JP5060264B2 - Human detection device - Google Patents

Description

  The present invention relates to a person detection device that detects a person based on a difference between an input image and a background image, and more particularly to accurately determine a person including a stationary person even when there is illumination fluctuation or background fluctuation. The present invention relates to a person detection device capable of

  A background difference that obtains a background image that is a background image in advance and extracts an area that has changed from the background by taking the difference between the background image and the input image (hereinafter referred to as “background different area”). The law is known.

  When this background subtraction method is used for detecting a moving object such as a person, lighting fluctuation due to ON / OFF of a lighting fixture or a change in sunlight becomes a problem. This is because the background subtraction method calculates the difference between the background image and the input image, so if there is a difference between the background image at the time of acquisition and the actual background, the background that has changed due to illumination fluctuations is detected as a moving object. Because it will do.

  For this reason, a method for preventing erroneous detection due to illumination variation has been proposed. For example, Patent Literature 1 discloses a method for determining whether a background different area is caused by illumination fluctuation or by passing an object such as a person. Specifically, the fluctuation per unit time of the optical feature parameter value in a predetermined small area is calculated as a statistical feature quantity, and the calculated statistical feature quantity is a statistical feature quantity of illumination fluctuation obtained in advance. To identify lighting fluctuations.

JP-A-7-302328

  However, the technique of Patent Document 1 has a problem that the performance of moving object detection is greatly influenced by the size of the target small region and the value of the unit time in the statistical processing. For example, when the unit time is set small (for example, t seconds), it is difficult to detect a person who remains in the same place for t seconds or more. On the other hand, if the unit time is set to a large value, it is difficult to detect illumination fluctuation.

  In addition, when the size of the small area is set small, it is possible to detect a fine movement of the moving object, but it is difficult to detect a variation in illumination. The reason is that when the illumination variation is seen in a small area, the brightness changes little by little due to the influence of reflection or the like, so that it is difficult to distinguish from the fine movement of the moving object.

  As described above, although the technique of Patent Document 1 can distinguish a moving object from an illumination variation only when the size and unit time of a small region are set to appropriate values, it can be distinguished from an illumination variation. , Limited to moving objects with large movements such as walking people. That is, even when the technique of Patent Document 1 is used, it is difficult to detect a stationary person who stays at the same position by distinguishing it from illumination variation.

  By the way, when the background subtraction method is used, it is necessary to deal with not only illumination variation but also the background itself. For example, when changing the arrangement of the ropes for organizing the matrix according to the length of the matrix, such as a bank or an airport, it is necessary to distinguish the background fluctuations associated with the change of the rope arrangement from the moving object.

  For these reasons, it has become a major issue how to realize a person detection apparatus that can accurately determine a person including a stationary person even when there is a lighting fluctuation or a background fluctuation.

  The present invention has been made to solve the above-described problems caused by the prior art, and can detect a person including a stationary person with high accuracy even when there is illumination fluctuation or background fluctuation. An object is to provide an apparatus.

In order to solve the above-described problems and achieve the object, the present invention is a person detection device that detects a person based on a difference between an input image and a background image, and is a pixel composed of one or more pixels in the input image. For each unit , there is an alternating repetition of light and dark based on the difference between the average luminance value in the past first predetermined period and the current input luminance value, and this repetition is repeated a predetermined number of times within the second predetermined period, and the light and dark fluctuations A fine movement detecting means for detecting a fine movement of a luminance change due to a non-constant period, an edge detected from the input image, and an edge that is stationary for a third predetermined period and an edge that is continuous for a certain number of pixels in the same direction an edge detecting means for excluding, said fine said by the detecting means fine movement is detected, and detects the person based on the pixel where the edge is detected by said edge detection means Characterized in that a person detection means for.

Further, the present invention is a person detection device for detecting a person based on a difference between an input image and a background image, and setting means for setting a detection target area that is a target detection area on the input image; For each pixel unit composed of one or a plurality of pixels in the detection target region set by the setting means , alternating light and dark are repeated based on the difference between the past average luminance value in the first predetermined period and the current input luminance value. Yes, the repetition is repeated a predetermined number of times within the second predetermined period, and the fine movement detecting means for detecting the fine movement of the luminance change for detecting the fine movement due to the fact that the period of the brightness fluctuation is not substantially constant, and the fine movement detecting means Person detection means for detecting the detection target area as a person area where a person exists when the number of pixel units in which fine movement is detected is equal to or greater than a predetermined threshold. And features.

  Further, the present invention is characterized in that, in the above invention, the setting means sets a background difference area, which is an area where the input image and the background image are different, as the detection target area.

  Further, the present invention is characterized in that, in the above invention, the setting means sets, as the detection target area, a definition area that is an area defined in advance according to an operation place.

The present invention, in the above invention, further comprises edge detection means for detecting an edge from the input image and excluding an edge that is stationary for a third predetermined period and an edge that is continuous for a certain number of pixels in the same direction, The person detecting means detects the detection target area when the fine movement is detected by the fine movement detecting means and the number of the pixel units in which the edge is detected by the edge detecting means is equal to or larger than a predetermined threshold. It is detected as a person area.

  Further, the present invention provides the update means for updating the background image by taking in the background image a non-detection area that is the detection target area that the person detection means did not detect as the person area in the above invention. It is further provided with a feature.

  Further, according to the present invention, in the above invention, on the condition that the person detecting means detects the non-detection area as the person area within a predetermined time after the updating means updates the background image. An update canceling unit for canceling the update of the background image by the image updating unit is further provided.

In addition, the present invention is a person detection device that detects a person based on a difference between an input image and a background image, and a definition area that is a predefined area according to an operation place is a person detection target area. A setting means for setting as a detection target area, and an average luminance value in the past first predetermined period and a current input luminance value for each pixel unit composed of one or a plurality of pixels in the detection target area set by the setting means . There is an alternating repetition of light and dark based on the difference, this repetition is repeated a predetermined number of times within the second predetermined period, and fine movement detecting means for detecting fine movement of the luminance change due to the period of the light and dark fluctuation being not substantially constant ; and it detects an edge from the input image, an edge detection unit, except for the edges successively a predetermined pixel or the third predetermined time period still to have an edge and the same direction, the said fine detection means Said fine has been detected, and the number of the pixels in which the edge is detected by the edge detecting means has a counting means for counting the number of the detection target area is above a predetermined threshold value as a number Te It is characterized by.

In addition, the present invention is a person detection device that detects a person based on a difference between an input image and a background image, and a definition area that is a predefined area according to an operation place is a person detection target area. A setting means for setting a number counting area, which is a target for counting the number of persons, and setting as a detection target area, and a past number for each pixel unit composed of one or a plurality of pixels in the detection target area set by the setting means . 1 There is a light and dark alternating repetition based on the difference between the average luminance value for a predetermined period and the current input luminance value, this repetition is repeated a predetermined number of times within the second predetermined period, and the period of light and dark fluctuations is substantially constant. fine movement detecting means for detecting a fine movement of luminance change by the absence, and detects an edge from the input image, a certain pixel over a third predetermined period still to have an edge and the same direction communicating An edge detecting unit, excluding the edges, the said fine by fine movement detection means is detected and the number of the pixels in which the edge is detected by said edge detection means of the people counting area is above a predetermined threshold value And a counting means for counting the number as the number of persons.

  Further, the present invention is the above invention, wherein the number of persons counted by the counting means and the area in which the number of persons is counted are associated with each other and stored in time series, and the time series stored in the storage means And generating means for generating a person's stay time density map based on the number of persons.

According to the present invention, for each pixel unit consisting of one or a plurality of pixels in the input image , there is an alternating repetition of light and dark based on the difference between the average luminance value of the past first predetermined period and the current input luminance value, This repetition is repeated a predetermined number of times within the second predetermined period, and the brightness fluctuation is not substantially constant, thereby detecting a slight change in luminance, detecting an edge from the input image, and resting for the third predetermined period. Except for edges that are present and continuous edges of a certain number of pixels or more in the same direction , and the person is detected based on the pixel unit from which the edge was detected. Performs human detection by combining fine movements that tend to appear and edges that appear in the contours of objects that include people, enabling high accuracy for people including stationary people even when there are variations in lighting and background. An effect that can be discriminated.

Further, according to the present invention, a detection target region that is a target region for human detection is set on the input image, and a past first predetermined period is set for each pixel unit including one or a plurality of pixels in the set detection target region. There is an alternating repetition of light and dark based on the difference between the average luminance value and the current input luminance value, and this repetition is repeated a predetermined number of times within the second predetermined period, and the period of fluctuation of light and dark is not substantially constant. When the number of pixel units in which fine movement is detected is equal to or greater than a predetermined threshold, the detection target area is detected as a person area where a person exists. The presence of a person in each detection target area is highly accurate even if a slight movement is mistakenly detected by making the area where a person exists in the detection target area where the fine movement that is likely to appear in the outline of the object is recognized above a predetermined threshold. In An effect that it is possible to separate.

  In addition, according to the present invention, since the background difference area, which is an area where the input image and the background image are different, is set as the detection target area, it is possible to detect a person based only on the camera image. Play.

  In addition, according to the present invention, since the definition area that is a predefined area according to the operation location is set as the detection target region, it is possible to detect a person regardless of the operation location. .

In addition, according to the present invention, edges are detected from the input image, and fine movement is detected and edges are detected except for edges that are stationary for a third predetermined period and edges that are continuous for a certain number of pixels in the same direction. When the number of pixel units is equal to or greater than a predetermined threshold, the detection target area is detected as a person area. Therefore, the fine movement that easily appears in the outline of a person including a stationary person and the outline of an object including a person By using the number of pixel units in which the edge appearing in the image appears as a person detection criterion, there is an effect that a person including a stationary person can be determined with high accuracy.

  In addition, according to the present invention, the background image is updated by moving the article or the like because the background image is updated by capturing the non-detection area, which is the detection target area not detected as the person area, into the background image. Even if it is a case, there exists an effect that the person including a still person can be discriminate | determined with high precision.

  Further, according to the present invention, the background image update is canceled on the condition that the non-detection area is detected as a person area within a predetermined time after the background image is updated. Even if it is a case, there exists an effect that the detection accuracy of a person can be improved by returning to the background image before erroneously updating.

In addition, according to the present invention, a definition area that is a predefined area according to an operation place is set as a detection target area that is a target area for person detection, and is composed of one or a plurality of pixels in the set detection target area. For each pixel unit , there is an alternating repetition of light and dark based on the difference between the average luminance value of the past first predetermined period and the current input luminance value, and this repetition is repeated a predetermined number of times within the second predetermined period. Detects slight changes in luminance due to the fact that the period of fluctuation is not substantially constant, detects edges from the input image, and removes fine edges except for edges that are stationary for a third predetermined period and edges that are continuous for a certain number of pixels in the same direction. There are detected, and the number of pixels that the edge has been detected was counting the number of detection target region is above a predetermined threshold value as a number, a contour portion of a person, including a stationary person By tends to appear fine movement counts as number the number of detection target region recognized more than a predetermined threshold value, an effect that it is possible to determine the number of persons conveniently.

Further, according to the present invention, a definition area that is a predefined area according to the operating location is set as a detection target area that is a person detection target area, and a number counting area that is a target for counting the number of persons is set. Then, for each pixel unit composed of one or a plurality of pixels in the set detection target region , there is an alternating repetition of light and dark based on the difference between the past average luminance value in the first predetermined period and the current input luminance value, 2 This repetition is repeated a predetermined number of times within a predetermined period, and the brightness fluctuations are not substantially constant, so that a slight change in luminance is detected, an edge is detected from the input image, and still for a third predetermined period. except for edges successively certain pixels over the edge and the same direction, the fine movement is detected and the number of pixels where the edge is detected a number of people counting area is above a predetermined threshold number By counting by an effect that it is possible to determine the number of persons flexible.

  Further, according to the present invention, the counted number of people and the area in which the number of people is counted are stored in time series, and a person staying time density diagram is generated based on the stored time series number of persons. Therefore, the effect that the change of the number of persons can be provided visually and clearly is produced.

  Exemplary embodiments of a person detection apparatus according to the present invention will be described below in detail with reference to the accompanying drawings. In the following, a case will be described in which human detection is performed using a fisheye camera that captures an area to be detected from above from above.

  First, feature points of the person detection apparatus according to the present embodiment will be described with reference to FIG. FIG. 1 is a diagram illustrating the characteristics of the person detection apparatus according to the present embodiment. In the figure, a case where person detection is performed at an ATM (Automated Teller Machine) corner installed in a bank or the like is shown.

  As shown in the figure, the “background image” includes an ATM corner in an unattended state and a procession rope. On the other hand, the “input image” shows a person who performs an operation at each ATM and a person who waits for the turn of the ATM. In the person detection apparatus according to the present embodiment, a difference image between the “background image” and the “input image” (hereinafter referred to as “background difference image”) is generated.

  As shown in the figure, the “background difference image” includes a block made up of a square aggregate. Here, each square is a pixel group composed of a predetermined number of pixels (hereinafter referred to as “pixel unit”), and is a minimum unit of person detection processing. Details of this pixel unit will be described later with reference to FIG.

  The blocks included in the “background difference image” include a block corresponding to a person and a block corresponding to a person other than the person (for example, an article brought in, an article moved, a shadow due to sunlight, etc.). In this case, it is necessary to detect only the block corresponding to the person.

  Therefore, the person detection apparatus according to the present embodiment uses the fact that the “fine movement” of the luminance change in the above-described pixel unit is likely to appear in the contour portion of the person, and uses a block corresponding to a person other than a block corresponding to a person. It was decided to distinguish. Here, the “fine movement” indicates that the difference between the average luminance value and the input luminance value over a predetermined time is alternately repeated. In addition, a pixel unit that is recognized as “finely moving” is referred to as a pixel unit in which “pixel feature” appears.

  Since such a pixel feature is likely to appear in a contour portion of a person including a stationary person, a block in which the pixel feature is recognized can be a block corresponding to the person. However, in the person detection apparatus according to the present embodiment, the pixel feature and the “edge feature” are combined in order to improve the accuracy of person detection. Here, the “edge feature” refers to a detected edge obtained by removing an edge that is stationary for a predetermined time and an edge that is continuous for a certain number of pixels in the same direction. The edge refers to a portion where the luminance change in the image is abrupt, and is detected by using, for example, a Laplacian filter.

  Such an “edge feature” appears in the contour portion of an object including a person. On the other hand, because “pixel features” are likely to appear in the outline of a person as described above, by taking the common parts of the two features of “pixel features” and “edge features”, the misdetected portions of each feature are complemented. It becomes possible to meet each other.

  That is, in the human detection device according to the present embodiment, edge processing for detecting edge features is performed (see (1A) in the figure), and fine motion detection processing for detecting pixel features is performed (see (1B) in the figure). The person detection process is performed by combining the detected features (see (2) in the figure). Therefore, even if there are illumination fluctuations and background fluctuations, it is possible to determine a person including a stationary person with high accuracy.

  Next, the configuration of the person detection apparatus according to the present embodiment will be described with reference to FIG. FIG. 2 is a diagram illustrating a configuration of the person detection device 1 according to the present embodiment. As shown in the figure, the person detection device 1 includes a camera 2, a display 3, a control unit 10, and a storage unit 20. In addition, the control unit 10 includes an image input unit 11, a setting unit 12, a background difference area detection unit 13, a fine movement detection unit 14, an edge detection unit 15, a person detection unit 16, and a background image update unit 17. The storage unit 20 stores an input image 21, a background image 22, and a detection target area 23.

  The camera 2 is a camera such as a CCD (Charge Coupled Devices) camera that picks up an image of a person detection target region from above. In this embodiment, a fisheye camera is used to image a wide viewing angle region, but there is no restriction on the size of the viewing angle. Moreover, it is good also as installing in a side instead of installing a camera upwards.

  The display 3 is a display device that displays detection results by the person detection unit 16 in the form of characters or graphs. In the present embodiment, the detection result by the person detection unit 16 is displayed on the display 3, but data related to the detection result may be transmitted to another device (for example, a personal computer).

  The control unit 11 is a processing unit that performs processing for detecting a person based on fine movement and edges detected from an image captured by the camera 2. The image input unit 11 is a processing unit that performs processing for receiving an image captured by the camera 2 and storing the image as the input image 21 in the storage unit 20. The image input unit 11 updates the input image 21 as needed using the image received from the camera 2. In addition, when the operation of the person detection device 1 is started, the input image 21 generated first is used as the background image 22.

  The setting unit 12 performs a process of notifying the background difference region detection unit 13 of a definition region that is a region defined in advance according to an operation place or the like. The definition area is expressed as a set of closed areas. For example, the definition area is area information in which the entire path shown in the input image 21 is divided without gaps, or the arrangement positions of ATM corners are divided at equal intervals. It is represented as a set of determination areas to be detected as “features” and “pixel features”. Since areas not included in this definition area are not subject to person detection, the processing load can be reduced by omitting image processing related to parts that do not need to be detected (for example, walls or tables where no person exists). Can be reduced.

  The background difference area detection unit 13 detects a background difference area by taking a difference between the input image 21 and the background image 22 and detects an overlapping area between the detected background difference area and the definition area by the setting unit 12 as a detection target area. 23 is a processing unit that performs processing to be stored as 23 in the storage unit 20. In this embodiment, the definition information by the setting unit 12 is used. However, the detected background difference area may be used as the detection target area 23 without using the definition information.

  The fine movement detection unit 14 is a processing unit that performs a process of detecting the fine movement in which the above-described “pixel feature” appears from the detection target region 23. Further, the detection result by the fine motion detection unit 14 is passed to the person detection unit 16. Here, the detail of the process in which the fine movement detection part 14 detects a fine movement is demonstrated using FIG. 3 and FIG. FIG. 3 is a diagram showing an outline of the fine motion detection process, and FIG. 4 is a diagram showing a modification of the fine motion detection process.

  The upper part of FIG. 3 is a graph showing a change in luminance value of a predetermined pixel in the input image 21. In this graph, the vertical axis represents luminance values (0 to 255), and the horizontal axis represents time. Here, the horizontal axis indicates the past time as it goes to the left in the figure with reference to the present. Also, the lower part of the figure shows a graph showing a fine movement detection / non-detection state.

  The “condition A” illustrated in FIG. 3 indicates a condition that the difference between the “average luminance value for the past T1 seconds” and the “input luminance value” is “Th” or more. Here, when this condition A is expressed by an expression, expression 1 “average luminance value in the past T1 seconds−Th> input luminance value” or expression 2 “average luminance value in the past T1 seconds + Th” <input luminance value ” It will be the case that either expression is satisfied. The case where Expression 1 is satisfied is the case where the input value satisfies the condition A on the darker side than the average value, and the case where the input value is satisfied satisfies the condition A on the side where the input value is brighter than the average value. Is the case. The value of the time “T1” is, for example, “5 seconds”, and the value of the predetermined value “Th” is “6” in order to detect a person wearing clothes of the same color as the floor. , Respectively.

  “Condition B” refers to a condition “when the condition A is alternately repeated three times within T2 seconds, the detection state is maintained for T3 seconds”. However, when the period in which the brightness varies is substantially constant, that is, when “Ta” and “Tb” shown in the figure satisfy the expression “| Ta−Tb | ≦ 0.2 seconds”, the rope This is excluded because it is considered to be a periodic luminance change caused by the fluctuation of the image.

  In the condition B, the detection state is maintained for a predetermined time, particularly in the case of a stationary person, such a fine movement occurs only sporadically. Therefore, in order to determine that a person exists, the following fine movement is performed. This is because it is necessary to maintain the detection state until it occurs. The value of time “T2” is, for example, “5 seconds” as the maximum time during which reciprocation of fine movement can be detected, and the value of time “T3” is that some part of the human body moves slightly. For example, “10 seconds” is selected as each possible time interval. In the present embodiment, the condition A is repeated three times alternately in the light and dark in the “condition B”, but may be changed to another number.

  As shown in FIG. 3, when either condition A or condition B is satisfied, the fine movement is detected, and when neither condition A or condition B is satisfied, the fine movement is not detected. . In this way, the presence or absence of fine movement is detected for each pixel in the input image 21. If there is even one pixel in which fine movement is detected in the above-described pixel unit, it is determined that the pixel unit has a pixel feature.

  In the fine movement detection process described above, the presence / absence of fine movement is determined after excluding the pixels whose luminance value changes periodically, but the presence / absence of fine movement is determined without excluding the pixels whose luminance value changes periodically. It is good to do. For example, when the number of pixels determined to have fine movement is larger than the number of pixels whose luminance value periodically changes, the pixel unit may be determined to have pixel characteristics.

  By the way, in FIG. 3, the fine movement is detected based on the binary values of the fine movement detection state and the fine movement non-detection state, but a state called a holding state may be added. Therefore, a modified example in which a hold state is added will be described with reference to FIG.

  The upper part of FIG. 4 is a graph showing a change in the luminance value of a predetermined pixel in the input image 21, which is the same as FIG. On the other hand, “pending state” is added to the graph showing the fine movement detection / non-detection state shown in the lower part of FIG. As shown in FIG. 4, when the condition B is not satisfied and the condition A is satisfied, the “holding state” is set, and when the condition B is satisfied, the “detection state” is set.

  In other words, the “pending state” refers to a case where it is not yet clear whether the person is “person” or “background” (that is, the condition A is satisfied but the condition B is not satisfied). Since the illumination variation does not repeat the condition A alternately between light and dark, the condition B is not satisfied. Therefore, by providing the “hold state”, it is possible to effectively prevent erroneous detection of fine movement immediately after the occurrence of illumination fluctuation.

  Returning to FIG. 2, the edge detection unit 14 will be described. The edge detection unit 14 is a processing unit that performs processing for detecting an edge where the above-described “edge feature” appears from the detection target region 23. Here, the edge detection unit 14 detects edges using, for example, a Laplacian filter, and uses a Sobel filter to detect the direction of the edges. Then, an edge that is stationary for a certain time (for example, stationary for 30 seconds) or an edge that is continuous for a certain number of pixels in the same direction is removed, and the remaining edge is defined as an edge in which an “edge feature” appears. The detection result by the edge detection unit 14 is passed to the person detection unit 16.

  The person detection unit 16 is a processing unit that performs processing for detecting the presence or absence of a person in the detection target region 23 based on the detection result by the fine movement detection unit 14 and the detection result by the edge detection unit 15. Here, an outline of the person detection processing by the person detection unit 16 will be described with reference to FIG. FIG. 5 is a diagram showing an outline of the person detection process. The background difference image shown in the figure includes a plurality of detection target areas 23 (see the areas labeled “A” to “H” in the figure).

  For example, the detection target area 23 of “G” shown in FIG. 3 is expressed as a set of a plurality of “pixel units”. Each pixel unit is composed of, for example, 8 pixels × 8 pixels. Here, each pixel is not set as a minimum unit for detection processing, but a pixel unit composed of a plurality of pixels is set as a minimum unit, as shown in FIG. This is because even if they do not appear in the same pixel, a region where both features appear is detected.

  For example, if the number of pixels detected as having a pixel feature is one in the pixel unit, the person detection unit 16 sets this pixel unit as the pixel unit having the pixel feature. In addition, for example, if there are three or more pixels detected as having edge features in the pixel unit, the person detection unit 16 sets this pixel unit as the pixel unit having the edge feature. Here, the reason why the number of pixel units is three or more is that a case where an edge related to a stationary object that should have been removed remains as noise is considered.

  Then, the person detection unit 16 sets a pixel unit determined to have a pixel feature and an edge feature as a pixel unit in which both features appear. Then, the same processing is repeated for all the pixel units included in each detection target area 23, and when the number of pixel units in which both features appear is equal to or greater than a predetermined value, a person is present in the detection target area 23. judge.

  In this way, the person detection unit 16 determines the presence / absence of a person in each detection target area 23 and outputs it to the display 3. In this embodiment, it is determined that there is a person in the detection target area 23 in which the number of pixel units in which both the pixel feature and the edge feature appear is equal to or greater than a predetermined value. However, the pixel unit in which both features appear It is also possible to determine that there is a person on the condition that the ratio obtained by dividing the number of pixels by the number of all pixel units included in the detection target area 23 is equal to or greater than a predetermined value.

  Returning to the description of FIG. 2, the background image update unit 17 will be described. The background image update unit 17 is a processing unit that performs a process in which the person detection unit 16 updates the background image 22 using the detection target region 23 that is not the region where the person exists (person region). In this manner, by updating the background image 22, even if there is a change in the background due to a change in the position of the matrix arrangement rope or a change in the position of the stationary object, the background image 22 changes due to factors other than these persons. By excluding the area where the occurrence of the error occurs from the background difference area, it is possible to detect a person with high background accuracy. The background image update unit 17 also performs a process of canceling the update of the background image 22 in order to deal with an erroneous update of the background image 22. Here, an overview of the background image update processing will be described with reference to FIG.

  FIG. 6 is a diagram showing an outline of the background image update process. As shown in the figure, the background image update unit 17 has two generations of background images 22 of “current background image” representing the current background image and “previous background image” representing the immediately preceding background image as the background image 22. Refer to It is assumed that the “current background image” and the “previous background image” are stored in the storage unit 20 as the background image 22.

  As shown in the figure, “Comparison 1” is performed to compare the input image and the current background image, and the person detection unit 16 uses the detection target area 23 that is not the area where the person exists (person area). “Update 1” for updating the background image 22 is executed. On the other hand, “Comparison 2” for comparing the input image and the foreground / background image is also performed together with “Comparison 1” described above.

  Then, three conditions such as “a person is not detected in the input image”, “the input image and the current background image are substantially the same”, and “the input image and the foreground / background image are different” are set for 10 seconds, for example. If it is satisfied, “Update 2” is executed with the current background image as the foreground image. This is because it is unnatural that the input image differs from the foreground / background image even though there is no person in the input image.

  On the other hand, the three conditions of “a person is detected in the input image”, “the input image and the current background image are substantially the same”, and “the input image and the foreground / background image are different” are satisfied for 5 seconds, for example. In such a case, a “revival” process is executed to cancel the update by replacing the current background image on which “update 1” has been executed with the previous background image. As described above, since the erroneous update can be canceled, it is possible to return to a normal background image even when a person is captured in the background image.

  The storage unit 20 is a storage unit configured by a storage device such as an HDD (Hard Disk Drive) or a RAM (Random Access Memory). The storage unit 20 stores an input image 21, a background image 22, and a detection target area 23. The input image 21 is a captured image by the camera 2 that is updated as needed by the image input unit 11. The background image 22 is an input image 21 captured without a person, or an image that is updated as needed by the background image update unit 17. As described above, the background image 22 includes the current background image and the foreground background image. The detection target area 23 is a set of person detection areas determined by the background difference area detection unit 13 (see “A” to “H” in FIG. 5).

  Next, a definition area, which is an area defined in advance according to the operation location or the like, will be described with reference to FIG. FIG. 7 is a diagram illustrating an example of the definition area. As shown in (1) of the figure, as the definition area, each definition area can be a position where persons are lined up at equal intervals (for example, 60 cm intervals) in a passage partitioned by a rope. Moreover, as shown in (2) of the same figure, it is good also as dividing | segmenting the whole channel | path divided by the rope without gap, and setting it as each definition area | region.

  Note that areas that are not included in this definition area are not subject to human detection, so human detection can be achieved by omitting image processing for parts that do not require human detection (for example, walls and tables where no human exists). The processing load accompanying processing can be reduced. In addition, the definition area can be flexibly set according to the installation location of the person detection device 1.

  Next, a case where the person detection unit 16 processes a person detection result into a graph or the like and outputs the result to the display 3 will be described with reference to FIG. FIG. 8 is a diagram illustrating an example of a stay time density diagram. In FIG. 81, a count area is set around the product display shelf. In FIG. 82, a stay time density diagram representing the number of visitors per unit time around the product display shelf is shown. Show.

  Here, the counting area shown in FIG. 81 may be the same as the definition area shown in FIG. 7, or may be newly set as an area different from the definition area shown in FIG. For example, each determination area shown in “(2) Definition area example 2” in FIG. 7 is set as an area that does not overlap each other, but each count area may be set so as to straddle each determination area. it can.

  When the person detection unit 16 detects a person in each counting area indicated by 81 in the figure, the number of detected persons is counted up. Here, counting up the number of people is performed by measuring the time and number of people staying in each counting area, taking into account the number of people entering and leaving each counting area. Then, the number of staying people at each time is associated with each counting area, for example, stored in the storage unit 20, and graphed over a specified time range, thereby obtaining a staying time density diagram 82 shown in FIG. It is done.

  In the stay time density diagram 82 shown in the figure, it is shown that the stay time density is high in the left part of the upper display shelf, the central part sandwiched between both display shelves, and the like. By generating such a stay time density diagram, it becomes possible to analyze the popularity and the attention level of the products arranged on the display shelf. In FIG. 8, merchandise display shelves are counted, but booths at exhibitions and merchandise presentations, attraction queues, and the like may be counted.

  Next, the processing procedure of the fine motion detection process performed by the fine motion detection unit 14 will be described with reference to FIG. FIG. 9 is a flowchart showing the processing procedure of the fine motion detection processing. Here, the processing target in the figure is each pixel in the input image 21, and the processing shown in the figure is performed in parallel with each update of the input image 21 (for each frame) for each pixel included in the detection target region 23. Done. The processing procedure shown in FIG. 9 corresponds to FIG.

  As shown in the figure, it is determined whether or not the difference between the average luminance value for T1 seconds and the input luminance value is equal to or greater than Th (step S101). If the difference is equal to or greater than Th (step S101, Yes), a fine movement detection state is set (step S102). In step S101, when the difference is less than Th (step S101, No), the process ends.

  Subsequently, it is determined whether or not the fine motion detection state has been detected by repeating a predetermined number of times alternately within T2 seconds (step S103), and if the determination condition is satisfied (step S103, Yes), the light-dark fluctuation cycle is determined. It is determined whether or not it is substantially constant (step S104). If the determination condition in step S103 is not satisfied (No in step S103), the process is terminated.

  If the determination condition of step S104 is not satisfied (step S104, No), a fine movement detection state holding command is issued for T3 seconds (step S105), and the process ends. Here, for the T3 second hold command, for example, FLG is raised, the FLG is processed by a separate timer, and turned off after T3 seconds. If a hold command is generated again within T3 seconds, the timer is once initialized, and T3 seconds are measured again from that point. If the determination condition of step S104 is satisfied (step S104, Yes), the process ends without performing the process of step S105.

  Next, the procedure of the person detection process performed by the person detection unit 16 will be described with reference to FIG. FIG. 10 is a flowchart showing the processing procedure of the person detection processing. The processing target in the figure is a pixel unit composed of a plurality of pixels (see “pixel unit” in FIG. 5), and the processing shown in the figure is performed in each detection target region 23 (“A” to “A” in FIG. H ”).

  As shown in the figure, the input image 21 corresponding to the specific detection target area 23 is moved to a predetermined pixel unit (step S201). Subsequently, it is determined whether or not fine movement (pixel feature) is detected by the fine movement detection unit 14 for the pixel unit (step S202). If fine movement (pixel feature) is detected (step S202, Yes), It is determined whether or not an edge (edge feature) has been detected by the edge detection unit 15 for the pixel unit (step S203). If fine movement (pixel feature) is not detected in step S202 (No in step S202), the process after step S201 is repeated after moving to another pixel unit.

  If an edge (edge feature) is detected in step S203 (step S203, Yes), the pixel unit is counted (step S204). On the other hand, when an edge (edge feature) is not detected in step S203 (step S203, No), the process after step S201 is repeated after moving to another pixel unit.

  Subsequently, it is determined whether or not the determination is completed for all the pixel units (step S205). If the determination condition of step S205 is satisfied (step S205, Yes), the counted pixel unit is equal to or greater than a predetermined value. It is determined whether or not there is (step S206). On the other hand, when the determination condition of step S205 is not satisfied (No at step S205), the process after step S201 is repeated after re-selecting another detection target area 23.

  If the determination condition of step S206 is satisfied (step S206, Yes), it is determined that person detection is performed for the detection target area 23 (step S207), and the process ends. On the other hand, when the determination condition of step S206 is not satisfied (step S206, No), it is determined that no person is detected in the detection target area 23 (step S208), and the process is terminated.

  As described above, in this embodiment, the fine movement detection unit detects fine movement of luminance change for each pixel unit including one or a plurality of pixels in the input image, and the edge detection unit detects an edge from the input image. The person detection device is configured such that the person detection unit detects a person based on the fine movement detected by the fine movement detection unit and the edge detected by the edge detection unit.

  Therefore, even if there is illumination fluctuation or background fluctuation, the person is detected by combining the fine movement that tends to appear in the outline of the person including the stationary person and the edge that appears in the outline of the object including the person. Can be determined with high accuracy.

  As described above, the person detection apparatus according to the present invention is suitable for a case where it is desired to detect a person with high accuracy using the background subtraction method. The present invention can also be applied to queue counting in waiting for a cash register or outdoor attraction, vacant seat detection such as vacant seat display, seat occupancy rate calculation, etc., and analysis of the visitor stay status in front of a product display shelf or in front of an exhibition booth.

It is a figure which shows the characteristic of the person detection apparatus which concerns on a present Example. It is a figure which shows the structure of the person detection apparatus which concerns on a present Example. It is a figure which shows the outline | summary of a fine motion detection process. It is a figure which shows the modification of a fine motion detection process. It is a figure which shows the outline | summary of a person detection process. It is a figure which shows the outline | summary of a background image update process. It is a figure which shows an example of a definition area | region. It is a figure which shows an example of a stay time density diagram. It is a flowchart which shows the process sequence of a fine motion detection process. It is a flowchart which shows the process sequence of a person detection process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Person detection apparatus 2 Camera 3 Display 10 Control part 11 Image input part 12 Setting part 13 Background difference area | region detection part 14 Fine motion detection part 15 Edge detection part 16 Person detection part 17 Background image update part 20 Storage part 21 Input image 22 Background image 23 Detection target area

Claims (10)

A person detection device that detects a person based on a difference between an input image and a background image,
For each pixel unit consisting of one or a plurality of pixels in the input image , there is an alternating repetition of light and dark based on the difference between the average luminance value of the past first predetermined period and the current input luminance value, and within the second predetermined period And the fine movement detecting means for detecting the fine movement of the luminance change due to the repetition of this repetition a predetermined number of times and the period of the brightness fluctuation being not substantially constant ,
Edge detection means for detecting edges from the input image and excluding edges that are stationary for a third predetermined period and edges that are continuous for a certain number of pixels in the same direction ;
A person detection apparatus comprising: person detection means for detecting a person based on the pixel unit in which the fine movement is detected by the fine movement detection means and the edge is detected by the edge detection means . A person detection device that detects a person based on a difference between an input image and a background image,
Setting means for setting a detection target area, which is a target detection area, on the input image;
For each pixel unit composed of one or a plurality of pixels in the detection target region set by the setting means , alternating light and dark are repeated based on the difference between the past average luminance value in the first predetermined period and the current input luminance value. There is a fine movement detecting means for detecting the fine movement when the repetition is repeated a predetermined number of times within the second predetermined period and the period of the light / dark fluctuation is not substantially constant ,
Human detection means for detecting the detection target area as a person area where a person exists when the number of pixel units in which the fine movement is detected by the fine movement detection means is equal to or greater than a predetermined threshold. A person detection apparatus characterized by the above. The setting means includes
The person detection apparatus according to claim 2, wherein a background difference area, which is an area where the input image and the background image are different, is set as the detection target area. The setting means includes
The person detection apparatus according to claim 2, wherein a definition area, which is an area defined in advance according to an operation place, is set as the detection target area. An edge detecting means for detecting an edge from the input image and excluding an edge that is stationary for a third predetermined period and an edge that is continuous for a certain number of pixels in the same direction ;
The person detecting means is
The detection target area is detected as the person area when the fine movement is detected by the fine movement detection unit and the number of the pixel units in which the edge is detected by the edge detection unit is equal to or larger than a predetermined threshold. The person detection device according to claim 2, 3, or 4. The update apparatus which updates the said background image by taking in into the said background image the non-detection area | region which is the said detection object area | region which the said person detection means did not detect as the said person area | region. The person detection device according to any one of? Canceling the update of the background image by the background image update means on condition that the person detection means detects the non-detection area as the person area within a predetermined time after the update means updates the background image The person detecting apparatus according to claim 6, further comprising update canceling means. A person detection device that detects a person based on a difference between an input image and a background image,
A setting means for setting a definition area, which is an area defined in advance according to the operation location, as a detection target area that is a target area of person detection;
For each pixel unit composed of one or a plurality of pixels in the detection target region set by the setting means , alternating light and dark are repeated based on the difference between the past average luminance value in the first predetermined period and the current input luminance value. There is a fine movement detecting means for detecting the fine movement of the luminance change when the repetition is repeated a predetermined number of times within the second predetermined period and the period of the brightness fluctuation is not substantially constant ,
Edge detection means for detecting edges from the input image and excluding edges that are stationary for a third predetermined period and edges that are continuous for a certain number of pixels in the same direction;
Counting means for counting the number of the detection target areas in which the number of pixel units in which the fine movement is detected by the fine movement detection means and the edge is detected by the edge detection means are equal to or larger than a predetermined threshold as the number of persons; A person detection apparatus comprising: A person detection device that detects a person based on a difference between an input image and a background image,
A setting means for setting a definition area, which is an area defined in advance according to the operation location, as a detection target area, which is a target area for person detection, and setting a number counting area, which is a target for counting the number of people,
For each pixel unit composed of one or a plurality of pixels in the detection target region set by the setting means , alternating light and dark are repeated based on the difference between the past average luminance value in the first predetermined period and the current input luminance value. There is a fine movement detecting means for detecting the fine movement of the luminance change when the repetition is repeated a predetermined number of times within the second predetermined period and the period of the brightness fluctuation is not substantially constant ,
Edge detection means for detecting edges from the input image and excluding edges that are stationary for a third predetermined period and edges that are continuous for a certain number of pixels in the same direction;
Counting means for counting the number of the person counting areas in which the number of pixel units in which the fine movement is detected by the fine movement detecting means and the edge is detected by the edge detecting means is equal to or greater than a predetermined threshold as the number of persons; A person detection apparatus comprising: Storage means for storing the number of persons counted by the counting means and the area in which the number of persons is counted in association with each other in time series;
The person detecting apparatus according to claim 8, further comprising: generating means for generating a person staying time density diagram based on the number of people in time series stored in the storage means.

Images