatus generates a low gradation image with a reduced number of gradations of an image, extracts edges from the low gradation image generated by the gradation reduction unit, and extracts an original image And a contour image generation unit that generates a contour image in which the contour line of the subject is emphasized.
- This image processing apparatus has an estimated existence probability of a tracking object in a first image frame out of a first image frame and a second image frame included in an image stream that constitutes moving image data obtained by capturing an object to be tracked. Based on the distribution, a candidate contour determining unit that determines the candidate contour of the tracking target in the second image frame, the candidate contour determined by the candidate contour determining unit, and the contour of the second image frame generated by the contour image generating unit An observation unit that matches the images and observes the likelihood of the candidate contour; a tracking result acquisition unit that estimates the existence probability distribution of the object in the second image frame based on the likelihood observed by the observation unit; May be further provided.
- Still another aspect of the present invention relates to an image processing method.
- an image stored in a memory is read, a low gradation image with a reduced number of gradations is generated, an edge is extracted from the low gradation image, and a contour line of a subject in the original image is obtained. Generating a contour image with emphasis on.
- FIG. 1 is a diagram illustrating a configuration example of a visual tracking system in Embodiment 1.
- FIG. It is a figure which shows the structure of the tracking apparatus in Embodiment 1 in detail.
- 3 is a flowchart showing a procedure of tracking processing in the first embodiment.
- FIG. 3 is a diagram illustrating the configuration of an observation unit in the first embodiment in more detail.
- FIG. 6 is a diagram schematically showing a state in which image data of a region cut out from a contour image stored in an image storage unit in the first embodiment is copied to each local memory.
- FIG. 6 is a diagram schematically illustrating a process transition when the first processing unit, the second processing unit,..., The Nth processing unit of the contour search unit perform the contour search processing in the first embodiment.
- 3 is a diagram illustrating a detailed configuration of a contour image generation unit according to Embodiment 1.
- FIG. It is a figure for demonstrating the difference of a general edge extraction process and the outline image generation process in this Embodiment. It is a figure which shows the example of the original image which is a process target. It is a figure which shows the edge image produced
- FIG. 12 is a diagram showing a gradation-reduced image obtained as an intermediate image when the contour image generation process of the present embodiment is performed on the original image shown in FIG. 11 in the first embodiment. It is a figure which shows the outline image produced
- 10 is a flowchart showing a procedure of tracking processing in the second embodiment. 9 is a flowchart illustrating a procedure for setting environment-dependent parameters in the second embodiment. It is a figure which shows typically a mode that the tracking environment setting part in Embodiment 2 determines an environment dependence parameter.
- FIG. 10 is a diagram for describing a setting order when provisionally setting environment-dependent parameters in a tracking environment setting unit according to the second embodiment.
- FIG. 1 is a diagram for explaining a visual tracking method when a person is a tracking target.
- the person image 150 is one of the image frames constituting the image stream of the moving image generated by the moving image or the computer graphics that is actually captured, and the person 152 to be tracked is captured.
- an ⁇ -shaped curve 154 that approximates the shape of the head contour of the person 152 is described in a known expression.
- the person image 150 including the person 152 is subjected to edge extraction processing to obtain an edge image. Then, by changing the shape and position while changing the parameter defining the curve 154 and searching for an edge in the vicinity thereof, the value of the parameter estimated to be the best match with the head contour of the person 152 is specified.
- the tracking of the person 152 progresses by repeating the above processing for each frame.
- the edge generally refers to a portion having an abrupt change in image density or color.
- a probability distribution prediction technique using a particle filter is introduced.
- the number of samplings of the curve 154 is increased or decreased according to the probability distribution of the object in the parameter space in the immediately preceding frame, and the tracking candidates are narrowed down.
- Non-Patent Document 3 (ICondensation: Unifying low-level and high-level tracking in a stochastic framework, Michael Isard and Andrew Blake, Proc 5th European Conf. Computer Vision, 1998).
- the description will be focused on the points according to the present embodiment.
- the ⁇ -shaped curve 154 is described as a B-spline curve.
- the B-spline curve is defined by n control point sequences (Q0,..., Qn) and knot sequences (s0,..., Sn). These parameters are set in advance so as to form a basic curve shape, in this case, an ⁇ -shaped curve.
- the curve obtained by the setting at this time is hereinafter referred to as template Q0.
- the template Q0 has an ⁇ shape, but the shape is changed depending on the tracking target. That is, if the tracking target is a ball, the shape is circular, and if the tracking target is a palm, the shape is a hand.
- a shape space vector x is prepared as a conversion parameter for changing the shape of the template.
- the shape space vector x is composed of the following six parameters.
- (shift x , shift y ) is a translation amount in the (x, y) direction
- (extend x , extend y ) is a magnification
- ⁇ is a rotation angle.
- the template can be translated, stretched and rotated by appropriately changing the six parameters constituting the shape space vector x, and the shape and position of the candidate curve Q can be changed variously depending on the combination. it can.
- a plurality of candidate curves expressed by changing the parameters of the template Q 0 such as the interval between the control point sequence and the knot sequence and the six parameters constituting the shape space vector x are persons in the vicinity of each knot. Search for 152 edges. Thereafter, the likelihood density distribution in the six-dimensional space spanned by the six parameters constituting the shape space vector x is estimated by obtaining the likelihood of each candidate curve from the distance to the edge or the like.
- FIG. 2 is a diagram for explaining a probability density distribution estimation method using a particle filter.
- the change of a certain parameter x1 is shown on the horizontal axis, but actually the same processing is performed in the 6-dimensional space. Is called.
- the image frame whose probability density distribution is to be estimated is the image frame at time t.
- a particle is a materialization of the value of the parameter x1 to be sampled and the sampling density. For example, in the region of the parameter x1, where the probability density was high at time t-1, sampling is performed by increasing the particle density. In the range where the probability density is low, sampling is not performed by reducing the number of particles. Thereby, for example, many candidate curves are generated near the edge of the person 152, and matching is performed efficiently.
- the predetermined motion model is, for example, a Gaussian motion model, an autoregressive prediction motion model, or the like.
- the former is a model in which the probability density at time t is Gaussian distributed around each probability density at time t-1.
- the latter is a method that assumes a second-order or higher-order autoregressive prediction model acquired from sample data. For example, it is estimated from a change in past parameters that a person 152 is moving at a certain speed. In the example of FIG. 2, the motion in the positive direction of the parameter x1 is estimated by the autoregressive prediction type motion model, and each particle is changed in that way.
- the likelihood of each candidate curve is obtained by searching for the edge of the person 152 in the vicinity of the candidate curve determined by each particle using the edge image at time t, and the probability density distribution at time t is obtained.
- Estimate (S16) As described above, the probability density distribution at this time is a discrete representation of the true probability density distribution 400 as shown in S16. Thereafter, by repeating this, the probability density distribution at each time is represented in the parameter space. For example, if the probability density distribution is unimodal, that is, if the tracked object is unique, the final parameter is the weighted sum of each parameter value using the obtained probability density. Thus, the contour curve closest to the tracking target is obtained.
- the probability density distribution p (x t i ) at time t estimated in S16 is calculated as follows.
- i is a number uniquely given to a particle
- x t i , u t ⁇ 1 ) is a predetermined motion model
- x t i ) is a likelihood.
- FIG. 3 shows a configuration example of the visual tracking system in the present embodiment.
- the visual tracking system 10 includes an imaging device 12 that images the tracking target 18, a tracking device 14 that performs tracking processing, and a display device 16 that outputs image data captured by the imaging device 12 and tracking result data.
- the tracking target 18 may vary depending on the purpose of use of the visual tracking system 10 such as a person, an object, or a part thereof, but in the following description, it is assumed that the person is a person as in the above example.
- connection between the tracking device 14 and the imaging device 12 or the display device 16 may be wired or wireless, and may be via various networks. Alternatively, any two or all of the imaging device 12, the tracking device 14, and the display device 16 may be combined and integrally provided. Depending on the usage environment, the imaging device 12 and the display device 16 may not be connected to the tracking device 14 at the same time.
- the imaging device 12 acquires data of an image including the tracking target 18 or an image of a certain place regardless of the presence or absence of the tracking target 18 at a predetermined frame rate.
- the acquired image data is input to the tracking device 14, and the tracking processing of the tracking target 18 is performed.
- the processing result is output as output data to the display device 16 under the control of the tracking device 14.
- the tracking device 14 may also serve as a computer that performs another function, and may implement various functions by using data obtained as a result of the tracking process, that is, position information and shape information of the tracking target 18. .
- FIG. 4 shows the configuration of the tracking device 14 in the present embodiment in detail.
- the tracking device 14 includes an image acquisition unit 20 that acquires input image data input from the imaging device 12, an image storage unit 24 that stores the input image data and the contour image data, and a contour image that generates a contour image from the input image data.
- a generation unit 22 a tracking start / end determination unit 28 that determines the start and end of tracking, a tracking processing unit 26 that performs tracking processing using a particle filter, a result storage unit 36 that stores final tracking result data, and a tracking result
- the output control part 40 which controls the output to the display apparatus 16 is included.
- each element described as a functional block for performing various processes can be configured by a CPU, a memory, and other LSI in terms of hardware, and a program for performing image processing in terms of software. It is realized by. Therefore, it is understood by those skilled in the art that these functional blocks can be realized in various forms by hardware only, software only, or a combination thereof, and is not limited to any one.
- the contour image generation unit 22 extracts the contour line to be tracked from the image frame of the input image, and generates a contour image.
- the contour image is stored in the image storage unit 24 and later used in the observation of the candidate curve in the observation unit 30 of the tracking processing unit 26.
- “contour lines” are treated as “edges” in an edge image, and thus, with the conventional technique, likelihood observation using an “edge image” has been performed.
- edge extraction filters many edges are extracted in addition to the contour line of an object depending on the input image, so it is considered that likelihood observation cannot be performed accurately due to matching with edges other than the contour line. It is done. Further, if the threshold value for edge extraction is set high and the number of edges is reduced, the contour line is cut off, and there is a possibility that the likelihood observation is not accurately performed.
- the contour image generation unit 22 generates not only a general “edge image” but also an image that can accurately perform likelihood observation by focusing on the “contour” of an object in the input image. To do. Although a specific method will be described later, in the following description, an image generated by the contour image generation unit 22 is distinguished from a general “edge image” as a “contour image”. Further, the contour image generation unit 22 may be mounted with a foreground extractor (not shown) using a background difference. Then, the contour of the tracking target may be efficiently extracted by extracting the foreground including the tracking target from the input image as preprocessing of the contour image generation processing.
- the tracking start / end determining unit 28 evaluates, for example, the contour line or the foreground shape obtained by the contour image generating unit 22 and determines whether to start or end tracking according to a predetermined condition.
- end may include a temporary stop of tracking by occlusion or the like. Tracking starts when the tracking target appears within the viewing angle of the imaging device or when it appears from behind the object, etc.When the tracking target leaves the viewing angle of the imaging device or enters the shadow, etc. To finish. When it is determined that the tracking is to be started, the tracking processing unit 26 is notified to that effect and the tracking process is started.
- the tracking processing unit 26 includes a sampling unit 29, an observation unit 30, and a result acquisition unit 34.
- the sampling unit 29 performs particle generation and extinction processing based on the probability density distribution estimated for the previous image frame at time t-1. Then, a predetermined motion model is applied to all the particles to cause the particles to transition on the parameter space. Thereby, a plurality of candidate curves in the image frame at time t are determined.
- the sampling unit 29 starts the process when receiving a signal indicating the start of tracking from the tracking start / end determining unit 28, and ends the process when receiving a signal indicating the end of tracking.
- the observation unit 30 observes the likelihood of the candidate curve defined by each particle generated, disappeared, and transitioned by the sampling unit. For example, when the candidate curve defined by each particle is expressed by a B-spline curve, for each knot of the B-spline curve, the nearest contour line is searched for in the contour image generated by the contour image generation unit 22 to obtain the distance. To score the knots according to a predetermined rule. Then, the likelihood of the candidate curve is obtained based on the scores of all knots constituting the candidate curve. The observation unit 30 executes this search process in parallel using a plurality of processor units.
- a processing unit obtained by dividing the contour search processing for each knot is used as one processing unit, and a plurality of processor units perform parallel processing.
- Each processor unit copies only the image data of a part of the contour image including the knot and the search region to the subordinate local memory in order to search for the contour line closest to one knot. .
- the number of processing units of (number of particles) ⁇ (number of knots constituting candidate curve) is processed in a short time for each tracking target.
- the score of each knot acquired in parallel by each processor unit is integrated for each candidate curve, and the likelihood is calculated. Conventional techniques can be used for scoring and likelihood calculation.
- the result acquisition unit 34 calculates a probability density distribution p (x t i ) as shown in Expression 3 based on the likelihood observed by the observation unit 30, and tracks the curve data obtained by the weighted average parameter.
- the result is calculated and stored in the result storage unit 36.
- the data is returned to the sampling unit 29 for use in the tracking process at the next time t + 1.
- the data stored in the result storage unit 36 may be the value of each parameter that has been weighted and averaged, and may be any of an image composed only of a curve determined by that, or image data formed by combining a curve and an input image But you can.
- the result acquisition unit 34 may further track each tracking target using a template prepared for each, and combine the tracking results into one tracking result. Further, a case where a plurality of tracking targets overlap is detected based on the tracking result, and the tracking target hidden behind is taken out of the tracking processing target at a predetermined timing. As a result, even if the observation likelihood is temporarily lowered due to the tracking target moving behind another tracking target, it is possible to avoid outputting an inappropriate tracking result.
- the result storage unit 36 stores, for example, moving image data including the tracking result.
- the result storage unit 36 stores, for example, moving image data including the tracking result.
- the display device 16 by outputting the moving image data to the display device 16 under the control of the output control unit 40, it is possible to display how the curve of the template moves in the same manner as the movement of the tracking target.
- processing such as output to another arithmetic module may be appropriately performed according to the purpose of tracking.
- the imaging device 12 captures a place to be captured at a predetermined frame rate.
- the captured image is input as input image data to the image acquisition unit 20 of the tracking device 14 and stored in the image storage unit 24. In such a state, the following tracking process is performed.
- FIG. 5 is a flowchart showing the procedure of the tracking process in the present embodiment.
- the tracking target is a person
- an ⁇ -type template is prepared for the tracking device 14 as described above.
- the template expression method is not limited to the B-spline curve, but may be any description format that can express a desired curve.
- the template shape deformation method may be appropriately selected from methods that can be adapted to the description format and that can be flexibly deformed as described above by changing several types of parameters.
- the tracking start / end determination unit 28 reads the input image data stored in the image storage unit 24 for each frame and determines whether or not to start tracking (S20, S22). For example, when an object having a predetermined size and shape that can be estimated as a person appears as a foreground extracted from an image frame, it is determined to start tracking.
- the size and shape of the foreground used as a judgment criterion are determined in advance logically or experimentally.
- Foreground extraction processing may use a foreground extractor (not shown) mounted on the contour image generation unit 22. In this case, the tracking start / end determining unit 28 requests the contour image generating unit 22 for foreground extraction processing.
- the tracking start / end determination unit 28 may be equipped with a foreground extractor.
- the tracking processing unit 26 starts the tracking process.
- the sampling unit 29 requests the contour image generation unit 22 for a contour image generation process
- the sampling unit 29 may also request a contour image generation process for the subsequent frame, and the contour image generation unit 22 may sequentially perform the process.
- the sampling unit 29 performs sampling by arranging particles evenly in a predetermined region of the parameter space, for example, and the observation unit 30 observes the likelihood by matching the candidate curve defined by each particle and the contour image, and the result
- the acquisition unit 34 calculates the initial value p (x 0 i ) of the rate density distribution using Equation 3 (S28, S30, S32).
- the sampling unit 29 generates a number of particles corresponding to the generated initial value p (x 0 i ) of the probability density distribution on the parameter space, and performs sampling by transitioning the particles based on a predetermined motion model ( S28).
- the number of particles to be generated is controlled in consideration of the processing load based on the amount of computation resources of the tracking device 14 and the required result output speed.
- a motion model that can be obtained with high tracking accuracy from a Gaussian motion model, an autoregressive motion model, or the like is determined in advance according to the type of tracking target.
- the observation unit 30 observes the likelihood p (y t
- search processing is allocated to a plurality of processors for each knot. Details will be described later.
- the tracking start / end determining unit 28 determines whether to continue or end the tracking process (S34). For example, the tracking end is determined when a target having a predetermined size and shape that can be estimated as a person does not appear as a foreground for a predetermined time. Alternatively, the end of tracking is determined when the occlusion state continues for a predetermined time, such as when a tracking target in real space moves behind another tracking target. The occlusion state may be estimated from past tracking processing results, or may be detected by a distance measurement system (not shown). Further, a situation in which the tracking target deviates from the angle of view of the imaging device 12 continues for a predetermined time is also detected by a method similar to occlusion and the tracking end is determined.
- FIG. 6 shows the configuration of the observation unit 30 in more detail.
- the observation unit 30 includes a contour image cutout unit 50, a contour search task queue 52, a contour search unit 56, and a likelihood acquisition unit 54.
- the contour image cutout unit 50 cuts out a region corresponding to each knot from the contour image based on the knot coordinates of the curve representing the candidate curve.
- the area corresponding to each knot is an area including the knot and a search area for the knot. It may be equal to the search area, and is also referred to as “search area” in the following description.
- a processing request for contour search including coordinate information of knots and information related to the corresponding region is issued. The issued processing request is added to the processing queue in the contour search task queue 52.
- the contour search unit 56 includes N processing units of a first processing unit 58a, a second processing unit 58b, a third processing unit 58c,..., An Nth processing unit 58n, and local memories 60a and 60b connected to the N processing units. , 60c, ..., 60d.
- Each processing unit sequentially reads out contour search processing requests from the contour search task queue 52 and executes the contour search processing for the requested knots. Specifically, the image data of the area specified by the processing request is copied from the contour image stored in the image storage unit 24 to the subordinate local memory. Based on the coordinates of the designated knot, the contour line closest to the knot is searched in the area copied to the local memory and scored according to a predetermined rule.
- a generally used edge search technique can be used for contour search.
- the search area can be determined by the selected search method and the accuracy required for matching.
- the N-th processing unit 58n executes one processing request, the obtained score is output to the likelihood acquisition unit 54. To do. Then, the next contour search processing request is read from the contour search task queue 52, and the same processing is repeated.
- the likelihood acquisition unit 54 integrates the scores of the knots respectively input from the first processing unit 58a, the second processing unit 58b, the third processing unit 58c, ..., the N-th processing unit 58n of the contour search unit.
- the likelihood for each candidate curve is calculated.
- the scores of all knots constituting the candidate curve are collected and summed, an average value is calculated, or substituted into a predetermined conversion formula.
- the scores for the knots are output one after another from each processing unit of the contour search unit 56, and information that associates the identification information of the knots with the identification information of the candidate curve to which the knot belongs is shared in the observation unit 30.
- the result can be integrated for each candidate curve by storing it and including necessary identification information when processing requests and scores are output.
- FIG. 6 schematically shows how the image data of the designated area is copied from the contour image to the local memories 60a, 60b, 60c,.
- the image storage unit 24 stores a contour image 90 generated by the contour image generation unit 22.
- the coordinates of the knots 92 of the candidate curve are determined by the particles determined by the sampling unit 29.
- the contour image cutout unit 50 When the contour image cutout unit 50 acquires the coordinates of the knots 92, the contour image cutout unit 50 cuts out the search area 94 of the knots for each knot.
- the search area is determined in consideration of accuracy, memory capacity, processing speed, and the like. In the example of FIG. 7, a square whose center of gravity is the coordinate of the knot 92 is cut out as the search area 94. From the viewpoint of search accuracy, it is desirable to change the size of the search area 94 according to the size of the candidate curve. For example, the maximum data size of the search area 94 with respect to the maximum size that can be taken by the candidate curve is set to be equal to the maximum data size that can be stored in the copy areas of the local memories 60a, 60b, 60c,.
- the size of the square is determined by changing the search area 94 according to the ratio of the sizes of the candidate curves.
- the size of the candidate curve can be determined based on the magnification (extend x , extend y ) among the parameters of each particle.
- the method of determining the search area 94 is not limited to that shown in FIG. As described later, the time may be increased or decreased in consideration of the time for copying the image data of the area to the local memories 60a, 60b, 60c,.
- the center of gravity of the search area 94 may not have knots.
- the movement of the tracking target may be estimated using an autoregressive prediction model used when the particles are transitioned by the motion model, and the search region 94 may be provided widely in the direction estimated to move.
- the search area 94 may not be a square, and may be other rectangles, rhombuses, horizontal and vertical lines of pixels, etc., depending on the search method and the characteristics of movement of the tracking target.
- the contour image cutout unit 50 puts a contour search processing request including the coordinates of the knot 92 and the information of the search area 94 corresponding thereto, for example, the coordinates of one corner of the square and the length of one side into the contour search task queue 52.
- the Nth processing unit 58n of the contour search unit read one contour search processing request from the contour search task queue 52, Based on the square information included in the request, only the image data of the square area is copied from the contour image 90 stored in the image storage unit 24 to the subordinate local memory.
- the amount of data required for one processing unit is limited by setting the processing unit to each knot.
- dividing the processing for each knot significantly reduces the amount of data in the search area and affects the size of the candidate curve. It is hard to receive.
- storage in the local memory is possible regardless of the size of the tracking target.
- the local memory generally has a small capacity and can be accessed at high speed. Therefore, by defining a search area for each knot and copying only the image data in that area, high-speed tracking processing can be performed along with the effect of parallel processing. This effect can be obtained by any information processing apparatus having a plurality of processors. In particular, such a configuration facilitates application to an information processing apparatus that implements “heterogeneous multi-core”.
- Heterogeneous multi-core is an information processing device that mounts different types of cores, and has the characteristics that the memory capacity used by each core is small and that data necessary for processing needs to be copied to the memory of each core.
- the search area 94 is defined as described above, the size of the image data can be made smaller than the memory capacity of each core.
- the present embodiment can be applied to the heterogeneous multi-core, and high-speed tracking processing can be realized without limiting the apparatus.
- any of the plurality of processor units that realize the functions of the first processing unit 58a to the Nth processing unit 58n included in the contour search unit 56 may also serve as the contour image cutout unit 50 and the likelihood acquisition unit 54.
- each function other than the observation unit 30 included in the tracking device 14 may be realized by any one of the plurality of processor units.
- FIG. 8 schematically shows a process transition when the first processing unit 58a, the second processing unit 58b,..., And the Nth processing unit 58n of the contour searching unit 56 perform the contour searching process.
- the right direction of the figure is the time axis, and from time T1, the N processing units of the first processing unit 58a to the N-th processing unit 58n perform knot 1, knot 2, ..., knot N of a certain candidate curve.
- Execute search processing request When the search processing request is read from the contour search task queue 52, each processing unit copies the image data of the area specified by the search processing request from the contour image in the image storage unit 24 to the subordinate local memory and starts the search processing. To do.
- the time required for copying and the time required for searching are each represented by a rectangle.
- pipeline processing is performed by starting a copy of the area specified by the search processing request read from the contour search task queue 52 next time.
- Reduce processing time In the example of the figure, at time T1, the first processing unit 58a corresponds to the knot 1, the second processing unit 58b corresponds to the knot 2,..., The Nth processing unit 58n corresponds to the knot N.
- the copying of the image data of the area to be started is started.
- the search for the contour line in the copied area is started.
- the first processing unit 58a, the second processing unit 58b,..., The N-th processing unit 58n are knots N + 1, knots N + 2,.
- the search area is copied by pipeline processing.
- FIG. 8 shows a case where the time required for copying and the time required for search processing are almost the same, but the present embodiment is not limited to this. That is, copying of the next area to be processed is started at any timing in the time zone during which the contour search process is performed for a certain area, and when the previous search is completed and the copy is also completed, the search process is performed for the area. Any other embodiment may be used. However, as shown in FIG. 8, when the copy time and the search processing time are substantially equal, it is possible to absorb the overhead that the search processing is not started due to the completion of the copy. When copying the entire contour image and continuously performing contour search for all knots, it takes time to copy image data of a large size first. As described above, the processing time can be shortened.
- the size of the search area may be adjusted so that the time required for copying and the time required for search processing are approximately the same. For example, before the actual tracking process, an experiment is performed in a search area of various sizes using a test image having a similar image configuration, the number of cycles required for each process is measured, and the area is cut out so that they are approximately equal The size may be determined. At this time, the contour image cutout unit 50 controls the contour search unit 56 to actually perform the search process, and acquires the time required for the copy and search process, thereby determining the most efficient size of the search region. Feedback control may be performed.
- the contour search for one knot needs to be shortened as the number of tracking objects increases. Even in such a case, it is possible to shorten the copy and search processing time by adjusting the size of the region by experimenting prior to the actual tracking processing and reducing the size.
- a table in which the optimum size of the search area is set for various numbers of tracking targets is prepared in advance, and after starting tracking, the contour image cutout unit 50 refers to the table when the number of tracking targets is acquired.
- the size of the search area may be determined.
- a table that can determine the size of the search area from parameters that determine the ease of tracking, such as the shape of the candidate curve, the complexity of the operation, and the contrast of the image, and various factors such as the required tracking accuracy May be prepared.
- FIG. 9 shows a detailed configuration of the contour image generation unit 22.
- the contour image generation unit 22 includes a smoothing filter 62, a gradation reduction unit 64, and an edge extraction filter 66.
- the contour image generation unit 22 reads the image frame of the input image stored in the image storage unit 24, smoothes it with the smoothing filter 62, lowers the gradation with the gradation reduction unit 64, and extracts the edge with the edge extraction filter 66. Are applied in this order to generate a contour image.
- FIG. 10 is a diagram for explaining a difference between a general edge extraction process and a contour image generation process in the present embodiment.
- the horizontal axis of the graph in the figure all represents the position of the image, that is, the arrangement of pixels, and the range is common.
- the uppermost graph represents the distribution of luminance values of the original image that is the input image.
- the figure shows a pixel row in which an image to be tracked exists in a region 110 indicated by an arrow, and a pixel A and a pixel B have a contour to be tracked.
- the change in the luminance value in the vicinity of the pixel A is gentler with respect to the position than the change in the luminance value in the vicinity of the pixel B.
- Such a situation can frequently occur when the tracking target and the background color are similar in the vicinity of the pixel A, or only the pixel A side is in the shade.
- the luminance value is not constant because the color is changed or the shadow is formed in the region 110 of the image to be tracked, and the luminance value varies like the unevenness 112.
- the edge When generating an edge image of an original image showing such a luminance value distribution, the edge is generally extracted by filtering with an edge extraction filter such as a Laplacian filter. At this time, a threshold value is set for the magnitude of the change of the luminance value with respect to the image plane, and a portion where a change exceeding the threshold value is seen is extracted as an edge.
- the second row of FIG. 10 shows a graph when the edge image is generated as described above. That is, the magnitude of the amount of change in luminance value as shown in the graph is calculated as an edge value, and a portion having an edge value exceeding a preset threshold value 116, that is, a pixel in the vicinity of the pixel B is extracted as an edge. .
- the vicinity of pixel A which is the other contour, is not extracted as an edge because its edge value is smaller than the threshold value 116.
- the threshold value 116 is set to a small value, but by doing so, a portion having a relatively large edge value due to the unevenness 112 of the luminance value that is not related to the contour. 114 is also extracted as an edge.
- candidate curves are set for the contour of the tracking target, and the likelihood is observed by matching with the actual contour to estimate the position of the tracking target. Therefore, if there is a portion that is not extracted from the contour line, or if many lines other than the contour are extracted, the tracking accuracy naturally decreases.
- the optimum value changes for each image, and depending on the image, it can be said that it is an optimum value in the first place. There can be situations where there is no value.
- the present embodiment it is possible to extract the “outline of an object” rather than “the edge in the image” by roughly capturing the object as a surface from detailed information such as lines and gradations in the image.
- a low gradation image in which the luminance value of the original image is reduced is generated.
- the luminance value of the original image is represented by three gradations for easy understanding.
- the luminance value of the area 110 where the tracking target exists discontinuously changes from the luminance value of the other area, and becomes information indicating the presence of the tracking target as the area. .
- the contour image generation unit 22 first smoothes the image frame of the input image by the smoothing filter 62.
- the smoothing filter 62 a general smoothing filter such as a Gaussian filter, a median filter, a simple averaging filter, or a low-pass filter can be used. This removes excess high-frequency components and makes it easier to grasp the surface of the object as a region.
- the gradation-lowering unit 64 generates a gradation-reduced image as described above.
- the gradation reduction unit 64 can be realized by a general bit shift operation, and divides luminance values at predetermined boundaries, and converts the luminance values in each division into one luminance value.
- the luminance value may be equally divided from the bottom, or a luminance value that makes the number of pixels uniform when the color histogram of the image frame is created and divided may be used as the boundary.
- a general posterization method may be used.
- the number of gradations of the low gradation image can be about 8 to 32 gradations, for example.
- the number of gradations of the low gradation image may be reset depending on the tracking target, background, image content, type, and the like.
- a low gradation image having various gradations is generated by using a test image similar to that at the time of actual tracking, and a gradation that can generate a contour image with the highest accuracy or that does not fail tracking is obtained.
- an edge extraction filter 66 is applied to the low gradation image obtained by the gradation reduction unit 64 to generate a contour image.
- a general edge extraction filter such as a Laplacian filter, a Sobel filter, or a Canny edge filter can be used. Thereby, a binary image having different values in the contour portion and other portions is obtained as the contour image.
- FIG. 11 shows an example of an original image to be processed.
- 12 and 13 show the result of generating an edge image for the original image by a general edge image generation method.
- FIG. 12 shows a case where the threshold is low, and
- FIG. 13 shows a case where the threshold is high. It is an edge image.
- FIG. 12 in addition to the contour of the person who is the subject, many edges such as clothes patterns, wrinkles and facial parts are extracted, making it difficult to distinguish from the contour. Also, almost no edge is extracted from the shadowed portion on the left side of the person's head.
- the threshold value is increased, as shown in FIG. 13, the number of extracted edges decreases, and the outline is hardly extracted.
- FIG. 14 shows a low gradation image obtained by processing the original image shown in FIG. 11 by the smoothing filter 62 and the gradation reduction part 64 of the contour image generation unit 22 in the present embodiment.
- the low gradation image detailed information such as a clothing pattern as seen in the edge image of FIG. 12 is omitted, and the image is obtained by capturing the surface of a person or an object as an area.
- FIG. 15 shows a contour image generated by performing edge extraction processing on the low gradation image by the edge extraction filter 66.
- the outline of the person or object is represented by a continuous line, and in FIG. 12, the outline is also extracted from the left side of the person's head where no edge is extracted.
- the main purpose of this embodiment is to track the movement of a person or object in an image, priority is given to the presence of a contour line and information about the approximate position over detailed information of the image at the pixel level. By taking it out, it is possible to correctly detect the movement without mistaking or losing the tracking target.
- the gradation reduction process has the meaning of roughly dividing the image into regions based on the luminance value, and the boundaries of the regions generated thereby are regarded as contours, so that the lines are not interrupted in the middle and the search is easy. .
- a general edge image often appears with a certain width for pixels whose edge value exceeds a threshold value. This is because, as shown in the second stage of FIG. 10, the normal edge value changes approximately continuously with multiple gradations and reaches the peak with a certain width. Further, the lower the threshold value 116 is set so as to extract the edge reliably, the wider the extracted edge becomes.
- the gradation changes greatly even in adjacent pixels, and the edge value peaks in one pixel. Therefore, the extracted portion is a pixel unit, and the extraction result is linear. become. If the edge has a width, processing for thinning the edge is required to match the candidate curve. However, in the case of the contour line in this embodiment, this processing is not necessary, and tracking is performed at a high speed with a simpler configuration. Processing can be performed.
- the contour search process is divided for each knot. Assign to multiple processors and process in parallel. Since the contour search process is independent for each knot, allocation to processors and integration of results are easy. In addition, since the image data necessary for the contour search process for one knot is a limited area near the knot, the data size is small, and each processor copies the data to the local memory and performs the contour search process. Can be executed. Therefore, access to the contour image data can be performed at high speed, and the processing speed can be further improved.
- the number of processors required varies depending on the number of tracking targets, so if the number of tracking targets is smaller than the number of processors, the processing capacity of the device will be used up sufficiently. Disappear. Conversely, if the number of tracking targets is larger than the number of processors, some of the processing that could not be allocated will be executed later, eventually resulting in the possibility of surplus processing capacity.
- the image data size required for each search process varies greatly depending on the size of the tracking target, a memory capacity shortage or a data transfer time fluctuation may occur. Thus, if the processing time and the required memory size vary greatly depending on the contents of the input image, it may become an obstacle to determining the specifications of the apparatus, and the versatility will be poor.
- the search processing is divided for each knot to make the granularity of processing units finer and the number of processing units is increased, so that even if the tracking conditions such as the number of tracking people and the size of the tracking target change, There is little influence on processing time and required memory size, and it is easy to estimate those parameters. Therefore, the specification of the apparatus can be easily determined, and the tracking process can be performed in a suitable environment regardless of the contents of the input image. Similarly, parallel processing is possible with the same processing procedure regardless of the number of processors, and processing units of several thousand orders or more are generated per frame, so that they can be easily allocated to each processor. Since the processing unit is completed in a relatively short time, the scalability is high.
- the boundary between the surfaces can be extracted as a contour line.
- many extra edges other than the contour line are extracted depending on the threshold value for extraction, or the contour line is interrupted, which affects the tracking accuracy.
- the boundary between the surfaces is emphasized as described above, the allowable range of the threshold is wide, and the contour line can be easily extracted.
- the smoothing filter, low gradation part, and edge extraction filter used for contour image generation can all be processed by raster scan, and are independent processing for each line. Is possible. Further, since the contour image generated in the present embodiment appears in a line shape having a width corresponding to one pixel, it is not necessary to perform a thinning process for likelihood observation.
- Embodiment 2 In order for the visual tracking system 10 to accurately perform the visual tracking using the particle filter described above, it is important that the contour line to be tracked is appropriately obtained as an edge image. In addition, it is desirable that the initial arrangement of candidate curves and the motion model do not deviate from the actual tracking target position and movement. However, the optimum edge extraction condition changes depending on the brightness of the input image, and the position and movement of the tracking target may change greatly if the tracking target is different. Therefore, the visual tracking system 10 according to the present embodiment uses an actual input image to adjust edge extraction conditions, initial arrangement of candidate curves, internal parameters of the motion model, and the like, and tracks in an optimal state under any environment. Enable processing.
- FIG. 16 shows the configuration of the tracking device 14 in the present embodiment in detail.
- the tracking device 14 includes an image acquisition unit 20 that acquires input image data input from the imaging device 12, an image storage unit 24 that stores the input image data, and a tracking environment setting unit that detects a tracking target and adjusts various parameters. 124, an image processing unit 120 that generates an edge image from an input image, a tracking processing unit 132 that performs tracking processing using a particle filter, a result storage unit 36 that stores data of final tracking results, and a display device 16 for tracking results
- the output control part 40 which controls the output to is included.
- each element described as a functional block for performing various processes can be configured with a CPU, a memory, and other LSIs in terms of hardware, and a program for performing image processing in terms of software. It is realized by. Therefore, it is understood by those skilled in the art that these functional blocks can be realized in various forms by hardware only, software only, or a combination thereof, and is not limited to any one.
- the tracking environment setting unit 124 includes a tracking target area detection unit 126, an initial contour estimation unit 128, and an initial value determination unit 130.
- the tracking environment setting unit 124 detects a tracking target included in the input image and uses the input image to perform tracking processing.
- the unit 132 adjusts various parameters necessary for the tracking process.
- the parameters adjusted by the tracking environment setting unit 124 are those whose setting values affect the accuracy and efficiency of tracking, and the optimum values vary depending on the shooting environment of the input image and the tracking target.
- an edge extraction parameter which is an edge extraction condition for generating an edge image, an initial value range of the shape space vector x, an internal parameter of the motion model, and the like correspond to this.
- environment-dependent parameters such parameters are referred to as “environment-dependent parameters”.
- the edge extraction parameter is a parameter that changes the density of the edge in the edge image. For example, when a Gaussian filter is used in the image smoothing process before edge extraction, the size and standard deviation of the kernel correspond to it. It is desirable that the edge image is close to the state where only the contour line to be tracked appears continuously. If there are too few lines that appear as edges, there is a high possibility that the edge search will fail. Likelihood reliability decreases due to matching with a line.
- the initial value range of the shape space vector x is a range in which particles to be transitioned when the probability density distribution is first obtained are arranged in the parameter space. If the candidate contour represented by the particle is completely different from the position and size of the tracking target, the likelihood of that particle is naturally observed to be low, but such particles are not generated from the beginning, but only in the vicinity of the tracking target. By arranging, tracking processing with high accuracy can be performed from the tracking start time.
- the optimal values of the internal parameters of the motion model vary depending on the distance the tracking target moves between frames. For example, when a Gaussian motion model is applied, this is the standard deviation of the Gaussian distribution. By reducing the standard deviation when the tracking target hardly moves, and increasing the standard deviation when moving largely, it is possible to transition particles only to a highly probable range while covering the movement of the tracking target, Particles can be used efficiently.
- the environment-dependent parameters are not limited to the parameters described above, and may be parameters used for edge image generation and tracking processing, and may be appropriately determined by an edge extraction method, a motion model, or the like.
- the final number of gradations may be used, and when the observed likelihood is corrected by some reference, the reference may be set as an environment-dependent parameter.
- the gradation reduction and likelihood correction will be described later.
- the tracking target area detection unit 126 performs image analysis for each image frame of the input image data stored in the image storage unit 24 to detect a tracking target area, and determines whether to start or end tracking. For example, a foreground extractor (not shown) using a background difference is mounted, and the presence or absence of a tracking target is determined from the foreground shape extracted from the image frame, and the region is detected. At this time, if the tracking target is a human head, face detection technology may be further applied. Alternatively, an area having a color different from the background color or a specific color may be detected as a tracking target by the color detector. Alternatively, the region to be tracked may be detected by pattern matching with a preset object shape.
- the visual tracking system 10 is provided with a temperature sensor that measures the heat distribution of the space to be imaged and a piezoelectric sensor that acquires the contact area of the tracking object two-dimensionally. An area to be tracked may be detected.
- Existing technology can be applied to the detection of an object by a temperature sensor or a piezoelectric sensor.
- the tracking target region detection unit 126 determines the start of tracking and acquires predetermined information (hereinafter referred to as a feature amount) representing the feature of the detected region. For example, when a tracking target region is detected by the foreground extractor, since the contour line of the tracking target is acquired, the centroid position of the region, the range in the image in which the region exists, and the inclination of the region are feature quantities. Get as.
- the feature amount may be a color histogram, texture, temperature distribution, or the like of the region, and may be determined as appropriate according to the detection method of the region to be tracked. The feature value is used to adjust environment dependent parameters.
- the “end” of tracking determined by the tracking target area detection unit 126 may include a suspension of tracking due to occlusion or the like. Tracking is determined to start when the tracking target appears within the viewing angle of the imaging device or when it appears from behind the object, etc., and when the tracking target leaves the viewing angle of the imaging device or enters the shadow. In some cases, it is determined to end.
- the initial contour estimation unit 128 temporarily sets each environment-dependent parameter, and then estimates the contour of the tracking target using the particle filter by the same processing procedure as that of the normal visual tracking method described above. Specifically, an edge image of the target input image is generated using the temporarily set edge extraction parameters, and the particles are evenly arranged within the initial value range of the temporarily set shape space vector x. Then, the arranged particles are transited by a motion model to which the internal parameters of the motion model set temporarily are applied, for example, a Gaussian motion model, and the likelihood is observed using the generated edge image. A probability density distribution is calculated based on the result, and the contour is estimated.
- a motion model to which the internal parameters of the motion model set temporarily are applied, for example, a Gaussian motion model
- the initial value determination unit 130 compares the feature amount obtained from the tracking target contour estimated by the initial contour estimation unit 128 with the feature amount of the tracking target region detected by the tracking target region detection unit 126, and Evaluate the degree. If the degree of coincidence does not satisfy a predetermined condition, the initial contour estimation unit 128 is requested to reset parameters and estimate the contour of the tracking target.
- the feature value compared by the initial value determination unit 130 is the feature value of the region of the same part in the region to be tracked.
- the shape of the head is represented by an ⁇ -shaped curve 154, and thus the contour estimated by the initial contour estimation unit 128 is the head contour. It becomes. Therefore, the tracking target area detection unit 126 acquires the feature amount of the head portion of the tracking target area, that is, the area formed by the body of the person 152.
- the positional relationship between the head and other parts may be defined in advance, and the region of the specific part may be derived from the estimated contour. For example, by estimating the contour of the head and deriving the upper body region based thereon, the estimated upper body feature amount may be compared with the actual upper body feature amount detected by the tracking target region detection unit 126.
- the method for calculating the degree of matching of feature quantities naturally varies depending on the feature quantities. For example, if the feature amount is the center of gravity of the region, the distance between the centers of gravity, if the region is the range of the region, the position and size of the rectangle that circumscribes the region, if the slope is the angle difference, if the color histogram A scale corresponding to the feature amount such as histogram intersection is set in advance.
- the degree of matching of each feature quantity is comprehensively evaluated according to predetermined rules, such as assigning points or weighting.
- the initial value determination unit 130 Uses the value of the environment-dependent parameter at that time as the final set value.
- the edge extraction parameters are transmitted to the image processing unit 120.
- Parameters necessary for particle transition and observation such as internal parameters of the motion model and likelihood correction criteria, are transmitted to the tracking processing unit 132.
- the probability density distribution calculated when the initial contour estimation unit 128 estimates the contour is also transmitted to the tracking processing unit 132 so as to be used for generation and disappearance of particles at the next time.
- the environment-dependent parameter is set when the tracking target area detection unit 126 newly detects the tracking target.
- the feature quantity of the tracking target may be compared with the estimated feature quantity inside the contour.
- the environment-dependent parameter may be reset by the same method.
- the result obtained by the tracking processing unit 132 can be used as the feature amount based on the estimated contour. If the setting values of environment-dependent parameters are updated as needed, the brightness of the image changes not only at the start of tracking, but also due to changes in the situation such as the tracking target moving from place to place, weather changes, curtains open, etc. Even so, the tracking accuracy can be maintained.
- the image processing unit 120 generates an edge image from the input image. Specifically, edge extraction processing is performed for each image frame of the input image data stored in the image storage unit 24 using the edge extraction parameter acquired from the tracking environment setting unit 124.
- edge extraction processing is performed for each image frame of the input image data stored in the image storage unit 24 using the edge extraction parameter acquired from the tracking environment setting unit 124.
- one of general smoothing filters such as a Gaussian filter, median filter, simple averaging filter, and low-pass filter
- one of general edge extraction filters such as a Laplacian filter, Sobel filter, and Canny edge filter may be used. it can.
- the image processing unit 120 may acquire foreground data from the tracking target area detection unit 126 and perform edge extraction processing only on the foreground to efficiently extract the tracking target outline.
- the edge image in the present embodiment is generated to obtain the contour line to be tracked.
- the image processing unit 120 has the same configuration as that of the contour image generation unit 22 of the first embodiment, and in addition to the smoothing filter and the edge extraction filter, the image is reduced in gradation. Processing may be performed. For example, after smoothing the input image with a smoothing filter, the gradation is reduced to about 8 to 32 gradations, and then an edge extraction filter is applied.
- the luminance value of the tracking target area changes discontinuously from the luminance value of the other areas, so the tracking target area is represented as a plane. be able to. Then, even when the change in luminance value in the contour line is scarce in the original image, the contour line can be easily extracted by the edge extraction filter.
- the gradation reduction can be realized by a general bit shift operation. The luminance value is divided at a predetermined boundary, and the luminance value in each division is converted into one luminance value.
- the luminance values may be equally divided from the bottom, or the luminance values that make the number of pixels uniform when the color histogram of the image frame is created and divided may be used as the boundary.
- a general posterization method may be used.
- the final number of gradations can be included in the environment-dependent parameter as described above.
- the image processing unit 120 may extract the contour of the tracking target by a method other than the general edge extraction method described above.
- a curve constituting the outer periphery of the foreground may be extracted as the contour, or a plurality of methods may be combined.
- the contours extracted by these methods are included in the “edge”. Therefore, the “edge image” generated by the image processing unit 120 is synonymous with the “contour extracted image”.
- the tracking processing unit 132 includes a sampling unit 29, an observation unit 136, and a result acquisition unit 138.
- the function of the sampling unit 29 is the same as that described in the first embodiment.
- the observation unit 136 applies the motion model to all particles as described above to cause the particles to transition in the parameter space and observe the likelihood of the candidate curve determined by each particle.
- the internal parameters and initial probability density distribution of the motion model used here those obtained from the tracking environment setting unit 124 are used.
- the observation unit 136 may correct the obtained likelihood according to a predetermined rule. For example, even if the candidate curve has a shape far from the actual tracking target, there is a chance that a high likelihood is observed just because an edge exists in the vicinity. Likelihood correction is performed so that such likelihood does not adversely affect the calculation of the probability density distribution. For example, an allowable range is set for the size and aspect ratio of the estimated contour, and when the likelihood of the estimated contour that deviates from it exceeds a predetermined threshold value, the likelihood is reduced by a predetermined ratio. When such correction is performed, an allowable range to be set can be included in the environment-dependent parameter.
- the result acquisition unit 138 has the same function as the result acquisition unit 38 in the first embodiment.
- the result acquisition unit 38 further transmits estimated contour data to be tracked to the tracking environment setting unit 124.
- the result storage unit 36 stores, for example, moving image data including the tracking result.
- the result storage unit 36 stores, for example, moving image data including the tracking result.
- the display device 16 under the control of the output control unit 40, it is possible to display how the curve of the template moves in the same manner as the movement of the tracking target.
- processing such as output to another arithmetic module may be appropriately performed according to the purpose of tracking.
- the imaging device 12 captures a place to be captured at a predetermined frame rate.
- the captured image is input as input image data to the image acquisition unit 20 of the tracking device 14 and stored in the image storage unit 24. In such a state, the following tracking process is performed.
- FIG. 17 is a flowchart showing the procedure of the tracking process in the present embodiment.
- the tracking target is a person
- an ⁇ -type template is prepared for the tracking device 14 as described above.
- the template expression method is not limited to the B-spline curve, but may be any description format that can express a desired curve.
- the template shape deformation method may be appropriately selected from methods that can be adapted to the description format and that can be flexibly deformed as described above by changing several types of parameters.
- the tracking target area detection unit 126 of the tracking environment setting unit 124 reads the input image data stored in the image storage unit 24 for each frame, and determines whether to start tracking (S120, S122). For example, when an object having a predetermined size and shape that can be estimated as a person appears as a foreground extracted from an image frame, it is determined to start tracking.
- the size and shape of the foreground used as a judgment criterion are determined in advance logically or experimentally.
- a person may be detected by a known method such as color detection, pattern matching, face detection, heat distribution detection, or contact area detection as described above.
- Steps S120 and S122 are repeated until it is determined that the tracking is started. If it is determined that the tracking is started (Y in S122), the tracking environment setting unit 124 adjusts the environment-dependent parameter to obtain the optimum value, and the image processing unit 120 or Settings are made for the tracking processing unit 132 (S124). A processing procedure for setting environment-dependent parameters will be described later with reference to FIG.
- the sampling unit 29 generates particles using the probability density distribution p (x 0 i ) acquired when the tracking environment setting unit 124 determines the environment-dependent parameter (S128).
- the observation unit 136 causes each particle to transition based on a predetermined motion model, and observes the likelihood p (y t
- a value determined by the tracking environment setting unit 124 is used as an internal parameter that defines the motion model.
- the edge search may use a method generally used in a condensation algorithm or the like.
- desired tracking result data is generated and stored in the result storage unit, such as generating image frame data superimposed on the original input image frame (S132).
- the tracking target area detection unit 126 determines whether to continue or end the tracking process (S134). For example, the tracking end is determined when a target having a predetermined size and shape that can be estimated as a person does not appear as a foreground for a predetermined time. At this time, the tracking target area detection unit 126 is provided with a timer (not shown) to measure the elapsed time from the time when the foreground disappears. Then, the end of tracking is determined at the determination timing immediately after the flag is set, for example, by setting a flag when a predetermined time determined by an experiment or the like has elapsed.
- particles are generated or disappeared (S128).
- the processing from S128 to S132 is repeated for each frame until the tracking target area detection unit 126 determines the end of tracking (N in S134).
- moving image data in which the ⁇ -shaped curve changes with time with the same movement and shape as the head of the visitor to be tracked is stored in the result storage unit 36.
- the output control unit 40 outputs the data to the display device 16 or a module that provides another function, the user can use the tracking result in a desired form.
- FIG. 18 is a flowchart showing a processing procedure for setting environment-dependent parameters in S124 of FIG.
- the tracking target area detection unit 126 acquires feature amounts such as the center of gravity, size, and inclination of the tracking target area detected by itself (S140).
- the feature amount may be a color histogram, texture, temperature distribution, or the like of the tracking target region depending on the tracking target detection method or the tracking target.
- the initial contour estimation unit 128 temporarily sets the value of each environment-dependent parameter (S142). The value set at this time may be determined in advance for each environment-dependent parameter as will be described later.
- the initial contour estimation unit 128 generates an edge image from the image determined to start tracking using the temporarily set edge extraction parameter, and generates particles within the range of the initially set shape space vector x. To do. (S144). Then, the particles are transitioned by the motion model defined by the internally set internal parameters, the likelihood density distribution is obtained by observing the likelihood using the edge image generated in S144, and the contour is estimated (S146).
- the initial value determination unit 130 compares the feature amount obtained from the contour estimated by the initial contour estimation unit 128 in S146 with the feature amount of the tracking target region acquired by the tracking target region detection unit 126 in S140. (S148), the degree of coincidence between the two is confirmed (S150). If the degree of coincidence is smaller than the reference value (N in S150), it is determined that the current environment-dependent parameter value is inappropriate, and the initial contour estimation unit 128 temporarily resets the environment-dependent parameter value. A request is made (S142). If the degree of matching is equal to or greater than the reference value (Y in S150), it is determined that the temporarily determined environment-dependent parameter is appropriate, and the value of the environment-dependent parameter is determined as the final value (S152).
- the probability density distribution at the previous time is used to generate and extinguish particles, and transition and observation are performed to estimate the contour at the next time. And it compares with the feature-value of the area
- the accuracy of the estimated contour can be evaluated even with respect to the movement of the tracking target, so that the reliability of the environment-dependent parameter to be set increases and the applied motion model itself is also added to the environment-dependent parameter. Can do.
- FIG. 19 schematically shows how the tracking environment setting unit 124 determines environment-dependent parameters.
- the edge image 160 is generated in S144 using the edge extraction parameter temporarily set in S142 of FIG.
- a region 162 is a range in which candidate curves determined by each particle are arranged when particles are generated in S144.
- the range is, for example, a rectangle including an area obtained by multiplying the tracking target area detected by the tracking target area detection unit 126 by a constant number.
- the shape space vector x is provisionally set in consideration that each candidate curve is positioned within the region 162.
- the contour estimated in S146 of FIG. 18 using the particles thus generated is the ⁇ -type estimated contour 164. If the tracking target area detection unit 126 acquires the center of gravity of the tracking target face area acquired in S140, the intersection of the broken line 170, and the initial outline estimation unit 128 determines that the center of gravity of the face area based on the estimated contour 164 determined in S146 is the intersection of the solid line 166. As shown in the figure, it can be determined that the closer the intersection of the two is, the closer the currently calculated probability density distribution is to the true value, that is, the current value of the temporary environment-dependent parameter is appropriate.
- the degree of coincidence does not meet the standard in S150 of FIG. 18, it is necessary to sequentially change the value of the environment-dependent parameter.
- the kernel size of the Gaussian filter is used, provisional setting is performed using values such as 3 ⁇ 3, 5 ⁇ 5, 7 ⁇ 7, and 9 ⁇ 9.
- the standard deviation of the kernel is 0.01 to 10.0, and the color information at the time of gradation reduction processing is temporarily set at one of gradations of 1 to 7 bits if the original image is 8-bit color.
- two or more environment-dependent parameters may be changed simultaneously. For example, a table describing a set of kernel standard deviation and size is prepared, and at the time of adjustment, both values are changed according to the table.
- FIG. 20 is a diagram for explaining the setting order when the environment-dependent parameter is temporarily set in the tracking environment setting unit 124.
- this figure shows a case where the standard deviation of the kernel in the Gaussian filter is set.
- the horizontal axis is the standard deviation
- the vertical axis is the edge extraction frequency
- the edge extraction frequency is the ease with which an edge is extracted. For example, when extracting a portion where the rate of change in pixel value is equal to or greater than a threshold value as an edge, the lower the threshold value, Extraction frequency increases.
- the edge extraction frequency changes as indicated by a line 180
- the number of edges detected in the vicinity of the candidate curve determined by each particle, and thus the average value of likelihood changes as indicated by a line 182. That is, when the edge extraction frequency is low and there are few edges in the edge image, there are few edges in the vicinity of the candidate curve, so the average likelihood value is small, and the edge extraction frequency increases as the edge extraction frequency increases. Increases, and the average likelihood value increases.
- the edge image includes more edge lines other than the contour of the object, and the reliability of the likelihood decreases.
- the edge extraction is performed with the minimum extraction frequency at which the outline of the object is extracted as an edge.
- provisional setting is performed in the direction of the arrow at the bottom of the graph, that is, in order from the largest standard deviation to the smallest.
- the edge extraction frequency increases monotonously. Then, before the reliability of the likelihood decreases, the degree of coincidence of the feature amounts exceeds the reference with an appropriate edge extraction frequency, and the optimum value of the standard deviation is obtained.
- the edge extraction frequency increases monotonously by setting the final gradation number in order from the lowest. Even when the edge extraction frequency does not change monotonously when the edge extraction parameter is changed monotonously, an order in which the edge extraction frequency increases monotonously is obtained in advance. The same applies when a plurality of parameters are set simultaneously.
- an edge extraction parameter in visual tracking using a particle filter, using an actual input image, an edge extraction parameter, a range of an initial value of a shape space vector x, an internal parameter of a motion model, etc. Adjust environment dependent parameters.
- first, foreground extraction, face detection, pattern matching, heat distribution, pressure distribution, and the like are applied to the input image to acquire the region of the object.
- the contour of the tracking target is estimated by the particle filter using the edge image generated with the temporarily set edge extraction parameters and other environment-dependent parameters set temporarily.
- the degree of coincidence between the area defined by the estimated contour and the actually detected area to be tracked is estimated by comparing the feature amounts, and whether or not the parameter set temporarily is determined.
- the setting order of the edge extraction parameters is determined in such a direction that the edge extraction frequency monotonously increases.
- the optimum value of the parameter can be acquired before the edge is excessively extracted and the reliability of the likelihood is impaired, and high-accuracy tracking can be realized without imposing a large calculation load.
- the contour image generation unit 22 includes a smoothing filter 62, a gradation reduction unit 64, and an edge extraction filter 66 as shown in FIG. 9, and generates a contour image by these processes.
- the contour image generation unit 22 may be a general edge extractor.
- the contour image generation unit 22 may include only the smoothing filter 62 and the edge extraction filter 66.
- the surface or background of the tracking target is not complicated, it is possible to generate an outline image with an edge extractor. Also in this case, the likelihood can be observed by performing an edge search using the generated edge image. High-speed tracking processing becomes possible by parallel processing.
- the contour image generation unit 22 may generate a contour image using a foreground extractor (not shown). For example, when the movement of a player is tracked using an image of a soccer game as an input image, the main background is the ground and the player wears a uniform, so each pixel value is limited. In such a case, the outline of the tracking target can be accurately extracted by a general foreground extraction process. A table associating them may be prepared so that the outline image generation unit 22 can determine which processing is performed according to the type of the input image. Or you may enable a user to change a setting.
- the contour image generation unit 22 may be provided in an image processing device other than the tracking device 14.
- an apparatus for automatically taking a photograph may be provided, and the image of the subject may be once captured and the contour image of the subject may be generated by the contour image generation unit 22.
- the position of the subject is calculated from the contour image, and the information is fed back to the camera orientation and position control device to automatically adjust the subject so that it appears in a desired position such as the middle of the photograph.
- the feature amounts are compared while changing each parameter so that the edge extraction frequency monotonously increases, and the degree of matching exceeds the reference value.
- Parameters were determined as optimum values.
- the environment-dependent parameter may be set with all predetermined values, and the parameter value with the best feature value matching rate may be determined as the optimum value.
- tracking processing can be performed with appropriate parameters according to changes in the shooting environment and the object, and as a result, tracking accuracy can be maintained under any environment.
- the present invention can be used for information processing apparatuses such as a visual tracking apparatus, a computer, a game machine, an imaging apparatus, and a moving image reproduction apparatus.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
Description
Contour tracking by stochastic propagation of conditional density, Michael Isard and Andrew Blake, Proc. European Conf. on Computer Vision, vol. 1, pp.343-356, Cambridge UK (1996) CONDENSATION - conditional density propagation for visual tracking, Michael Isard and Andrew Blake, Int. J. Computer Vision, 29, 1, 5-28 (1998) ICondensation: Unifying low-level and high-level tracking in a stochastic framework, Michael Isard and Andrew Blake, Proc 5th European Conf. Computer Vision, 1998
初めに、本実施の形態の特徴および効果を明らかにするために、パーティクルフィルタによる視覚追跡について概説する。図1は人物を追跡対象とした場合の視覚追跡手法を説明するための図である。人物画像150は実写した動画像やコンピュータグラフィックスなどにより生成された動画像の画像ストリームを構成する画像フレームのひとつであり、追跡対象である人物152が写っている。
視覚追跡システム10によって、上述したパーティクルフィルタによる視覚追跡を精度よく行うためには、エッジ画像として追跡対象の輪郭線が適切に得られていることが重要である。また、候補曲線の初期配置や運動モデルが実際の追跡対象の位置や動きから乖離していないことが望ましい。しかしながら最適なエッジ抽出条件は入力画像の明るさによって変化し、追跡対象の位置や動きは追跡対象が異なれば大きく変わることがあり得る。そこで本実施の形態の視覚追跡システム10は、実際の入力画像を用いてエッジ抽出条件や候補曲線の初期配置、運動モデルの内部パラメータなどを調整し、どのような環境下でも最適な状態で追跡処理を行えるようにする。
Claims (18)
- 追跡したい対象物を撮影した動画像を構成する第1の画像フレームおよび第2の画像フレームのうち、第1の画像フレームにおける対象物の推定存在確率分布に基づき、前記第2の画像フレームにおける対象物の候補輪郭を定め、前記第2の画像フレームのエッジ画像とマッチングして前記候補輪郭の尤度を観測し、前記第2の画像フレームにおける対象物の存在確率分布を推定することにより対象物の追跡処理を行う追跡処理部と、
前記第1の画像フレームにおける対象物の領域を所定の分析手法により検出し、当該領域を表す所定の特徴量を取得する追跡対象領域検出部と、
前記追跡処理部が追跡処理に用いるパラメータの少なくともいずれかの値を仮に設定して前記第1の画像フレームにおける対象物の推定存在確率分布を求め、それに基づき対象物の輪郭を推定する初期輪郭推定部と、
前記初期輪郭推定部が推定した輪郭に基づく対象物の領域の前記特徴量と、前記追跡対象領域検出部が取得した前記特徴量とを比較し、比較結果が所定の条件を満たすとき、前記初期輪郭推定部が仮に設定したパラメータの値を適用して前記追跡処理部に追跡処理を開始させる初期値判定部と、
を備えたことを特徴とする対象物追跡装置。 - 前記初期輪郭推定部が値を仮に設定するパラメータは、画像フレームのエッジ画像を生成する際に用いるエッジ抽出パラメータを含むことを特徴とする請求項1に記載の対象物追跡装置。
- 前記追跡処理部が尤度を観測するために前記候補輪郭とマッチングするエッジ画像を生成する画像処理部をさらに備え、
前記画像処理部は、
対象となる画像フレームの階調数を下げた低階調画像を生成する低階調化部と、
前記低階調化部が生成した低階調画像からエッジを抽出して、元の画像における被写体の輪郭線を強調したエッジ画像を生成する輪郭画像生成部と、
を備え、
前記初期輪郭推定部が値を仮に設定するパラメータは、前記低階調化部が生成する低階調画像における階調数を含むことを特徴とする請求項1に記載の対象物追跡装置。 - 前記画像処理部は、
対象となる画像フレームの周波数帯域幅を低くして前記低階調化部へ出力する平滑化フィルタをさらに備えたことを特徴とする請求項3に記載の画像処理装置。 - 前記初期輪郭推定部は、前記初期値判定部において比較結果が所定の条件を満たすまで、パラメータの値を変化させて対象物の輪郭推定を繰り返し、当該パラメータが前記エッジ抽出パラメータであった場合は、エッジ画像生成時のエッジの抽出頻度が増加する方向で当該パラメータの値を変化させることを特徴とする請求項2に記載の対象物追跡装置。
- 前記比較結果が所定の条件を満たすとき、前記初期値判定部は、前記初期輪郭推定部が求めた推定存在確率分布を、前記第1の画像フレームにおける推定存在確率分布として前記追跡処理部に追跡処理を開始させることを特徴とする請求項1に記載の対象物追跡装置。
- 前記追跡処理部は、前記第1の画像フレームにおける対象物の推定存在確率分布に基づき発生させた曲線を所定の運動モデルに従い遷移させることにより前記第2の画像フレームにおける対象物の候補輪郭を定め、
前記初期輪郭推定部が値を仮に設定するパラメータは、前記運動モデルを規定するパラメータを含むことを特徴とする請求項1に記載の対象物追跡装置。 - 前記初期値判定部が比較する特徴量は、対象物の領域の重心、画像内での範囲、傾き、カラーヒストグラム、の少なくともいずれかを含むことを特徴とする請求項1に記載の対象物追跡装置。
- 前記追跡対象領域検出部は、背景差分、顔検出、パターンマッチング、温度分布検出、接触領域検出のいずれかにより対象物の領域を検出することを特徴とする請求項1に記載の対象物追跡装置。
- 前記初期輪郭推定部は、前記追跡対象領域検出部が検出した対象物の領域を含む所定の大きさの領域内に発生させた曲線を、所定の運動モデルに従い遷移させることにより対象物の候補輪郭を定め、尤度を観測して対象物の推定存在確率分布を求めることを特徴とする請求項1に記載の対象物追跡装置。
- コンピュータが、追跡したい対象物を撮影した動画像を構成する第1の画像フレームおよび第2の画像フレームのうち、第1の画像フレームにおける対象物の推定存在確率分布に基づき、前記第2の画像フレームにおける対象物の候補輪郭を定め、前記第2の画像フレームのエッジ画像とマッチングして前記候補輪郭の尤度を観測し、前記第2の画像フレームにおける対象物の存在確率分布を推定することにより対象物の追跡処理を行う方法であって、
動画像を記憶したメモリから前記第1の画像フレームを読み出し、当該画像フレームにおける対象物の領域を所定の分析手法により検出し、当該領域を表す所定の特徴量を取得するステップと、
追跡処理に用いるパラメータの少なくともいずれかの値を仮に設定して前記第1の画像フレームにおける対象物の推定存在確率分布を求め、それに基づき対象物の輪郭を推定するステップと、
前記特徴量を取得するステップにおいて取得した特徴量と、前記輪郭を推定するステップにおいて推定した輪郭に基づく対象物の領域の前記特徴量とを比較し、比較結果が所定の条件を満たすとき、前記仮に設定したパラメータの値を適用して追跡処理を開始するステップと、
を含むことを特徴とする対象物追跡方法。 - コンピュータに、
追跡したい対象物を撮影した動画像を構成する第1の画像フレームおよび第2の画像フレームのうち、第1の画像フレームにおける対象物の推定存在確率分布に基づき、前記第2の画像フレームにおける対象物の候補輪郭を定め、前記第2の画像フレームのエッジ画像とマッチングして前記候補輪郭の尤度を観測し、前記第2の画像フレームにおける対象物の存在確率分布を推定することにより、対象物の追跡処理を実現させるコンピュータプログラムであって、
動画像を記憶したメモリから前記第1の画像フレームを読み出し、当該画像フレームにおける対象物の領域を所定の分析手法により検出し、当該領域を表す所定の特徴量を取得する機能と、
追跡処理に用いるパラメータの少なくともいずれかの値を仮に設定して前記第1の画像フレームにおける対象物の推定存在確率分布を求め、それに基づき対象物の輪郭を推定する機能と、
前記特徴量を取得する機能が取得した特徴量と、前記輪郭を推定する機能が推定した輪郭に基づく対象物の領域の前記特徴量とを比較し、比較結果が所定の条件を満たすとき、前記仮に設定したパラメータの値を適用して追跡処理を開始する機能と、
をコンピュータに実現させることを特徴とするコンピュータプログラム。 - 画像の階調数を下げた低階調画像を生成する低階調化部と、
前記低階調化部が生成した低階調画像からエッジを抽出して、元の画像における被写体の輪郭線を強調した輪郭画像を生成する輪郭画像生成部と、
を備えたことを特徴とする画像処理装置。 - 画像データの周波数帯域幅を低くして前記低階調化部へ出力する平滑化フィルタをさらに備えたことを特徴とする請求項13に記載の画像処理装置。
- 追跡したい対象を撮影した動画像データを構成する画像ストリームに含まれる第1の画像フレームおよび第2の画像フレームのうち、前記第1の画像フレームにおける追跡対象物の推定存在確率分布に基づき、前記第2の画像フレームにおける追跡対象物の候補輪郭を決定する候補輪郭決定部と、
前記候補輪郭決定部が決定した候補輪郭と、前記輪郭画像生成部が生成した前記第2の画像フレームの輪郭画像とをマッチングして、前記候補輪郭の尤度を観測する観測部と、
前記観測部が観測した尤度に基づき、前記第2の画像フレームにおける対象物の存在確率分布を推定する追跡結果取得部と、
をさらに備えたことを特徴とする請求項13または14に記載の画像処理装置。 - メモリに保存された画像を読み出し、階調数を下げた低階調画像を生成するステップと、
前記低階調画像からエッジを抽出して、元の画像における被写体の輪郭線を強調した輪郭画像を生成するステップと、
を含むことを特徴とする画像処理方法。 - メモリに保存された画像を読み出し、階調数を下げた低階調画像を生成する機能と、
前記低階調画像からエッジを抽出して、元の画像における被写体の輪郭線を強調した輪郭画像を生成する機能と、
をコンピュータに実現させることを特徴とするコンピュータプログラム。 - メモリに保存された画像を読み出し、階調数を下げた低階調画像を生成する機能と、
前記低階調画像からエッジを抽出して、元の画像における被写体の輪郭線を強調した輪郭画像を生成する機能と、
をコンピュータに実現させることを特徴とするコンピュータプログラムを記録した、コンピュータ読み取り可能な記録媒体。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US12/595,634 US8331619B2 (en) | 2008-03-14 | 2009-01-28 | Image processing apparatus and image processing method |
JP2009527636A JP4756660B2 (ja) | 2008-03-14 | 2009-01-28 | 画像処理装置および画像処理方法 |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2008065199 | 2008-03-14 | ||
JP2008-065199 | 2008-03-14 | ||
JP2008227643 | 2008-09-04 | ||
JP2008-227643 | 2008-09-04 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2009113231A1 true WO2009113231A1 (ja) | 2009-09-17 |
Family
ID=41064905
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/000317 WO2009113231A1 (ja) | 2008-03-14 | 2009-01-28 | 画像処理装置および画像処理方法 |
Country Status (3)
Country | Link |
---|---|
US (1) | US8331619B2 (ja) |
JP (1) | JP4756660B2 (ja) |
WO (1) | WO2009113231A1 (ja) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2010140323A (ja) * | 2008-12-12 | 2010-06-24 | Toshiba Corp | 画像処理装置及び画像処理方法 |
EP2372655A1 (en) * | 2010-02-04 | 2011-10-05 | Sony Corporation | Image processing device and method, and program therefor |
JP2014056535A (ja) * | 2012-09-14 | 2014-03-27 | Oki Electric Ind Co Ltd | 物体追跡装置及び物体追跡方法 |
JP2016081251A (ja) * | 2014-10-15 | 2016-05-16 | 株式会社ソニー・コンピュータエンタテインメント | 画像処理装置および画像処理方法 |
CN109583347A (zh) * | 2018-11-22 | 2019-04-05 | 华南理工大学 | 一种针对移动平台长时间跟踪的方法 |
WO2019116518A1 (ja) * | 2017-12-14 | 2019-06-20 | 株式会社日立製作所 | 物体検出装置及び物体検出方法 |
CN112464948A (zh) * | 2020-11-11 | 2021-03-09 | 常州码库数据科技有限公司 | 一种基于仿生学的自然场景目标轮廓提取方法及系统 |
CN112651256A (zh) * | 2019-10-12 | 2021-04-13 | 大族激光科技产业集团股份有限公司 | 二维码识别方法、装置、计算机设备及存储介质 |
CN116309537A (zh) * | 2023-04-24 | 2023-06-23 | 东莞市京品精密模具有限公司 | 一种极耳模具表面油污的缺陷检测方法 |
CN116612112A (zh) * | 2023-07-17 | 2023-08-18 | 鑫脉(山东)工贸有限公司 | 一种水桶表面缺陷视觉检测方法 |
Families Citing this family (64)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP5507962B2 (ja) * | 2009-11-05 | 2014-05-28 | キヤノン株式会社 | 情報処理装置及びその制御方法、プログラム |
JP5394296B2 (ja) * | 2010-03-25 | 2014-01-22 | 富士フイルム株式会社 | 撮像装置及び画像処理方法 |
US9628722B2 (en) | 2010-03-30 | 2017-04-18 | Personify, Inc. | Systems and methods for embedding a foreground video into a background feed based on a control input |
CN102236899B (zh) * | 2010-05-07 | 2013-12-04 | 株式会社理光 | 物体检测方法和装置 |
US8649592B2 (en) | 2010-08-30 | 2014-02-11 | University Of Illinois At Urbana-Champaign | System for background subtraction with 3D camera |
US9256802B2 (en) * | 2010-11-26 | 2016-02-09 | Nec Corporation | Object or shape information representation method |
US20150117712A1 (en) * | 2011-05-31 | 2015-04-30 | Pointgrab Ltd. | Computer vision based control of a device using machine learning |
CN102750707A (zh) * | 2011-08-29 | 2012-10-24 | 新奥特(北京)视频技术有限公司 | 一种基于感兴趣区域的图像处理方法和装置 |
KR101913336B1 (ko) * | 2011-10-06 | 2018-10-31 | 삼성전자주식회사 | 이동 장치 및 그 제어 방법 |
US9275472B2 (en) * | 2011-11-18 | 2016-03-01 | Disney Enterprises, Inc. | Real-time player detection from a single calibrated camera |
US9213781B1 (en) | 2012-09-19 | 2015-12-15 | Placemeter LLC | System and method for processing image data |
RU2522044C1 (ru) * | 2013-01-09 | 2014-07-10 | Федеральное государственное бюджетное образовательное учреждение высшего профессионального образования "Южно-Российский государственный университет экономики и сервиса" (ФГБОУ ВПО "ЮРГУЭС") | Устройство выделения контуров объектов на текстурированном фоне при обработке цифровых изображений |
JP6074272B2 (ja) * | 2013-01-17 | 2017-02-01 | キヤノン株式会社 | 画像処理装置および画像処理方法 |
JP6312991B2 (ja) * | 2013-06-25 | 2018-04-18 | 株式会社東芝 | 画像出力装置 |
US9158996B2 (en) * | 2013-09-12 | 2015-10-13 | Kabushiki Kaisha Toshiba | Learning image collection apparatus, learning apparatus, and target object detection apparatus |
JP6347589B2 (ja) * | 2013-10-30 | 2018-06-27 | 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 |
NO336680B1 (no) * | 2013-12-04 | 2015-10-19 | Global Maritime As | Fremgangsmåte for estimering av risiko for minst én utilsiktet sluppet last fra minst én kran på en plattform eller et fartøy på undersjøiske rørledninger og annet undersjøisk utstyr, samt anvendelser av fremgangsmåten |
US9414016B2 (en) | 2013-12-31 | 2016-08-09 | Personify, Inc. | System and methods for persona identification using combined probability maps |
CN103678719B (zh) * | 2013-12-31 | 2017-05-31 | 冠捷显示科技(厦门)有限公司 | 一种主动透过环境感测提供反馈意见的推荐系统 |
US9485433B2 (en) | 2013-12-31 | 2016-11-01 | Personify, Inc. | Systems and methods for iterative adjustment of video-capture settings based on identified persona |
US9792491B1 (en) * | 2014-03-19 | 2017-10-17 | Amazon Technologies, Inc. | Approaches for object tracking |
US9542127B2 (en) * | 2014-05-19 | 2017-01-10 | Canon Kabushiki Kaisha | Image processing method and image processing apparatus |
JP2017525064A (ja) | 2014-05-30 | 2017-08-31 | プレイスメーター インコーポレイテッドPlacemeter Inc. | ビデオデータを用いた活動モニタリングのためのシステム及び方法 |
US9501915B1 (en) * | 2014-07-07 | 2016-11-22 | Google Inc. | Systems and methods for analyzing a video stream |
US9170707B1 (en) | 2014-09-30 | 2015-10-27 | Google Inc. | Method and system for generating a smart time-lapse video clip |
US10127783B2 (en) | 2014-07-07 | 2018-11-13 | Google Llc | Method and device for processing motion events |
US9213903B1 (en) | 2014-07-07 | 2015-12-15 | Google Inc. | Method and system for cluster-based video monitoring and event categorization |
US9449229B1 (en) | 2014-07-07 | 2016-09-20 | Google Inc. | Systems and methods for categorizing motion event candidates |
US10140827B2 (en) | 2014-07-07 | 2018-11-27 | Google Llc | Method and system for processing motion event notifications |
USD782495S1 (en) | 2014-10-07 | 2017-03-28 | Google Inc. | Display screen or portion thereof with graphical user interface |
US9977565B2 (en) | 2015-02-09 | 2018-05-22 | Leapfrog Enterprises, Inc. | Interactive educational system with light emitting controller |
US11334751B2 (en) | 2015-04-21 | 2022-05-17 | Placemeter Inc. | Systems and methods for processing video data for activity monitoring |
US10043078B2 (en) | 2015-04-21 | 2018-08-07 | Placemeter LLC | Virtual turnstile system and method |
US9916668B2 (en) | 2015-05-19 | 2018-03-13 | Personify, Inc. | Methods and systems for identifying background in video data using geometric primitives |
US9563962B2 (en) | 2015-05-19 | 2017-02-07 | Personify, Inc. | Methods and systems for assigning pixels distance-cost values using a flood fill technique |
US9361011B1 (en) | 2015-06-14 | 2016-06-07 | Google Inc. | Methods and systems for presenting multiple live video feeds in a user interface |
US9607397B2 (en) | 2015-09-01 | 2017-03-28 | Personify, Inc. | Methods and systems for generating a user-hair-color model |
US11100335B2 (en) | 2016-03-23 | 2021-08-24 | Placemeter, Inc. | Method for queue time estimation |
US10506237B1 (en) | 2016-05-27 | 2019-12-10 | Google Llc | Methods and devices for dynamic adaptation of encoding bitrate for video streaming |
US9883155B2 (en) | 2016-06-14 | 2018-01-30 | Personify, Inc. | Methods and systems for combining foreground video and background video using chromatic matching |
US10380429B2 (en) | 2016-07-11 | 2019-08-13 | Google Llc | Methods and systems for person detection in a video feed |
US10002284B2 (en) * | 2016-08-11 | 2018-06-19 | Ncku Research And Development Foundation | Iterative matching method and system for partial fingerprint verification |
US9881207B1 (en) | 2016-10-25 | 2018-01-30 | Personify, Inc. | Methods and systems for real-time user extraction using deep learning networks |
US10713792B1 (en) * | 2017-01-13 | 2020-07-14 | Amazon Technologies, Inc. | System and apparatus for image processing |
US11783010B2 (en) | 2017-05-30 | 2023-10-10 | Google Llc | Systems and methods of person recognition in video streams |
US10664688B2 (en) | 2017-09-20 | 2020-05-26 | Google Llc | Systems and methods of detecting and responding to a visitor to a smart home environment |
US10878578B2 (en) * | 2017-10-30 | 2020-12-29 | Qualcomm Incorporated | Exclusion zone in video analytics |
CN109819178B (zh) * | 2017-11-21 | 2022-07-08 | 虹软科技股份有限公司 | 一种用于帧处理的方法和装置 |
CN110033476A (zh) * | 2018-01-11 | 2019-07-19 | 富士通株式会社 | 目标速度估计方法、装置和图像处理设备 |
KR102045753B1 (ko) * | 2018-05-02 | 2019-11-18 | 주식회사 한글과컴퓨터 | 이미지의 배경 영역에 대한 투명화 처리를 지원하는 이미지 편집 처리 장치 및 그 동작 방법 |
CN108960046A (zh) * | 2018-05-23 | 2018-12-07 | 北京图森未来科技有限公司 | 一种训练数据采样方法及其装置、计算机服务器 |
JP7143703B2 (ja) * | 2018-09-25 | 2022-09-29 | トヨタ自動車株式会社 | 画像処理装置 |
CN110008802B (zh) | 2018-12-04 | 2023-08-29 | 创新先进技术有限公司 | 从多个脸部中选择目标脸部及脸部识别比对方法、装置 |
CN111325217B (zh) * | 2018-12-14 | 2024-02-06 | 京东科技信息技术有限公司 | 数据处理方法、装置、系统和介质 |
CN111353929A (zh) * | 2018-12-21 | 2020-06-30 | 北京字节跳动网络技术有限公司 | 图像处理方法、装置和电子设备 |
US11055852B2 (en) * | 2019-02-15 | 2021-07-06 | Nokia Technologies Oy | Fast automatic trimap generation and optimization for segmentation refinement |
CN112017203A (zh) * | 2019-05-31 | 2020-12-01 | 广州市百果园信息技术有限公司 | 图像处理方法、视频处理方法、装置、设备及存储介质 |
CN110363170B (zh) * | 2019-07-22 | 2022-02-01 | 北京华捷艾米科技有限公司 | 视频换脸方法和装置 |
JP7414551B2 (ja) * | 2020-01-24 | 2024-01-16 | キヤノン株式会社 | 制御装置および制御方法 |
CN111862154B (zh) * | 2020-07-13 | 2024-03-01 | 中移(杭州)信息技术有限公司 | 机器人视觉跟踪方法、装置、机器人及存储介质 |
US11800056B2 (en) | 2021-02-11 | 2023-10-24 | Logitech Europe S.A. | Smart webcam system |
US11800048B2 (en) | 2021-02-24 | 2023-10-24 | Logitech Europe S.A. | Image generating system with background replacement or modification capabilities |
JP2023003735A (ja) * | 2021-06-24 | 2023-01-17 | キヤノン株式会社 | 視線検出装置、撮像装置、視線検出方法、プログラム、および、記憶媒体 |
CN113706498A (zh) * | 2021-08-25 | 2021-11-26 | 广东奥普特科技股份有限公司 | 一种锂电池极片视觉检测装置 |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0799581A (ja) * | 1992-09-25 | 1995-04-11 | Olympus Optical Co Ltd | 画像処理装置 |
JPH1020998A (ja) * | 1996-06-28 | 1998-01-23 | Osaka Kagaku Gijutsu Center | 位置指示装置 |
JP2002056393A (ja) * | 2000-08-14 | 2002-02-20 | Canon Inc | 画像処理方法及び装置並びに記憶媒体 |
JP2003216951A (ja) * | 2001-12-03 | 2003-07-31 | Microsoft Corp | 複数のキューを使用する複数の個人の自動検出および追跡の方法、システムおよびコンピュータ可読媒体 |
JP2005311725A (ja) * | 2004-04-21 | 2005-11-04 | Nippon Telegr & Teleph Corp <Ntt> | ステレオ画像の隠蔽装置、及びステレオ画像の隠蔽方法、隠蔽画像の復元装置、及び隠蔽画像の復元方法、並びに、そのプログラム及び記憶媒体 |
JP2006285358A (ja) * | 2005-03-31 | 2006-10-19 | Advanced Telecommunication Research Institute International | エッジトラッキング方法及びそのコンピュータプログラム |
JP2007285762A (ja) * | 2006-04-13 | 2007-11-01 | Nippon Telegr & Teleph Corp <Ntt> | 画像変化予測方法および画像変化予測装置 |
JP2007328746A (ja) * | 2006-06-09 | 2007-12-20 | Sony Computer Entertainment Inc | 対象物追跡装置および対象物追跡方法 |
JP2008026974A (ja) * | 2006-07-18 | 2008-02-07 | Mitsubishi Electric Corp | 人物追跡装置 |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5872864A (en) * | 1992-09-25 | 1999-02-16 | Olympus Optical Co., Ltd. | Image processing apparatus for performing adaptive data processing in accordance with kind of image |
US5760922A (en) * | 1993-10-08 | 1998-06-02 | Matsushita Electric Industrial Co., Ltd. | Area recognizing device and gradation level converting device employing area recognizing device |
JPH08329254A (ja) | 1995-03-24 | 1996-12-13 | Matsushita Electric Ind Co Ltd | 輪郭抽出装置 |
JPH09138471A (ja) | 1995-09-13 | 1997-05-27 | Fuji Photo Film Co Ltd | 特定形状領域の抽出方法、特定領域の抽出方法及び複写条件決定方法 |
JP4206036B2 (ja) * | 2003-12-09 | 2009-01-07 | 株式会社ゼンリン | 電子地図データを利用した風景画像の撮像位置の特定 |
JP2006322795A (ja) * | 2005-05-18 | 2006-11-30 | Olympus Corp | 画像処理装置、画像処理方法および画像処理プログラム |
-
2009
- 2009-01-28 JP JP2009527636A patent/JP4756660B2/ja active Active
- 2009-01-28 WO PCT/JP2009/000317 patent/WO2009113231A1/ja active Application Filing
- 2009-01-28 US US12/595,634 patent/US8331619B2/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0799581A (ja) * | 1992-09-25 | 1995-04-11 | Olympus Optical Co Ltd | 画像処理装置 |
JPH1020998A (ja) * | 1996-06-28 | 1998-01-23 | Osaka Kagaku Gijutsu Center | 位置指示装置 |
JP2002056393A (ja) * | 2000-08-14 | 2002-02-20 | Canon Inc | 画像処理方法及び装置並びに記憶媒体 |
JP2003216951A (ja) * | 2001-12-03 | 2003-07-31 | Microsoft Corp | 複数のキューを使用する複数の個人の自動検出および追跡の方法、システムおよびコンピュータ可読媒体 |
JP2005311725A (ja) * | 2004-04-21 | 2005-11-04 | Nippon Telegr & Teleph Corp <Ntt> | ステレオ画像の隠蔽装置、及びステレオ画像の隠蔽方法、隠蔽画像の復元装置、及び隠蔽画像の復元方法、並びに、そのプログラム及び記憶媒体 |
JP2006285358A (ja) * | 2005-03-31 | 2006-10-19 | Advanced Telecommunication Research Institute International | エッジトラッキング方法及びそのコンピュータプログラム |
JP2007285762A (ja) * | 2006-04-13 | 2007-11-01 | Nippon Telegr & Teleph Corp <Ntt> | 画像変化予測方法および画像変化予測装置 |
JP2007328746A (ja) * | 2006-06-09 | 2007-12-20 | Sony Computer Entertainment Inc | 対象物追跡装置および対象物追跡方法 |
JP2008026974A (ja) * | 2006-07-18 | 2008-02-07 | Mitsubishi Electric Corp | 人物追跡装置 |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2010140323A (ja) * | 2008-12-12 | 2010-06-24 | Toshiba Corp | 画像処理装置及び画像処理方法 |
EP2372655A1 (en) * | 2010-02-04 | 2011-10-05 | Sony Corporation | Image processing device and method, and program therefor |
JP2014056535A (ja) * | 2012-09-14 | 2014-03-27 | Oki Electric Ind Co Ltd | 物体追跡装置及び物体追跡方法 |
JP2016081251A (ja) * | 2014-10-15 | 2016-05-16 | 株式会社ソニー・コンピュータエンタテインメント | 画像処理装置および画像処理方法 |
WO2019116518A1 (ja) * | 2017-12-14 | 2019-06-20 | 株式会社日立製作所 | 物体検出装置及び物体検出方法 |
CN109583347A (zh) * | 2018-11-22 | 2019-04-05 | 华南理工大学 | 一种针对移动平台长时间跟踪的方法 |
CN112651256A (zh) * | 2019-10-12 | 2021-04-13 | 大族激光科技产业集团股份有限公司 | 二维码识别方法、装置、计算机设备及存储介质 |
CN112464948A (zh) * | 2020-11-11 | 2021-03-09 | 常州码库数据科技有限公司 | 一种基于仿生学的自然场景目标轮廓提取方法及系统 |
CN116309537A (zh) * | 2023-04-24 | 2023-06-23 | 东莞市京品精密模具有限公司 | 一种极耳模具表面油污的缺陷检测方法 |
CN116612112A (zh) * | 2023-07-17 | 2023-08-18 | 鑫脉(山东)工贸有限公司 | 一种水桶表面缺陷视觉检测方法 |
CN116612112B (zh) * | 2023-07-17 | 2023-09-22 | 鑫脉(山东)工贸有限公司 | 一种水桶表面缺陷视觉检测方法 |
Also Published As
Publication number | Publication date |
---|---|
JPWO2009113231A1 (ja) | 2011-07-21 |
JP4756660B2 (ja) | 2011-08-24 |
US8331619B2 (en) | 2012-12-11 |
US20100128927A1 (en) | 2010-05-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4756660B2 (ja) | 画像処理装置および画像処理方法 | |
JP5213486B2 (ja) | 対象物追跡装置および対象物追跡方法 | |
US9767568B2 (en) | Image processor, image processing method, and computer program | |
JP5520463B2 (ja) | 画像処理装置、対象物追跡装置および画像処理方法 | |
EP1969559B1 (en) | Contour finding in segmentation of video sequences | |
EP1969560B1 (en) | Edge-controlled morphological closing in segmentation of video sequences | |
US8355530B2 (en) | Liveness detection method and apparatus of video image | |
JP4415198B2 (ja) | 画像合成装置及びプログラム | |
JP4597391B2 (ja) | 顔領域検出装置およびその方法並びにコンピュータ読み取り可能な記録媒体 | |
JP2009087090A (ja) | 対象物追跡装置および対象物追跡方法 | |
EP1969561A1 (en) | Segmentation of video sequences | |
JP2016081252A (ja) | 画像処理装置および画像処理方法 | |
JP2005196359A (ja) | 移動物体検出装置、移動物体検出方法及び移動物体検出プログラム | |
JP2004157778A (ja) | 鼻位置の抽出方法、およびコンピュータに当該鼻位置の抽出方法を実行させるためのプログラムならびに鼻位置抽出装置 | |
JP4750758B2 (ja) | 注目領域抽出方法、注目領域抽出装置、コンピュータプログラム、及び、記録媒体 | |
JP3963789B2 (ja) | 眼検出装置、眼検出プログラム、そのプログラムを記録する記録媒体及び眼検出方法 | |
KR20040042500A (ko) | 얼굴 검출방법 및 그 장치 | |
WO2020212794A1 (en) | Augmented reality implementation method | |
US10671881B2 (en) | Image processing system with discriminative control | |
JP2005071125A (ja) | 被写体検出装置、被写体検出方法、被写体データ選定プログラムおよび被写体位置検出プログラム | |
CN118429388B (zh) | 基于图像处理的视觉追踪方法及装置 | |
JP7341712B2 (ja) | 画像処理装置、画像処理方法、撮像装置、およびプログラム | |
Malashin et al. | Restoring a silhouette of the hand in the problem of recognizing gestures by adaptive morphological filtering of a binary image | |
JP2004118718A (ja) | シーン変化点検出方法および装置並びにプログラム | |
Tucholke et al. | Real-Time Reflection Reduction from Glasses in Videoconferences |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
ENP | Entry into the national phase |
Ref document number: 2009527636 Country of ref document: JP Kind code of ref document: A |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 09718732 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 12595634 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 09718732 Country of ref document: EP Kind code of ref document: A1 |