WO2022201987A1 - 画像解析装置、画像解析システム、画像解析方法及びプログラム - Google Patents
画像解析装置、画像解析システム、画像解析方法及びプログラム Download PDFInfo
- Publication number
- WO2022201987A1 WO2022201987A1 PCT/JP2022/006213 JP2022006213W WO2022201987A1 WO 2022201987 A1 WO2022201987 A1 WO 2022201987A1 JP 2022006213 W JP2022006213 W JP 2022006213W WO 2022201987 A1 WO2022201987 A1 WO 2022201987A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- person
- images
- detected
- image
- posture
- Prior art date
Links
- 238000010191 image analysis Methods 0.000 title claims abstract description 58
- 238000003703 image analysis method Methods 0.000 title description 4
- 238000001514 detection method Methods 0.000 claims abstract description 26
- 230000036544 posture Effects 0.000 claims description 106
- 238000003384 imaging method Methods 0.000 claims description 40
- 238000000034 method Methods 0.000 description 39
- 238000010586 diagram Methods 0.000 description 11
- 238000004458 analytical method Methods 0.000 description 10
- 238000010801 machine learning Methods 0.000 description 6
- 238000010606 normalization Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 241000776450 PVC group Species 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
Definitions
- the present invention relates to an image analysis device, an image analysis system, an image analysis method and a program.
- the match determination device described in Patent Document 1 identifies a selected feature amount selected from one or more feature amounts for an analysis target included in an analysis group, and selects a combination of selected feature amounts between different analysis groups. Based on this, it is evaluated whether the analysis targets of the multiple analysis groups match. In addition, when the evaluation indicates that the analysis targets match between the analysis groups, the analysis targets of the different analysis groups are identified as the same target. Note that Non-Patent Document 1 describes a technique related to human skeleton estimation.
- Patent Document 1 when people overlap in part of a plurality of images or people are hidden behind an object such as a pillar, the actual People who are the same object may not be judged to be the same.
- the present invention has been made in view of the above circumstances, and an image analysis apparatus, an image analysis system, an image analysis method, and a program capable of accurately identifying the same person in a plurality of time-series consecutive images. intended to provide
- the image analysis apparatus includes: an image acquiring means for acquiring a plurality of images consecutive in time series; detection means for detecting a person and the posture of the person in each of the plurality of images; and determining means for determining the identity of the person detected in different images using the detected posture of the person.
- the image analysis system comprises one or more imaging means; and the image analysis device described above.
- An image analysis method comprises the computer Acquiring a plurality of consecutive images in time series; detecting a person and a posture of the person in each of the plurality of images; and determining the identity of the person between different images using the detected pose of the person.
- a program comprises to the computer, Acquiring a plurality of consecutive images in time series; detecting a person and a posture of the person in each of the plurality of images; and determining the identity of a person between different images using the detected posture of the person.
- FIG. 1 is a diagram showing a functional configuration example of an image analysis system according to an embodiment of the present invention
- FIG. FIG. 10 is a diagram schematically showing an example of images of each of imaging areas A1 to A2 captured at times T1 to T4
- 1 is a diagram showing a physical configuration example of an image analysis system according to an embodiment
- FIG. 4 is a flowchart showing an example of image analysis processing according to one embodiment of the present invention
- 4 is a flowchart showing an example of image analysis processing according to one embodiment of the present invention
- FIG. 10 is a diagram schematically showing an example of images captured at times T5 to T8 for each of the imaging areas A1 to A2
- FIG. 5 is a flowchart showing the details of the detection process shown in FIG. 4;
- FIG. 5 is a flowchart showing the details of the grouping process shown in FIG. 4;
- FIG. FIG. 10 is a diagram showing superimposed images at times T1 to T6 in which the photographing area A1 is photographed.
- FIG. 10 is a diagram showing an example in which a flow line is generated by grouping regions of people detected in images taken at times T1 to T6 in which an imaging region A1 is photographed;
- FIG. 6 is a flowchart showing the details of the combining process shown in FIG. 5;
- FIG. FIG. 6 is a flowchart showing the details of the combining process shown in FIG. 5;
- FIG. FIG. 11 is a diagram showing an example in which the interrupted flow lines shown in FIG. 10 are combined by a combining process;
- FIG. 10 is a diagram showing an example in which a flow line is generated by grouping areas of people detected in images of times T1 to T8 in which an imaging area A2 is imaged.
- FIG. 10 is a diagram showing an example in which flow lines of different imaging areas A1 and A2 are combined;
- An image analysis system determines the identity of a person between different images based on a plurality of images that are consecutive in time series, and based on the result of the determination, determines the flow line of the person. Perform processing such as asking for
- the image analysis system includes two cameras 101a-101b and an image analysis device 100, as shown in FIG.
- the image analysis device 100 includes an image acquisition unit 102 , a detection unit 103 , a determination unit 104 and an identification image output unit 105 .
- Each of the cameras 101a to 101b is an example of a photographing means that is provided at a station, building, facility, road, etc., and photographs a predetermined photographing area. As shown in FIG. 2, the cameras 101a and 101b generate image information representing a plurality of two-dimensional images that are continuous in time series by photographing fixed photographing areas A1 and A2.
- FIG. 2 shows images of each of the imaging regions A1 to A2 captured at times T1 to T4.
- P_T1 to P_T4 shown in the photographing area A1 schematically show areas of the person P photographed at times T1 to T4, respectively.
- Q_T1 to Q_T4 shown in the photographing area A1 schematically show areas of the person Q photographed at times T1 to T4, respectively.
- R_T1 to R_T4 shown in the photographing area A2 schematically show areas of the person R photographed at times T1 to T4, respectively.
- one or more cameras may be provided in the image analysis system.
- the image acquisition unit 102 acquires a plurality of time-series continuous images obtained by photographing the photographing areas A1 and A2.
- the image acquisition unit 102 acquires image information generated by each of the cameras 101a to 101b from each of the cameras 101a to 101b via a wired network, a wireless network, or a network configured by appropriately combining these. do.
- the detection unit 103 detects a person and the posture of the person in each of the plurality of images acquired by the image acquisition unit 102 .
- the detection unit 103 detects the region and posture of the person in each image based on the image information of each of the plurality of images.
- a known method may be used as a method for detecting the region and posture of a person from an image.
- a person's posture should be detected based on features such as the recognized person's joints, using skeleton estimation technology that uses machine learning.
- OpenPose described in Non-Patent Document 1 can be cited as an example of the skeleton estimation technology.
- the determination unit 104 uses the posture of the person detected by the detection unit 103 to determine the identity of the person detected between different images.
- the determination unit 104 includes a feature amount acquisition unit 106 and a specification unit 107, as shown in FIG.
- the feature quantity acquisition unit 106 uses the posture of the person detected by the detection unit 103 to obtain the posture feature quantity of the person.
- a posture feature amount is a value indicating a feature of a person's posture, and is, for example, a feature amount of a two-dimensional skeletal structure detected by the detection unit 103 .
- the posture feature amount may be the feature amount of the entire skeletal structure, the feature amount of a part of the skeletal structure, or may include a plurality of feature amounts like each part of the skeletal structure.
- the posture feature amount is a feature amount obtained by machine learning the skeletal structure, the size of the skeletal structure from the head to the foot on the image, and the like.
- the size of the skeletal structure is the vertical height and area of the skeletal region including the skeletal structure on the image.
- the vertical direction (height direction or vertical direction) is the vertical direction (Y-axis direction) in the image, for example, the direction perpendicular to the ground (reference plane).
- the left-right direction horizontal direction
- X-axis direction is the left-right direction in the image, for example, the direction parallel to the ground.
- the identification unit 107 identifies the same person detected in the different images based on whether the similarity of the posture feature amount obtained by the feature amount acquisition unit 106 is equal to or greater than a first reference value. do.
- the first reference value is a value predetermined for the degree of similarity of posture feature amounts as a criterion for determining whether or not postures are similar.
- the identifying unit 107 determines that persons detected in different images are the same person when all of the following conditions A to C are satisfied. Further, when at least one of the conditions A to C is not satisfied, the identification unit 107 determines that persons detected in different images are not the same person.
- Condition A The degree of similarity of the posture feature quantity is equal to or greater than the first reference value
- Condition B The same person does not overlap in terms of time
- Condition C Different persons overlap in place things that don't happen
- condition B and condition C may not be included in the conditions for identifying the same person.
- the identification unit 107 includes a grouping unit 108 and a combining unit 109.
- the grouping unit 108 determines whether or not the persons detected in the different images are the same person based on conditions A to C as described above. group images of people in each of the plurality of images so that they belong to . In this grouping process, for example, images having adjacent shooting times may be sequentially selected in chronological order as the “different images”.
- the grouping unit 108 generates a flow line of each person included in the plurality of images by connecting the image areas of the persons belonging to the same group in chronological order.
- This flow line is a line connecting predetermined points such as the center of gravity of the image of the person and the center of the shoulder.
- the grouping unit 108 may determine whether or not the persons included in the different images are the same person based on the conditions A to G, similar to the combining unit 109 described later.
- the connecting unit 109 connects the discontinued flow line.
- the interrupted line of flow is a line of flow that includes an end in the imaging area A1 or A2.
- both ends of many flow lines generally coincide with the boundary of the imaging area A1 or A2.
- a broken line of flow may occur.
- the combining unit 109 determines whether or not the persons included in the image at the end of the flow line, that is, the images before and after the break in the flow line are the same person. connect between the ends of the flow line.
- the combining unit 109 determines whether or not the persons included in the different images are the same person based on the above conditions A to C and the following conditions D to G. In this combining process, it is preferable to select images before and after the line of flow is interrupted as the “images different from each other”.
- Condition D The shooting time interval of the images before and after the line of flow is interrupted is within a predetermined time.
- Condition E The distance of the person detected in the images before and after the line of flow is interrupted is within a predetermined distance. The difference in the orientation of the person detected in the images before and after the break of the line is within a predetermined range. be equal to or higher than the reference value
- the image capturing time interval is the time interval at which the image was captured. Images that are continuous in time series are often captured at approximately constant time intervals, such as N every second (N is an integer equal to or greater than 1). , the number of images. Note that the predetermined time may be defined by, for example, the length of time.
- Whether the distance of the person is the predetermined distance may be determined, for example, based on the distance between the image areas of the person in the image (for example, the number of pixels). It may be determined based on the distance in the real space to be measured.
- An image feature amount is a value that indicates the feature of a person's image area as an image, and is a feature amount generated based on image information.
- the image feature amount may be the feature amount of the entire image of a person, the feature amount of a part of the image, or may include the feature amounts of a plurality of parts such as the face, trunk, and legs. Any method such as machine learning or normalization may be used as the method for calculating the image feature amount, and the minimum value or the maximum value may be obtained as the normalization.
- the image feature amount is the average brightness of each color component, the degree of matching with a color pattern such as a checkered pattern, stripes, or the like.
- the second criterion is a predetermined value for the degree of similarity of image feature amounts as a criterion for determining whether or not images are similar.
- the combining unit 109 determines that persons detected in mutually different images are the same person when all conditions A to G are satisfied. Also, if at least one of the conditions A to G is not satisfied, the combining unit 107 determines that persons detected in different images are not the same person.
- conditions B to G may not be included in the conditions for connecting interrupted flow lines.
- the identification image output unit 105 outputs identification image information based on the determination result of the determination unit 104 .
- the identification image information is information including an image in which a person detected in each of a plurality of images is associated with information for identifying the person (that is, identification information for identifying the same person).
- a method for outputting image information by the identification image output unit 105 is, for example, display and transmission of image information. That is, the identification image output unit 105 may display the image on a display, or may transmit the image to another device connected via a network configured by wire, wireless, or an appropriate combination thereof.
- the image analysis apparatus 100 has a bus 1010, a processor 1020, a memory 1030, a storage device 1040, a network interface 1050, and a user interface 1060, as shown in FIG.
- the bus 1010 is a data transmission path through which the processor 1020, memory 1030, storage device 1040, network interface 1050, and user interface 1060 mutually transmit and receive data.
- the method of connecting processors 1020 and the like to each other is not limited to bus connection.
- the processor 1020 is a processor realized by a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or the like.
- the memory 1030 is a main memory implemented by RAM (Random Access Memory) or the like.
- the storage device 1040 is an auxiliary storage device realized by a HDD (Hard Disk Drive), SSD (Solid State Drive), memory card, ROM (Read Only Memory), or the like.
- HDD Hard Disk Drive
- SSD Solid State Drive
- ROM Read Only Memory
- the storage device 1040 realizes a function of holding various information.
- the storage device 1040 includes functional units of the image analysis apparatus 100 (image acquisition unit 102, detection unit 103, determination unit 104 (feature acquisition unit 106, identification unit 107 (grouping unit 108, combining unit 109)), A program module for realizing the identification image output unit 105) is stored.
- the processor 1020 loads each program module into the memory 1030 and executes it, thereby realizing each functional unit corresponding to the program module.
- the network interface 1050 is an interface for connecting the image analysis device 100 to a network configured by wire, wireless, or a combination thereof.
- Image analysis apparatus 100 according to the present embodiment is connected to a network through network interface 1050, thereby communicating with cameras 101a to 101b and the like.
- the user interface 1070 is an interface for inputting information from the user and an interface for presenting information to the user. including.
- the functions of the image analysis apparatus 100 can be realized by executing a software program in cooperation with each physical component. Therefore, the present invention may be implemented as a software program (simply referred to as "program") or as a non-temporary storage medium in which the program is recorded.
- the image analysis process determines the identity of a person between different images based on a plurality of time-series consecutive images captured by the cameras 101a and 101b, and determines the flow line of the person based on the result of the determination. It is processing such as asking.
- the image analysis process is started, for example, when the user indicates an image to be processed.
- the image to be processed is indicated by, for example, a camera that takes the image and a shooting time including the start time and the end time of shooting.
- a camera that takes the image
- a shooting time including the start time and the end time of shooting.
- the image acquiring unit 102 acquires a plurality of time-series continuous images of each of the photographing areas A1 and A2 photographed by the cameras 101a and 101b (step S101).
- the image acquisition unit 102 acquires image information representing each image shown in FIGS. 2 and 6 from each of the cameras 101a to 101b.
- the image information preferably includes camera identification information for identifying the cameras 101a and 101b that have taken the image, and the time the image was taken.
- FIGS 2 and 6 are diagrams showing examples of images captured by each of the cameras 101a to 101b at the start time T1 to the end time T8, and FIG. 2 is captured at the time T1 to T4 as described above. An image of each of the photographing areas A1 and A2 is shown. FIG. 6 shows images of the imaging areas A1 to A2 captured at times T5 to T8.
- P_T5 to P_T6 shown in the photographing area A1 schematically show areas of the person P photographed at times T5 to T6, respectively.
- Q_T5 to Q_T6 shown in the photographed area A1 schematically show areas of the person Q photographed at times T5 to T6, respectively.
- P_T7 to P_T8 shown in the photographing area A2 schematically show areas of the person P photographed at times T7 to T8, respectively.
- the detection unit 103 and the determination unit 104 repeatedly execute the processing of steps S103 to S110 for each of the imaging areas A1 and A2 acquired in step S101 (step S102; loop A).
- each image at times T1 to T8 when the imaging region A1 is first captured is processed.
- the detection unit 103 and the feature amount acquisition unit 106 repeat the processing of steps S104 to S105 for each of the images that are continuous in time series (step S103; loop B). More specifically, for example, the processing in steps S104 to S105 is repeated for each image at times T1 to T8, with the imaging region to be processed being the imaging region A1.
- the detection unit 103 performs detection processing (step S104).
- FIG. 7 is a flowchart showing details of the detection process (step S104).
- the detection unit 103 detects a human area in each image (step S201). Specifically, for example, in the case of the image at time T1 shown in the upper left image of FIG. 2, the region of the person P indicated by the solid-line ellipse P_T1 and the region of the person Q indicated by the dotted-line ellipse Q_T1 are detected.
- the detection unit 103 obtains an image feature amount for each human region specified in step S201 (step S202). More specifically, for example, based on the image information of the person's area specified in step S201, an image feature amount indicating the image feature of each area is obtained.
- the detection unit 103 detects the posture of the person for each of the areas of the person specified in step S201 (step S203).
- the image of the region of the person specified in step S201 is input, and the state of the skeleton of the person is estimated using a skeleton estimation model that has been trained using machine learning, thereby obtaining the posture of the person.
- the detection unit 103 ends the detection process (step S104) and returns to the image analysis process shown in FIG.
- the feature quantity acquisition unit 106 obtains the posture feature quantity using the human posture detected in step S203 (step S105).
- the feature amount acquisition unit 106 receives the posture of the person detected in step S104 as an input, uses a posture feature amount calculation model that has been trained using machine learning, and obtains the posture feature amount of the person. Output. For example, in the case of the image at time T1 shown in the upper left image in FIG. 2, the posture feature amount of the person P indicated by the solid-line ellipse P_T1 and the posture feature amount of the person Q indicated by the dotted-line ellipse Q_T1 are obtained.
- step S201 may be used together with the person's posture as the input information for obtaining the posture feature amount.
- steps S104 and S105 are repeated for each of the images at times T1 to T8 that are consecutive in time series taken in the imaging area A1 that is the object of processing in loop A (step S102) ( Step S103; Loop B).
- the grouping unit 108 selects time-sequentially consecutive times T1 to T8 when the photographing area A1 targeted for processing in loop A (step S102) is photographed.
- the images of the person detected in each of the images are grouped (step S106).
- FIG. 8 is a flowchart showing details of the grouping process (step S106).
- the grouping unit 108 repeats the processing of steps S302 to S306 (loop C; step S301).
- the combination of time-sequential images to be processed in loop C includes images at times T6 and T5, images at times T5 and T4, images at times T4 and T3, and images at times T3 and T2. Images, images at times T2 and T1.
- a combination of images to be processed may be selected in chronological order. In the following, an example will be described in which a combination of images to be processed is selected from images later in time, that is, from a combination of images at times T6 and T5.
- the grouping unit 108 determines whether or not the degree of similarity between human posture feature amounts detected in different images is equal to or greater than a first reference value (step S302).
- the judgment processing regarding the posture feature amount in this step S302 corresponds to the judgment as to whether or not the condition A described above is satisfied.
- the grouping unit 108 obtains the similarity of the posture feature amounts obtained in step S105 for each combination of persons detected in the images at times T6 and T5.
- FIG. 9 is a diagram showing superimposed images at times T1 to T6 in which the imaging area A1 is photographed.
- the people detected in the images at times T6 and T5 are the people in areas P_T6, P_T5, Q_T6, and Q_T5.
- the degree of similarity is obtained for each of six combinations when two regions are extracted from these four regions.
- the combinations in this example are specifically combinations of regions P_T6 and P_T5, P_T6 and Q_T6, P_T6 and Q_T5, P_T5 and Q_T6, P_T5 and Q_T5, Q_T6 and Q_T5.
- the degree of similarity of posture feature values is, for example, the difference or ratio of posture feature values.
- Grouping section 108 then compares the similarity of the posture feature amount with the first reference value to determine whether the similarity is greater than or equal to the first reference value.
- step S302 If it is determined that the similarity of the posture feature amount is not equal to or greater than the first reference value (step S302; No), the grouping unit 108 determines that the persons related to the posture feature amount for which the similarity is obtained are not the same person. (step S303).
- the posture of a person who is moving does not change significantly in a short period of time, and the change in posture is often within a certain range. Therefore, when the similarity of the posture feature amount of a person detected in different images is less than the first reference value, the person is not the same person, and the similarity of the posture feature amount of the person is less than the first reference value. If so, it can be presumed that the person is the same person.
- the combinations of areas indicating persons who are determined not to be the same person are specifically areas P_T6 and Q_T6, areas P_T6 and Q_T5, areas P_T5 and Q_T6, and areas P_T5 and Q_T5.
- step S302 If it is determined that the similarity of the posture feature amount is equal to or greater than the first reference value (step S302; Yes), the grouping unit 108 divides the same person temporally or different persons geographically. It is determined whether or not it will exist (step S304). The determination processing regarding duplication of the same person in step S304 corresponds to determination of whether or not conditions B and C described above are satisfied.
- the similarity of the posture feature amount is equal to or greater than the first reference value for the combination of the areas P_T6 and P_T5 and the areas Q_T6 and Q_T5. is judged.
- regions P_T6 and P_T5 are included in the images at different times T6 and T5, the same person does not exist temporally.
- regions Q_T6 and Q_T5 are included in images at different times T6 and T5, so the same person does not exist temporally.
- regions P_T6 and P_T5 do not overlap geographically, and the regions Q_T6 and Q_T5 also do not overlap geographically.
- step S304 When it is determined that the same person does not overlap in time and that different people do not overlap in location (step S304; No), the grouping unit 108 It is determined that the person indicated by the area is the same person (step S305).
- step S304 If it is determined that the same person overlaps in time or different people overlap in location (step S304; Yes), the grouping unit 108 divides the region into is not the same person (step S303).
- step S201 it may not be possible to detect that the region Q_T4 is a human region.
- the persons detected in the images at times T5 and T4 are persons in areas P_T5, P_T4, and Q_T5.
- the degree of similarity may be greater than or equal to the first reference value.
- region P_T4 when the degree of similarity between both the human posture feature amounts of regions P_T5 and Q_T5 and the human posture feature value of region P_T4 is greater than or equal to the first reference value. At this time, it is determined that the person in the area P_T4 is the same person as the person P in the area P_T5 and the person Q in the area Q_T5. will exist in duplicate at the location indicated by . In other words, different people will exist in overlapping locations.
- the posture of the person in the region Q_T4 may not be detected as the actual correct posture in step S203.
- the degree of similarity may be greater than or equal to the first reference value.
- the degree of similarity between the human posture feature amount of region P_T5 and both of the human posture feature amounts of regions P_T4 and Q_T4 is equal to or greater than the first reference value.
- the persons in the areas P_T4 and Q_T4 are determined to be the same person as the person P in the area P_T5, so the same person P exists at time T4. That is, the same person will exist temporally.
- the degree of similarity between the human posture feature amount of region Q_T5 and both of the human posture feature amounts of regions P_T4 and Q_T4 may be greater than or equal to the first reference value.
- the persons in the areas P_T4 and Q_T4 are determined to be the same person as the person Q in the area Q_T5, so the same person Q exists at time T4.
- an error occurs in determining the identity of a person, it is not limited to the case where a person's area or posture cannot be detected correctly because the person is hidden behind another person. may not be detected correctly (not shown).
- the grouping unit 108 groups regions of persons who are determined to be the same person (step S306).
- FIG. 10 shows an example of grouping people detected in images taken at times T1 to T6 in which the shooting area A1 was shot.
- the grouping unit 108 divides the regions P_T6 and P_T5 into images of the same person. belongs to a group G1 showing images of Further, the grouping unit 108 assigns the regions Q_T6 and Q_T5 to the group G2 representing images of the same person.
- the grouping unit 108 executes the processing of steps S302 to S306 for each combination of images that are consecutive in time series among the images at times T1 to T6 (loop C; step S301). As a result, as shown in FIG. 10, the persons detected in the images at times T1 to T6 are grouped into groups G1 to G4.
- FIG. 10 shows an example of grouping of persons detected in images taken at times T1 to T6 of the shooting area A1.
- regions P_T6 and P_T5 are grouped into group G1, and regions Q_T6 and Q_T5 into group G2.
- the regions P_T1 to P_T3 are grouped into a group G3, and the regions Q_T1 to Q_T3 are grouped into a group G4.
- the grouping unit 108 ends the grouping process (step S106) and returns to the image analysis process shown in FIG.
- the grouping unit 108 generates flow lines for each group G1 to G4 (step S107).
- flow lines ML_1 to ML_4 are generated by smoothly connecting predetermined points (for example, points corresponding to the centers of shoulders) of regions of people belonging to groups G1 to G4, respectively.
- the directions indicated by the arrows of the flow lines ML_1 to ML_4 are directions of movement along the time series.
- flow line ML_1 connecting regions P_T5 to P_T6 is generated.
- the person indicated by the area belonging to the group G1 is moving substantially rightward while tilting slightly upward in the image.
- the connecting unit 109 determines whether or not the flow lines ML_1 to ML_4 generated in step S107 include a discontinued flow line (step S108).
- the end ML_1S of the flow line ML_1 corresponds to the position of the person indicated by the area P_T5 and is inside the imaging area A1. Therefore, the line of flow ML_1 is a broken line of flow.
- the end ML_2S of the flow line ML_2, the end ML_3E of the flow line ML_3, and the end ML_4E of the flow line ML_4 are all inside the imaging area A1. Therefore, all of the flow lines ML_2 to ML_4 are interrupted flow lines.
- all of the flow lines ML_1 to ML_4 are interrupted flow lines, and the connecting unit 109 determines that the interrupted flow lines are included.
- step S108 When it is determined that the interrupted flow line is not included (step S108; No), the detection unit 103 and the determination unit 104 execute the processing of steps S103 to S110 for the next imaging region (step S102; loop A). .
- step S108 When it is determined that a discontinued flow line is included (step S108; Yes), the combining unit 109 repeatedly executes the merging process (step S110) for the discontinued flow lines ML_1 to ML_4 (step S109; loop D).
- the combining unit 109 determines whether or not the above-described conditions A to G are satisfied for each combination of the interrupted flow lines ML_1 to ML_4. Then, when the conditions A to G are satisfied, the connecting unit 109 merges the groups and connects the ends of the flow lines between the merged groups.
- the ends ML_1S and ML_2S are based on the image at the common time T5
- the ends ML_3E and ML_4E are based on the image at the common time T3. Since it is impossible for the same person to be photographed twice in images taken at a common time, there is no need to perform merging or joining processing for combinations of flow lines having ends based on images taken at a common time.
- the combination of flow lines ML_1 to ML_4 to be processed in loop D is a combination of flow lines having ends ML_1S, ML_2S, ML_3E, and ML_4E included in images at different times. becomes. That is, the combinations of flow lines to be processed in loop D (step S109) are a set of flow lines ML_1 and ML_3, a set of flow lines ML_1 and ML_4, a set of flow lines ML_2 and ML_3, and a set of flow lines ML_2 and ML_3. It is a set of line ML_2 and flow line ML_4.
- 11 and 12 are flowcharts showing the details of the combining process (step S110).
- the combining unit 109 determines whether or not the image capturing time interval before and after the flow lines ML_1 to ML_4 are interrupted is within a predetermined time (step S401).
- the determination processing regarding the interrupted shooting time interval in step S401 corresponds to determination of whether or not the condition D described above is satisfied.
- the flow line of the same person when interrupted, it may be hidden by a fixed object such as a pillar as described above, or by a moving object such as a person.
- the image capturing time interval before and after the discontinuity corresponds to the time it takes for the camera 101a to pass through a fixed object or behind a person. For this reason, it is generally preferable to predetermine a time corresponding to the time it takes for a person to pass through a fixed object or another person's back.
- the number of interrupted images is one in any combination of flow lines ML_1 to ML_4. Therefore, when the predetermined time is determined to be three images, for example, the combining unit 109 determines that the image capturing time interval before and after the interruption is within the predetermined time for each combination of the flow lines ML_1 to ML_4.
- a different predetermined time may be determined according to the estimated cause of the interrupted flow line. For example, as described above, when hiding behind a fixed object and hiding behind a moving object, when the line of flow is interrupted due to the latter cause of both moving, the time interval between interruptions of the line of flow is shorter than when the former is the cause. It is considered to be.
- the cause of the discontinuity of the line of flow may be estimated by obtaining the position of the fixed object in advance from the image and determining from the image whether or not the line of flow is discontinued in the vicinity of the fixed object.
- step S401 If it is determined that the imaging time interval is within the predetermined time (step S401; Yes), the combining unit 109 determines whether the distance of the person detected in the images before and after the flow lines ML_1 to ML_4 are interrupted is within the predetermined distance. It is determined whether or not (step S402). The determination processing regarding the discontinued distance in step S402 corresponds to the determination of whether or not the condition E described above is satisfied.
- the predetermined distance it is preferable to adopt the distance that a person generally moves in the above-described predetermined time.
- the predetermined time is defined as 3 images
- the predetermined distance is determined according to the distance a person moves in 3/N [sec].
- the distance that a person moves in a certain period of time is preferably determined based on a general walking speed (for example, 5 km/h) or faster.
- the distance of the person detected in the images before and after the flow lines ML_1 to ML_4 are interrupted is not within the predetermined distance, it can be determined that the person is not the same person. Also, when the distance is within a predetermined distance, it can be determined that the person may be the same person.
- the combining unit 109 determines that the distance of the person detected in the images before and after the interruption is within a predetermined distance for each combination of the flow lines ML_1 to ML_4.
- step S402 If it is determined that the distance of the detected person is within the predetermined distance (step S402; Yes), the combining unit 109 determines that the difference in orientation of the person detected in the images before and after the break in the flow lines ML_1 to ML_4 is a predetermined distance. (step S403).
- the judgment processing regarding the orientation of the person in step S403 corresponds to the judgment as to whether or not the condition F described above is satisfied.
- the orientation of a person can be estimated by, for example, whether or not a face area is included in an image, the direction of a line segment connecting both shoulders, etc. It is particularly effective when determining the identity of a person in a case.
- the face area of a person walking away from the camera 101a along the imaging direction of the camera 101a is not captured by the camera 101a.
- the face area of a person walking in the same direction toward the camera 101a is captured by the camera 101a. In this way, the orientation of a person can be estimated depending on whether or not the face area is included in the image.
- a line segment connecting both shoulders of a person who moves along the imaging direction of the camera 101a generally faces the horizontal direction of the imaging area A1.
- a line segment connecting both shoulders of a person moving in a direction perpendicular to the photographing direction of the camera 101a is approximately above and below the photographing area A1. turn direction. In this way, the orientation of a person can be estimated from the direction of the line segment connecting both shoulders.
- the lines of flow ML_1 and ML_3 are the lines of flow of people moving generally upward and to the right in the imaging area A1.
- Flow lines ML_2 and ML_4 are lines of flow of people who generally move upward and to the left in the imaging region A1. Therefore, it is often difficult to estimate the orientation of the person from the face area, because none of the persons on the flow lines ML_1 to ML_4 has a face area photographed or, even if an image is photographed, it is relatively small.
- the direction of the line segment connecting both shoulders does not differ as much in the angle between the person on the flow lines ML_1 and ML_3 and the person on the flow lines ML_2 and ML_4 as they are when they are orthogonal. Therefore, for each combination of the flow lines ML_1 to ML_4, the combining unit 109 determines that the difference in orientation of the person detected in the images before and after the interruption is within a predetermined range.
- step S403 If it is determined that the difference in orientation of the detected person is within the predetermined range (step S403; Yes), the combining unit 109 extracts the image features of the person detected in the images before and after the flow lines ML_1 to ML_4 break off. It is determined whether or not the degree of similarity in quantity is greater than or equal to the second reference value (step S404). The judgment processing regarding the image feature amount in this step S404 corresponds to the judgment as to whether or not the condition G described above is satisfied.
- step S404 If it is determined that the similarity of the image feature amount is equal to or greater than the second reference value (step S404; Yes), the combining unit 109, as shown in FIG. It is determined whether or not the degree of similarity of the posture feature amount of the person obtained is equal to or greater than a first reference value (step S405).
- the judgment processing regarding the posture feature amount in this step S405 corresponds to the judgment as to whether or not the condition A described above is satisfied.
- the combining unit 109 determines whether the similarity of the human posture feature amount is equal to or greater than the first reference value for the pair of the flow line ML_1 and the flow line ML_3 and the pair of the flow line ML_2 and the flow line ML_4. We judge that it is. For groups of flow lines other than these, the combining unit 109 determines that the degree of similarity of the person's posture feature amount is not equal to or greater than the first reference value.
- step S405 If it is determined that the similarity of the posture feature amount is equal to or higher than the first reference value (step S405; Yes), the combining unit 109 determines whether the same person overlaps temporally or different persons overlap geographically. It is determined whether or not to do so (step S406).
- the determination processing regarding duplication of the same person in step S406 corresponds to determination of whether or not conditions B and C described above are satisfied.
- the combining unit 109 prevents the same person from overlapping in time for the set of the flow line ML_1 and the flow line ML_3 and the set of the flow line ML_2 and the flow line ML_4. In addition, it is determined that different people do not overlap in terms of location.
- step S406 whether or not the same person overlaps in time and whether different persons overlap in location is determined by all groups included in each group. area, in other words, the entirety of each flow line.
- flow lines A, B, and C have ends at times TA, TB, and TC, respectively, and time TC is later than time TB, and time TB is later than time TA.
- time TC is later than time TB
- time TB is later than time TA.
- step S406 based on the entire flow line, it is determined that the same person overlaps in time.
- step S406 If it is determined that the same person does not overlap in terms of time and that different persons do not overlap in terms of location (step S406; No), the combining unit 109 selects It is determined that the combination of the flow lines ML_1 to ML_4 that are the same person.
- the combining unit 109 merges the groups of regions that constitute the flow lines ML_1 to ML_4 of the same person, that is, the groups of the same person (step S407). Further, the connecting unit 109 connects the ends of the flow lines ML_1 to ML_4 that are interrupted for the same person (step S408). After performing the process of step S408, the combining unit 109 ends the combining process (step S110).
- the set of flow lines ML_1 and ML_3 and the set of flow lines ML_2 and ML_4 satisfy conditions A to G as described above. Therefore, in step S407, the group G1 and the group G3 are merged as a group related to the person P. Group G2 and group G4 are merged as a group for person Q.
- step S408 as shown in FIG. 13, the flow line ML_1 and the flow line ML_3 are combined to generate a flow line ML_P related to the person P.
- the flow line ML_2 and the flow line ML_4 are combined to generate a flow line ML_Q for the person Q.
- step S110 the combining unit 109 terminates the combining process
- step S401 when it is determined that the shooting time interval is not within the predetermined time (step S401; No), when it is determined that the distance to the person is not within the predetermined distance (step S402; No), the orientation of the person is determined. is not within a predetermined range (step S403; No), or if it is determined that the similarity of the image feature quantity is not equal to or greater than the second reference value (step S404; No), the combining unit 109 The combining process (step S110) is terminated.
- step S110 when it is determined that the similarity of the posture feature amount is not equal to or greater than the first reference value (step S405; No), or the same person overlaps in time, or If it is determined that different people overlap in location (step S406; Yes), the combining unit 109 ends the combining process (step S110).
- step S110 When the combining process (step S110) ends, the process returns to the image analysis process shown in FIG. 5, and the detection unit 103 repeats the process of loop A (step S102). In this embodiment, the processing of steps S103 to S110 is further performed for the imaging area A2.
- regions R_T1 to R_T4 are grouped into a group G5 related to person R based on the images of time T1 to T8 (see FIGS. 2 and 6) of imaging region A2, and a flow line ML_R related to person R is determined. Generate. Also, the regions P_T7 to P_T8 are grouped into a group G6 to generate a flow line ML_5.
- loop A ends as shown in FIG.
- the combining unit 109 then combines the flow lines ML_P, ML_Q, ML_5, and ML_R between the different imaging areas A1 and A2 (step S111).
- the combination of flow lines to be processed in step S111 is a set of flow lines ML_P and ML_5, and a set of flow lines ML_Q and ML_5.
- the combining unit 109 determines that the flow lines belong to the same person, and determines that the flow lines belong to the same person. Merge between groups of regions that make up a line and join between the ends of the flow line. Also, if the similarity of the image feature amount is not equal to or greater than the second reference value, the combining unit 109 determines that the flow lines do not belong to the same person, does not merge the groups, and does not combine the flow lines.
- step S111 the flow lines of different imaging areas A1 and A2 are combined as shown in FIG. 13 shows an example in which a flow line ML_P is generated by combining the flow line ML_5 shown in FIG. 14 with the flow line ML_P shown in FIG. Further, between the imaging area A1 and the imaging area A2, the flow line ML_P is connected by a flow line that smoothly connects the flow line ML_P and the flow line ML_5 shown in FIG. 13 .
- groups may be merged or flow lines may be combined based on whether or not an appropriate combination of conditions A to G is satisfied.
- the identification image output unit 105 outputs identification image information based on the determination result of the determination unit 104 (step S112), and ends the image analysis process.
- the identification image information for example, the movement lines of people detected in a plurality of images are used as identification information, and the identification information is associated with the person shown in each image by connecting the areas of the people with the movement lines. (See FIG. 15). Also, for example, the image indicated by the identification image information has identification information such as numbers, symbols, codes (for example, P, Q, R) for identifying a person, and the identification information corresponds to the area of the person indicated in the image. This is the attached image.
- the determination result of the identity of a person detected in a plurality of consecutive images in time series is output as identification image information
- the determination result of identity is not limited to images.
- a person and the posture of the person are detected in each of a plurality of images that are consecutive in time series, and different images are created using the detected posture of the person.
- the identity of the detected person is determined.
- the posture of a person detected in each of the images shot in time series within a predetermined period of time among a plurality of images is used to identify the posture of the person detected in different images. determine gender. This makes it possible to determine the identity of a person more accurately. Therefore, it is possible to identify the same person more accurately in a plurality of images that are consecutive in time series.
- the identity of persons detected in different images is determined using postures of persons within a predetermined distance. This makes it possible to determine the identity of a person more accurately. Therefore, it is possible to identify the same person more accurately in a plurality of images that are consecutive in time series.
- the orientation of a person detected in each of a plurality of images is obtained. Then, among the persons detected in each of the plurality of images, the identity of the persons detected in the different images is determined using the postures of the persons whose orientation difference is within a predetermined range. to decide. This makes it possible to determine the identity of a person more accurately. Therefore, it is possible to identify the same person more accurately in a plurality of images that are consecutive in time series.
- the detected posture of the person is used to determine the posture feature amount of the person. Then, based on whether or not the degree of similarity of the obtained posture feature amount is equal to or greater than a predetermined reference value, the identity of the persons detected in the mutually different images is determined.
- the identity of a person from the image feature amount it is possible to determine the identity of the person even when it is difficult to track the person, such as when the line of flow of the person is interrupted. Therefore, it is possible to accurately identify the same person in a plurality of images that are consecutive in time series.
- the similarity of the calculated posture feature amount is equal to or greater than a predetermined reference value
- the same person overlaps temporally or different persons overlap spatially.
- it is determined that persons detected in different images are not the same person. This can prevent a person's identity from being determined in a way that cannot actually happen. Therefore, it is possible to identify the same person more accurately in a plurality of images that are consecutive in time series.
- the degree of similarity of the obtained posture feature values is equal to or greater than a predetermined reference value
- the same person does not overlap in terms of time, and different persons do not overlap with each other
- the persons detected in different images are determined to be the same person. This can prevent a person's identity from being determined in a way that cannot actually happen. Therefore, it is possible to identify the same person more accurately in a plurality of images that are consecutive in time series.
- an image in which information identifying a person in a plurality of images is associated with the person shown in each image is output based on the result of determination regarding the identity of the detected person.
- the present invention is not limited to these.
- the present invention also includes a form obtained by appropriately combining part or all of the embodiments and modifications described above, and a form obtained by appropriately modifying the form.
- an image acquiring means for acquiring a plurality of images consecutive in time series; detection means for detecting a person and the posture of the person in each of the plurality of images;
- An image analysis apparatus comprising: determining means for determining identity of persons detected in mutually different images using the detected person's posture. 2.
- the determining means determines identity of the person detected in the different images by using the posture of the person detected in each of the images taken within a predetermined time in chronological order among the plurality of images. 2.
- the image analysis device according to 1. 3. 1 or 2, wherein the determining means determines the identity of the person detected in each of the images different from each other by using the posture of the person within a predetermined distance among the persons detected in each of the plurality of images.
- the image analysis device according to . 4.
- the determination means obtains the orientation of the person detected in each of the plurality of images, and among the persons detected in each of the plurality of images, the difference in the obtained orientation is within a predetermined range. 4.
- the image analysis device according to any one of 1 to 3, wherein the identity of the person detected in the mutually different images is determined using a certain person's posture. 5.
- the determining means obtains an image feature amount of a person detected in each of the plurality of images, and determines in advance a similarity of the obtained image feature amount among the persons detected in each of the plurality of images. 5.
- the image analysis device according to any one of 1 to 4, wherein the identity of the person detected in the mutually different images is determined based on whether or not it is equal to or greater than the determined reference value. 6.
- the determination means is feature acquisition means for obtaining a posture feature of the person using the detected posture of the person; identifying means for determining the identity of the person detected in the mutually different images based on whether or not the similarity of the calculated posture feature amount is equal to or greater than a predetermined reference value; 6.
- the image analysis device according to any one of 5.
- the specifying means determines that the same person overlaps in time or different persons overlap in place when the similarity of the calculated posture feature amount is equal to or greater than a predetermined reference value. 7.
- the image analysis apparatus according to 6, wherein the persons detected in the mutually different images are not the same person when . 8.
- the identifying means prevents the same person from overlapping when the similarity of the calculated posture feature quantity is equal to or greater than a predetermined reference value, and prevents the same person from overlapping and the different persons from overlapping locations.
- identification image output means for outputting an image in which the person detected in each of the plurality of images is associated with information for identifying the person, based on the result of determination by the determination means. or the image analysis device according to one. 10. one or more imaging means; An image analysis system comprising the image analysis device according to any one of 1 to 9. 11.
- the computer Acquiring a plurality of consecutive images in time series; detecting a person and a posture of the person in each of the plurality of images; and determining the identity of a person between different images using the detected pose of the person. 12. to the computer, Acquiring a plurality of consecutive images in time series; detecting a person and a posture of the person in each of the plurality of images; and judging the identity of a person between different images using the detected posture of the person.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
Description
時系列的に連続する複数の画像を取得する画像取得手段と、
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出する検出手段と、
前記検出された人の姿勢を用いて、互いに異なる画像にて検出された人の同一性を判断する判断手段とを備える。
1つ又は複数の撮影手段と、
上記の画像解析装置とを備える。
コンピュータが、
時系列的に連続する複数の画像を取得することと、
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出することと、
前記検出された人の姿勢を用いて、互いに異なる画像間における人の同一性を判断することとを含む。
コンピュータに、
時系列的に連続する複数の画像を取得することと、
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出することと、
前記検出された人の姿勢を用いて、互いに異なる画像間における人の同一性を判断することとを実行させるためのプログラムである。
本発明の一実施の形態に係る画像解析システムは、時系列的に連続する複数の画像に基づいて互いに異なる画像間における人の同一性を判断し、その判断の結果に基づいて人の動線を求めるなどの処理を行う。
画像取得部102は、撮影領域A1~A2を撮影した時系列的に連続する複数の画像を取得する。本実施の形態では、画像取得部102は、カメラ101a~101bの各々によって生成された画像情報を、有線、無線又はこれらを適宜組み合わせて構成されたネットワークを介してカメラ101a~101bの各々から取得する。
条件B:同一人が時間的に重複して存在することにならないこと
条件C:異なる人が場所的に重複して存在することにならないこと
条件E:動線が途切れる前後の画像にて検出された人の距離が所定距離内であること
条件F:動線が途切れる前後の画像にて検出された人の向きの違いが所定の範囲内であること
条件G:動線が途切れる前後の画像にて検出された人の画像特徴量の類似度が第2基準値以上であること
ここから、本実施の形態に係る画像解析システムの物理的構成の例について、図を参照して説明する。
ここから、本発明の一実施の形態に係る画像解析処理について図を参照して説明する。
結合部109は、ステップS107にて生成された動線ML_1~ML_4に途切れた動線が含まれるか否かを判断する(ステップS108)。
撮影時間間隔が所定時間内であると判断した場合(ステップS401;Yes)、結合部109は、動線ML_1~ML_4が途切れる前後の画像にて検出された人の距離が所定距離内であるか否かを判断する(ステップS402)。このステップS402での途切れた距離に関する判断処理は、上述の条件Eを満たすか否かの判断に相当する。
検出された人の距離が所定距離内であると判断した場合(ステップS402;Yes)、結合部109は、動線ML_1~ML_4が途切れる前後の画像にて検出された人の向きの違いが所定の範囲内であるか否かを判断する(ステップS403)。このステップS403での人の向きに関する判断処理は、上述の条件Fを満たすか否かの判断に相当する。
検出された人の向きの違いが所定の範囲内であると判断した場合(ステップS403;Yes)、結合部109は、動線ML_1~ML_4が途切れる前後の画像にて検出された人の画像特徴量の類似度が第2基準値以上であるか否かを判断する(ステップS404)。このステップS404での画像特徴量に関する判断処理は、上述の条件Gを満たすか否かの判断に相当する。
画像特徴量の類似度が第2基準値以上であると判断した場合(ステップS404;Yes)、結合部109は、図12に示すように、動線ML_1~ML_4が途切れる前後の画像にて検出された人の姿勢特徴量の類似度が第1基準値以上であるか否かを判断する(ステップS405)。このステップS405での姿勢特徴量に関する判断処理は、上述の条件Aを満たすか否かの判断に相当する。
姿勢特徴量の類似度が第1基準値以上であると判断した場合(ステップS405;Yes)、結合部109は、同一人が時間的に重複して又は異なる人が場所的に重複して存在することになるか否かを判断する(ステップS406)。このステップS406での同一人の重複に関する判断処理は、上述の条件B及びCを満たすか否かの判断に相当する。
ステップS401~S406にて上述とは異なる判断をした場合、結合部109は、結合処理(ステップS110)を終了する。
識別画像出力部105は、判断部104による判断の結果に基づく識別画像情報を出力し(ステップS112)、画像解析処理を終了する。
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出する検出手段と、
前記検出された人の姿勢を用いて、互いに異なる画像にて検出された人の同一性を判断する判断手段とを備える
画像解析装置。
2.前記判断手段は、前記複数の画像のうち、時系列的に所定の時間内に撮影された画像それぞれにおいて検出された人の姿勢を用いて、前記互いに異なる画像にて検出された人の同一性を判断する
1に記載の画像解析装置。
3.前記判断手段は、前記複数の画像のそれぞれにおいて検出された人のうち、所定の距離内の人の姿勢を用いて、前記互いに異なる画像にて検出された人の同一性を判断する
1又は2に記載の画像解析装置。
4.前記判断手段は、前記複数の画像のそれぞれにおいて検出された人の向きを求め、前記複数の画像のそれぞれにおいて検出された人のうち、前記求められた向きの違いが予め定められた範囲内である人の姿勢を用いて、前記互いに異なる画像にて検出された人の同一性を判断する
1から3のいずれか1つに記載の画像解析装置。
5.前記判断手段は、前記複数の画像のそれぞれにおいて検出された人の画像特徴量を求め、前記複数の画像のそれぞれにおいて検出された人のうち、前記求められた画像特徴量の類似度が予め定められた基準値以上であるか否かに基づいて、前記互いに異なる画像にて検出された人の同一性を判断する
1から4のいずれか1つに記載の画像解析装置。
6.前記判断手段は、
前記検出された人の姿勢を用いて、当該人の姿勢特徴量を求める特徴量取得手段と、
前記求められた姿勢特徴量の類似度が予め定められた基準値以上であるか否かに基づいて、前記互いに異なる画像にて検出された人の同一性を判断する特定手段とを含む
1から5のいずれか1つに記載の画像解析装置。
7.前記特定手段は、前記求められた姿勢特徴量の類似度が予め定められた基準値以上である場合において、同一人が時間的に重複して又は異なる人が場所的に重複して存在することになるとき、前記互いに異なる画像にて検出された人を同一人ではないと判断する
6に記載の画像解析装置。
8.前記特定手段は、前記求められた姿勢特徴量の類似度が予め定められた基準値以上である場合において、同一人が重複して存在することにならず、かつ、異なる人が場所的に重複して存在することにならないとき、前記互いに異なる画像にて検出された人を同一人であると判断する
6又は7に記載の画像解析装置。
9.前記判断手段による判断の結果に基づいて、前記複数の画像それぞれにおいて検出された人に当該人を識別するための情報を対応付けた画像を出力する識別画像出力手段をさらに備える
1から8のいずれか1つに記載の画像解析装置。
10.1つ又は複数の撮影手段と、
1から9のいずれか1つに記載の画像解析装置とを備える
画像解析システム。
11.コンピュータが、
時系列的に連続する複数の画像を取得することと、
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出することと、
前記検出された人の姿勢を用いて、互いに異なる画像間における人の同一性を判断することとを含む
画像解析方法。
12.コンピュータに、
時系列的に連続する複数の画像を取得することと、
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出することと、
前記検出された人の姿勢を用いて、互いに異なる画像間における人の同一性を判断することとを実行させるためのプログラム。
101a,101b カメラ
102 画像取得部
103 検出部
104 判断部
105 識別画像出力部
106 特徴量取得部
107 特定部
108 グループ化部
109 結合部
Claims (12)
- 時系列的に連続する複数の画像を取得する画像取得手段と、
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出する検出手段と、
前記検出された人の姿勢を用いて、互いに異なる画像にて検出された人の同一性を判断する判断手段とを備える
画像解析装置。 - 前記判断手段は、前記複数の画像のうち、時系列的に所定の時間内に撮影された画像それぞれにおいて検出された人の姿勢を用いて、前記互いに異なる画像にて検出された人の同一性を判断する
請求項1に記載の画像解析装置。 - 前記判断手段は、前記複数の画像のそれぞれにおいて検出された人のうち、所定の距離内の人の姿勢を用いて、前記互いに異なる画像にて検出された人の同一性を判断する
請求項1又は2に記載の画像解析装置。 - 前記判断手段は、前記複数の画像のそれぞれにおいて検出された人の向きを求め、前記複数の画像のそれぞれにおいて検出された人のうち、前記求められた向きの違いが予め定められた範囲内である人の姿勢を用いて、前記互いに異なる画像にて検出された人の同一性を判断する
請求項1から3のいずれか1項に記載の画像解析装置。 - 前記判断手段は、前記複数の画像のそれぞれにおいて検出された人の画像特徴量を求め、前記複数の画像のそれぞれにおいて検出された人のうち、前記求められた画像特徴量の類似度が予め定められた基準値以上であるか否かに基づいて、前記互いに異なる画像にて検出された人の同一性を判断する
請求項1から4のいずれか1項に記載の画像解析装置。 - 前記判断手段は、
前記検出された人の姿勢を用いて、当該人の姿勢特徴量を求める特徴量取得手段と、
前記求められた姿勢特徴量の類似度が予め定められた基準値以上であるか否かに基づいて、前記互いに異なる画像にて検出された人の同一性を判断する特定手段とを含む
請求項1から5のいずれか1項に記載の画像解析装置。 - 前記特定手段は、前記求められた姿勢特徴量の類似度が予め定められた基準値以上である場合において、同一人が時間的に重複して又は異なる人が場所的に重複して存在することになるとき、前記互いに異なる画像にて検出された人を同一人ではないと判断する
請求項6に記載の画像解析装置。 - 前記特定手段は、前記求められた姿勢特徴量の類似度が予め定められた基準値以上である場合において、同一人が重複して存在することにならず、かつ、異なる人が場所的に重複して存在することにならないとき、前記互いに異なる画像にて検出された人を同一人であると判断する
請求項6又は7に記載の画像解析装置。 - 前記判断手段による判断の結果に基づいて、前記複数の画像それぞれにおいて検出された人に当該人を識別するための情報を対応付けた画像を出力する識別画像出力手段をさらに備える
請求項1から8のいずれか1項に記載の画像解析装置。 - 1つ又は複数の撮影手段と、
請求項1から9のいずれか1項に記載の画像解析装置とを備える
画像解析システム。 - コンピュータが、
時系列的に連続する複数の画像を取得することと、
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出することと、
前記検出された人の姿勢を用いて、互いに異なる画像間における人の同一性を判断することとを含む
画像解析方法。 - コンピュータに、
時系列的に連続する複数の画像を取得することと、
前記複数の画像それぞれにおいて、人及び当該人の姿勢を検出することと、
前記検出された人の姿勢を用いて、互いに異なる画像間における人の同一性を判断することとを実行させるためのプログラム。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2023508783A JP7525055B2 (ja) | 2021-03-23 | 2022-02-16 | 画像解析装置、画像解析方法及びプログラム |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2021048550 | 2021-03-23 | ||
JP2021-048550 | 2021-03-23 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022201987A1 true WO2022201987A1 (ja) | 2022-09-29 |
Family
ID=83396895
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2022/006213 WO2022201987A1 (ja) | 2021-03-23 | 2022-02-16 | 画像解析装置、画像解析システム、画像解析方法及びプログラム |
Country Status (4)
Country | Link |
---|---|
JP (1) | JP7525055B2 (ja) |
AR (1) | AR125574A1 (ja) |
TW (1) | TW202240470A (ja) |
WO (1) | WO2022201987A1 (ja) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005086626A (ja) * | 2003-09-10 | 2005-03-31 | Matsushita Electric Ind Co Ltd | 広域監視装置 |
WO2006013765A1 (ja) * | 2004-08-03 | 2006-02-09 | Matsushita Electric Industrial Co., Ltd. | 人物判定装置及び人物検索追跡装置 |
JP2010239992A (ja) * | 2009-03-31 | 2010-10-28 | Sogo Keibi Hosho Co Ltd | 人物同定装置、人物同定方法、及び人物同定プログラム |
JP2015002547A (ja) * | 2013-06-18 | 2015-01-05 | 富士通株式会社 | 画像処理装置、プログラム、および画像処理方法 |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3785456B2 (ja) | 2002-07-25 | 2006-06-14 | 独立行政法人産業技術総合研究所 | 駅ホームにおける安全監視装置 |
JP6377563B2 (ja) | 2015-03-27 | 2018-08-22 | Kddi株式会社 | 動線の部分削除処理を利用して物体を識別する物体追跡装置、プログラム及び方法 |
JP6794575B1 (ja) | 2020-07-28 | 2020-12-02 | 株式会社 日立産業制御ソリューションズ | 動画解析装置及び動画解析方法 |
-
2022
- 2022-02-16 WO PCT/JP2022/006213 patent/WO2022201987A1/ja active Application Filing
- 2022-02-16 JP JP2023508783A patent/JP7525055B2/ja active Active
- 2022-02-23 TW TW111106508A patent/TW202240470A/zh unknown
- 2022-03-23 AR ARP220100697A patent/AR125574A1/es unknown
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005086626A (ja) * | 2003-09-10 | 2005-03-31 | Matsushita Electric Ind Co Ltd | 広域監視装置 |
WO2006013765A1 (ja) * | 2004-08-03 | 2006-02-09 | Matsushita Electric Industrial Co., Ltd. | 人物判定装置及び人物検索追跡装置 |
JP2010239992A (ja) * | 2009-03-31 | 2010-10-28 | Sogo Keibi Hosho Co Ltd | 人物同定装置、人物同定方法、及び人物同定プログラム |
JP2015002547A (ja) * | 2013-06-18 | 2015-01-05 | 富士通株式会社 | 画像処理装置、プログラム、および画像処理方法 |
Also Published As
Publication number | Publication date |
---|---|
JPWO2022201987A1 (ja) | 2022-09-29 |
AR125574A1 (es) | 2023-07-26 |
JP7525055B2 (ja) | 2024-07-30 |
TW202240470A (zh) | 2022-10-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6295645B2 (ja) | 物体検出方法及び物体検出装置 | |
CN107431786B (zh) | 图像处理设备、图像处理***和图像处理方法 | |
CN102262725B (zh) | 三维场景的分析 | |
Gabriel et al. | The state of the art in multiple object tracking under occlusion in video sequences | |
CN104717481B (zh) | 摄像装置、图像处理装置、摄像方法 | |
JP5001260B2 (ja) | オブジェクト追跡方法及びオブジェクト追跡装置 | |
CN107438173A (zh) | 视频处理装置、视频处理方法和存储介质 | |
Khan et al. | Tracking in uncalibrated cameras with overlapping field of view | |
JP2007142527A (ja) | 移動体監視システム、移動体特徴量算出装置およびカメラ間移動体照合追跡装置 | |
JP2006343859A (ja) | 画像処理装置及び画像処理方法 | |
JP5101429B2 (ja) | 画像監視装置 | |
KR20200113743A (ko) | 인체 자세 추정 및 보정을 하는 방법 및 장치 | |
JPH11257931A (ja) | 物体認識装置 | |
JP2001084383A (ja) | 移動検出方法 | |
WO2022201987A1 (ja) | 画像解析装置、画像解析システム、画像解析方法及びプログラム | |
JP5416489B2 (ja) | 三次元指先位置検出方法、三次元指先位置検出装置、およびプログラム | |
JP2019096062A (ja) | 物体追跡装置、物体追跡方法、および、物体追跡プログラム | |
KR20220029946A (ko) | 영상에 포함된 객체를 추적하는 장치, 방법 및 컴퓨터 프로그램 | |
JP5448952B2 (ja) | 同一人判定装置、同一人判定方法および同一人判定プログラム | |
JP2005250692A (ja) | 物体の同定方法、移動体同定方法、物体同定プログラム、移動体同定プログラム、物体同定プログラム記録媒体、移動体同定プログラム記録媒体 | |
JP2009301242A (ja) | 頭部候補抽出方法、頭部候補抽出装置、頭部候補抽出プログラムおよびそのプログラムを記録した記録媒体 | |
US20020085747A1 (en) | Image processing apparatus and method, image capturing apparatus, and information provision medium | |
JP2004046464A (ja) | 移動物体3次元位置推定装置及びその方法、プログラム並びにその記録媒体 | |
JP6504711B2 (ja) | 画像処理装置 | |
JP6548306B2 (ja) | カメラの撮影画像に映る人物を追跡する画像解析装置、プログラム及び方法 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22774787 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2023508783 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18273136 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: 22774787 Country of ref document: EP Kind code of ref document: A1 |