Differentiating people in a crowd in an image

Abstract

The present disclosure relates to a method of distinguishing a person in a crowd in an image. A person distinguishing system identifies a detected object classified as a person in an image derived from a thermal camera adapted to capture a scene at a substantially forward looking angle, further identifies at least a first grouping of adjacent pixels in the image having an intensity within a predefinable intensity range that is not included in the detected person, determines a grouped pixel area of the at least first grouping in the image, determines an expected pixel area of a person head at at least a first vertical position in the image based on head size reference data for the at least first vertical position, compares at least a portion of the grouped pixel area to the expected head pixel area for the at least first vertical position, determines that the at least first grouping comprises at least a first overlapping person when at least a first comparison resulting from the comparison exceeds a predefinable compliance threshold. The present disclosure further relates to a person distinguishing system, a thermal camera, a computer program product and a non-transitory computer readable storage medium.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to distinguishing people in a crowd in an image. BACKGROUND

[0002] For various reasons, it can be desirable and / or of interest to attempt to count people appearing in an image. The image can be of any arbitrary scene in which a camera, e.g., a surveillance camera, has captured an image of people in, e.g., a room, an open area, and / or an open space.

[0003] For example, computer vision techniques can provide an estimate of the number of people in an image. A moving window-like detector can be used to identify people in the image and count how many are in it, which can require a well-trained classifier capable of extracting low-level features for detection. However, while these techniques can work well at detecting faces, they typically do not perform well on crowded images in which some of the target objects, here people, are not clearly visible. SUMMARY

[0004] It is therefore an object of embodiments herein to provide a method of distinguishing people in a crowd in an image in an improved and / or alternative way.

[0005] The above objects can be achieved by the subject matter disclosed herein. Embodiments are set forth in the appended claims, below description and attached drawings.

[0006] The disclosed subject matter relates to a method, performed by a person distinguishing system, for distinguishing people in a crowd in an image. The person distinguishing system identifies one or more detected objects classified as people in an image derived from a thermal camera adapted to capture a scene at a substantially forward-looking angle. The person distinguishing system further identifies at least a first group of contiguous pixels in the image having an intensity within a predefinable intensity range that is not included in the one or more detected people. In addition, the person distinguishing system determines at least a first group of group pixel areas in the image. Moreover, the person distinguishing system determines, based on head size reference data, an expected pixel area of a person head at at least a first vertical position in the image for the at least first vertical position. The person distinguishing system further compares at least a portion of the at least first group of pixel areas with the expected head pixel area for the at least first vertical position. In addition, the person distinguishing system determines that the at least first group comprises at least a first person when at least a first comparison resulting from said comparison exceeds a predefinable conformity threshold.

[0007] The disclosed subject matter also relates to a person distinguishing system for distinguishing a person in a crowd in an image. The person distinguishing system comprises a person identifying unit for and / or adapted to identify one or more detected objects classified as a person in an image derived from a thermal camera, the thermal camera being adapted to capture a scene at a substantially forward looking angle. The person distinguishing system further relates to a grouping identifying unit for and / or adapted to identify at least a first group of adjacent pixels in the image having an intensity within a predefinable intensity range not comprised in the one or more detected persons. Furthermore, the person distinguishing system comprises a region determining unit for and / or adapted to determine a region of the group pixels at at least a first vertical position of a head of the person. The person distinguishing system further comprises an expected region determining unit for and / or adapted to determine an expected region of the head of the person at the at least first vertical position based on head size reference data for the at least first vertical position in the image. Furthermore, the person distinguishing system comprises a comparison unit for and / or adapted to compare at least a portion of the group of pixels region with the expected head pixel region for the at least first vertical position. Moreover, the person distinguishing system comprises a conformity determining unit for and / or adapted to determine that the at least first group comprises at least a first person when at least a first comparison resulting from said comparison exceeds a predefinable conformity threshold.

[0008] Furthermore, the disclosed subject matter relates to a thermal camera comprising the person distinguishing system as described herein.

[0009] Furthermore, the disclosed subject matter relates to a computer program product comprising a computer program stored on a computer readable medium or carrier, the computer program comprising computer program code means which are arranged, when the computer program is run on a computer or processor, to implement the steps of the person distinguishing system as described herein.

[0010] The disclosed subject matter also relates to a non-transitory computer readable storage medium having stored thereon the computer program product.

[0011] Thus, a method is introduced that enables distinguishing a person from another in a crowd packed image. That is, as a result of identifying one or more detected objects classified as a person in an image derived from a thermal camera adapted to capture a scene at a substantially forward looking angle, the objects detected in the thermal camera image and classified as a person can be confirmed. Using a thermal camera also enables mitigating potential issues of light conditions such as shadows, backlight, darkness, camouflaged objects, and the like. Moreover, that is, as a result of identifying at least a first group of contiguous pixels in the image having an intensity within a predefinable intensity range that is not included in the one or more detected persons, a region of contiguous pixels in the image having a certain intensity (e.g., corresponding to a person's intensity) can be confirmed (excluding the detected persons) thereby indicating a possible presence of a person in the at least first group. Furthermore, that is, as a result of determining a group pixel region of at least a first group in the image, a distribution of regions of contiguous pixels identified as being within a predefinable range is established. Moreover, that is, as a result of determining an expected pixel region of a person's head at at least a first vertical position in the image based on head size reference data, a respective expected pixel region form and size of a person's head is established for one or more vertical positions in the image by cross-referencing the head size reference data. That is, an exemplary low value vertical position can indicate a proximity to the thermal camera, and thus an expected pixel region of a person's head at the exemplary low value vertical position can be larger in size than an exemplary higher value vertical position. Accordingly, the head size reference data comprises a mapping between one or more vertical positions and a corresponding expected size and form of a person's head in terms of an expected pixel region for the respective one or more vertical positions, thereby enabling determining a respective expected head size at a respective vertical position. Furthermore, that is, as a result of comparing at least a portion of the group pixel region to the expected head pixel region for at least a first vertical position, a selected region of pixels of the group pixel region is compared to a respective expected pixel region of a person's head for one or more vertical positions. Moreover, that is, as a result of determining that at least a first group comprises at least a first overlapping person when at least a first comparison resulting from said comparison exceeds a predefinable threshold of compliance, which sets a level of minimum compliance between a region of the group pixel region (e.g., at least a portion resembling a region of a person's head) and an expected head size, a determination can be made that at least a first group in the image comprises one or more overlapping persons, assuming that the comparison activity results in one or more exceedances of the threshold. Accordingly, persons that are indistinguishable and / or difficult to distinguish by, for example, commonly known object detection, can be distinguished in a crowded image by the introduced concept.

[0012] Thus, a method of distinguishing a person in a crowd in an image is provided in an improved and / or alternative manner.

[0013] The technical features of the above-mentioned method and corresponding advantages will be discussed in further detail below. BRIEF DESCRIPTION OF DRAWINGS

[0014] The respective aspects of the non-limiting embodiments, including particular features and advantages, will readily be appreciated from the following detailed description and drawings, in which:

[0015] Fig. 1-2 a schematic diagram illustrating an exemplary person distinguishing system according to embodiments of the present disclosure is shown;

[0016] Fig. 3 is a schematic block diagram illustrating an exemplary person distinguishing system according to embodiments of the present disclosure; and

[0017] Fig. 4 is a flowchart illustrating an exemplary method performed by a person distinguishing system according to embodiments of the present disclosure. DETAILED DESCRIPTION

[0018] Non-limiting embodiments of the present disclosure will now be described more fully hereinafter with reference to the accompanying drawings, in which current preferred embodiments of the present disclosure are shown. The present disclosure may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Like reference numerals refer to like elements throughout. The dashed lines of certain blocks in the figures indicate that these units or actions are optional, not mandatory.

[0019] In the following, according to embodiments herein relating to distinguishing a person in a crowd in an image, a method will be disclosed that enables distinguishing a person from another person in a crowded image of a crowd of people.

[0020] Reference will now be made to the drawings and, in particular, to Fig. 1-2 a schematic diagram illustrating an exemplary person distinguishing system 1 according to embodiments of the present disclosure is depicted. The person distinguishing system 1 is adapted and / or configured for distinguishing a person in a crowd in an image.

[0021] The “person distinguishing system” can refer to a “person detection system”, a “unique head determination system” and / or just an “evaluation system” or “control system”, while the phrase “for distinguishing a person in a crowd in an image” can refer to “for distinguishing a unique person or person head in a crowd in an image”, “for distinguishing persons from each other in close proximity and / or at least partially overlapping in an image”, “for distinguishing a person from another person at least partially overlapping with said person in an image” and / or “for distinguishing persons in a crowded image”. On the other hand, the term “crowd” can refer to “group of persons”.

[0022] The person distinguishing system 1 can be associated with a thermal camera 10 Fig. 3The thermal camera 10 is associated with, and optionally comprised in, a surveillance system (not shown) as will be described. Thus, by using the thermal camera 10, potential problems of light conditions such as shadows, backlight, darkness, camouflaged objects, etc. can be mitigated.

[0023] The thermal camera 10 can refer to any arbitrary (e.g. known) thermal camera supporting the creation of images based at least partly on the heat continuously radiating from all objects including humans. The thermal camera 10, which can be considered a combined visible and infrared (IR) camera, can also refer to any arbitrary camera supporting the creation of images using IR radiation in combination with visible light (i.e. being sensitive to at least a part of the IR range, e.g. the mid- wavelength infrared (MWIR) and / or long-wavelength infrared (LWIR) band, and to at least a part of the visible light range).

[0024] According to examples, the thermal camera 10 can be comprised in a thermal camera arrangement (not shown) comprising additional components, e.g. being part of an exemplary surveillance system (not shown). Thus, according to examples, the phrase "thermal camera comprising a person differentiation system" can refer to "thermal camera arrangement comprising a person differentiation system" and / or "thermal camera of a surveillance system comprising a person differentiation system".

[0025] The person differentiation system 1 (e.g. by means of the person identification unit 101 (shown in Fig. 3 and further described) is adapted and / or configured for identifying one or more detected objects 2 classified as humans in an image 3 derived from the thermal camera 10 adapted to capture a scene in a substantially forward looking angle. Thereby, objects detected in the thermal camera image 3 and classified as humans 2 can be confirmed. In the exemplary Fig. 1 arrangement, a first detected human 21 and a second detected human 22 are identified in an exemplary manner.

[0026] The objects 2 detected and classified as humans can (or have been) detected and classified in any arbitrary (e.g. known) manner, e.g. using a trained Histogram-of-Gradients (HoG) algorithm (e.g. under the support of well-known computer vision techniques, image processing, object detection and / or classifiers). The number of detected objects classified as humans 2 can naturally vary from image 3 to image 3 and can range up to tens, hundreds or even thousands of humans 2. However, it should be noted that it can also occur that there are no detectable and / or identified detected objects classified as humans 2 in an image 3, why a detected object classified as a human 2 can be considered a "potential" detected object classified as a human 2. On the other hand, the term "image" can refer to any arbitrary camera captured image of any arbitrary size (e.g. in pixels) and / or of any arbitrary image quality.

[0027] The thermal camera 10 is adapted and / or configured to capture a scene in a substantially front view angle which can be represented by any arbitrary position of the thermal camera 10 in which the thermal camera 19 can capture potential persons in the scene from a substantially side view and / or from a slightly downward tilted view opposite to e.g. a top view. Thus, the substantially front view angle can be represented by the thermal camera 10 capturing the scene in a position substantially considered to be the eye level of a person and / or from a position slightly above it, e.g. indicating a range of positions of the thermal camera 10 from less than one meter up to several meters from the floor and / or floor level. According to an example, the phrase "front view angle" can refer to "slightly above front view angle", thus indicating that the thermal camera 10 can be adapted to capture the scene (and subsequently potential persons thereof) from a slightly downward tilted and / or inclined position, e.g. ranging from above zero up to 45 degrees. On the other hand, the scene can be represented by any arbitrary surrounding, e.g. a room, an open area, an open space, etc.

[0028] The phrase "identifying one or more detected objects classified as persons" can refer to "confirming, classifying, filtering out, blocking and / or blacklisting one or more detected objects classified as persons". According to an example, the phrase can further refer to "counting one or more detected objects classified as persons" and / or "classifying one or more detected objects as persons". On the other hand, "detected objects" can refer to "potential detected objects" and / or just "objects", while "images" can refer to "camera images", "thermal camera images", "combined temperature and visible light images" and / or "image frames". The phrase "images derived from a thermal camera" can refer to "images obtained and / or extracted from a thermal camera" and / or just "images from and / or captured by a thermal camera". On the other hand, "in a substantially front view angle" can refer to "substantially in the level of a human eye and / or slightly downward tilted" and / or just "in a front view angle and / or slightly downward tilted". Furthermore, "scene" can refer to "surrounding", while "thermal camera adapted to capture a scene" can refer to "thermal camera adapted to capture images of a scene". According to an example, the phrase "in images derived from a thermal camera adapted to capture a scene in a substantially front view angle" can refer to "in images of a scene in a substantially front view angle derived from a thermal camera".

[0029] As Fig. 2 Further shown, the person differentiation system 1 is e.g. by means of the grouping identification unit 102 (in Fig. 3At least a first group 4 of adjacent pixels in the image 3 (as shown and further described in the exemplary embodiment) is adapted and / or configured to identify adjacent pixels in the image 3 having an intensity in a predetermined intensity range, which are not comprised in the one or more detected persons 2. Thereby, in addition to the already detected persons 2, an area of adjacent pixels 4 in the image 3 having a certain intensity (e.g. corresponding to the intensity of a person) can be confirmed, indicating that a person can be present in at least the first group 4. In the exemplary embodiment, the first group 4 is identified in an exemplary manner. However, it should be noted that in other exemplary images (not shown), further separated groups of adjacent pixels can be naturally identified and / or identifiable. Fig. 1

[0030] Identifying at least the first group 4 of adjacent pixels can be done in any arbitrary (e.g. known) manner, e.g. with the support of computer vision techniques and / or image processing. I.e. since the relative differences in intensity of infrared energy reflected or omitted from different objects can be detected and displayed in the image 3, a predetermined intensity range (e.g. comprising intensities corresponding to the intensity of a person and / or a person’s head) can be identified. The intensity range can refer to any range which can represent the intensity and / or thermal signature of a person and / or a person’s head. Further, the intensity range can be e.g. pre-stored such as in the person differentiation system 1, in the thermal camera 10, on a remote server, etc.

[0031] However, optionally, the intensity range can be based on an intensity measurement 211, 221 of at least one of the one or more detected persons 2 (e.g. in the exemplary embodiment, the exemplary first detected person 21 and the second detected person 21). Thus, the one or more intensities of the one or more already detected persons 2 can constitute the basis for defining the intensity range, e.g. the intensity range substantially equals or is close to said intensity measurement 221, 221 and / or a statistical intensity measurement variation thereof. “Based on the intensity measurement” can refer to “derived from the intensity measurement”, “based on a statistical intensity measurement” and / or “based on an intensity value”. Fig. 1

[0032] ​​The phrase "identifying at least first group of neighboring pixels" can refer to "identifying at least first group of substantially neighboring pixels" and / or "confirming, selecting, classifying and / or filtering out at least first group of neighboring pixels". On the other hand, "group" can refer to "region", "area", "distribution" and / or "group of interest", while "neighboring pixels" can refer to "adjacent pixels". Furthermore, the phrase "not included in one or more detected persons" can refer to "not included in one or more detected objects classified as persons", "not including pixels of one or more detected persons" and / or "excluding detected persons". On the other hand, "predetermined intensity range" can refer to "predetermined intensity range corresponding to and / or equivalent to intensity and / or thermal signature of a person and / or a person's head" and / or "only a person's head intensity range".

[0033] The person distinguishing system 1 is further adapted and / or configured to determine a group pixel region 40 of at least a first group 4 in the image 3, e.g. by means of a region determination unit 103 (shown in Fig. 3 and further described). Thereby, a region distribution 40 of neighboring pixels 4 identified as being within the predetermined range is established as outlined in the exemplary Fig. 1 .

[0034] Determining the group pixel region 40 can be done in any arbitrary (e.g. known) way, e.g. with support of computer vision techniques and / or image processing. Furthermore, the group pixel region 40 of neighboring pixels can be of any arbitrary (e.g. irregular) shape and / or form naturally depending on the person and / or the number of persons in the scene or other potential objects having an intensity within the predetermined intensity range. The phrase "determining the group pixel region" can refer to "computing and / or estimating the group pixel region", while on the other hand "pixel region" can refer to "pixel region distribution".

[0035] Furthermore, the person distinguishing system 1 is adapted and / or configured to determine, e.g. by means of an expected region determination unit 104 (shown in Fig. 3 and further described), based on the head size reference data 5, for at least a first vertical position y expected in the image 3, an expected pixel region x expected of a person's head at the at least first vertical position y expectedThus, by referring to the head size reference data 5, for one or more vertical positions y in the image 3, a respective expected pixel area x form and size of a human head is established. That is, an exemplary low value vertical position y can indicate a closer proximity to the thermal camera 10 than an exemplary high value vertical position y, and thus the size of the pixel area x of a human head can be expected to be larger at the exemplary low value vertical position y than at the exemplary high value vertical position y. Accordingly, in terms of the expected pixel area x for the respective one or more vertical positions y, the head size reference data 5 comprises a mapping between the one or more vertical positions y and the corresponding expected size and form of a human head, enabling a determination of a respective expected head size x at the respective vertical position y expected . expected .

[0036] at least a first vertical position y expected may be represented by any arbitrary vertical position y in the image 3. Furthermore, the head size reference data 5 can be represented by any data indicative of a relationship between a vertical position y in the image 3 and a corresponding expected pixel area x or human head. The head size reference data 5 may, for example, be comprised in a data table, further for example, be stored in the person distinguishing system 1, in the thermal camera 10, on a remote server, etc., and for example, also be pre-stored.

[0037] The phrase "determining an expected pixel area" can refer to "deriving, obtaining, calculating and / or interpolating an expected pixel area", and "based on head size reference data" can refer to "based on referring to head size reference data" and / or "based on data indicative of a relationship between a vertical position in the image and a corresponding expected pixel area of a human head". Furthermore, the phrase "head size reference data" can refer to "head pixel area reference data" and / or "head size mapping data", and "data" can refer to "mapping information". On the other hand, an "expected" pixel area can refer to an "expected corresponding" pixel area and / or an "estimated and / or calculated" pixel area.

[0038] Optionally, as Fig. 1As shown, the head size reference data 5 can be based on (for two 21, 22 or more detected persons 2) mapping respective vertical positions yi, y2and sum pixel areas xi, x2of the heads 212, 222 of the detected persons 21, 22 in the image 3. Further optionally, the head size reference data 5 can be based on interpolation from said mapping. Thereby, the first determined head size pixel area xi of the head 212 of the detected person 21 at the first determined vertical position yi and at least the second determined head size pixel area x2of the head 222 of the detected person 22 at the second determined vertical position y2may form the basis of the head size reference data 5. The head size reference data 5 may, for example, as exemplarily depicted in the exemplary coordinate plane in Fig. 1 depend on a linear graph 50 (or substantially linear graph) between the first head size pixel area xi / first vertical position yi and at least the second head size pixel area x2 / second vertical position y2. The mapped vertical positions yi, y2of the detected persons 21, 22 can be selected along respective extensions, i.e. heights of said detected persons 21, 22 deemed appropriate, e.g. as exemplarily illustrated in Fig. 1 the lower end of the respective head 212, 222, alternatively e.g. at the upper end of the respective head 212, 222 (or assuming line of sight level). Further, "mapping" can refer to "associating" and / or "linking", and "based on mapping" can refer to "calculated and / or derived from the mapping of". On the other hand, "vertical position and pixel area" can refer to "vertical head position and vertical head size pixel area".

[0039] As exemplarily illustrated in Fig. 2 Further, the person distinguishing system 1 is adapted and / or configured for comparing at least a portion of the grouping pixel area 40 with an expected head pixel area x expected for at least the first vertical position y expected , e.g. by means of the comparison unit 105 (shown in Fig. 3 and further described). Thereby, a selected region of pixels of the grouping pixel area 40 is compared with a respective expected pixel area x expected of a person head for one or more vertical positions y expected .

[0040] The comparison of the pixel areas can be repeated for any arbitrary number of vertical positions y. Moreover, the comparison activity can be accomplished in any arbitrary (e.g. known) manner, e.g. with the support of computer vision techniques and / or image processing. On the other hand, the at least one portion of the grouped pixel area 40 can be represented by any area of the grouped pixel area 40 and can further have any size, form and / or shape deemed suitable and / or feasible. For example, one or more areas of the grouped pixel area 40 at least somewhat (or a predetermined degree) similar to a human head can be selected for comparison. The phrase "comparing at least one portion" can refer to "comparing at least one area" and / or "comparing one or more selected areas", while the "expected head pixel area" can refer to "expected head size pixel area" and / or "expected pixel area of a human head".

[0041] Moreover, the person distinguishing system 1 is adapted and / or configured to determine that the at least first group 4 comprises at least a first overlapping person 6 if at least a first comparison resulting from said comparison exceeds a predetermined conformity threshold, e.g. by means of a conformity determination unit 106 (shown and further described in Fig. 3 Thereby, assuming that the comparison activity results in one or more exceedances of a threshold set level of minimum conformity between an area of the grouped pixel area 40 (e.g. an area at least somewhat similar to a human head) and the expected head size x, it can be determined that the at least first group 4 in the image 3 comprises one or more overlapping persons 6. Thus, persons that are indistinguishable and / or difficult to distinguish by means of e.g. generally known object detection in a crowded image 3 can be distinguishable by means of the introduced concept.

[0042] In the exemplary Fig. 2 determination can be made that the conformity threshold is exceeded in three exemplary scenarios out of potentially many comparisons; for a comparison of a head-like area (which can subsequently be identified as a first overlapping person 61) with an expected head size pixel area x at a first vertical position y expected expected and in a similar manner, for respective second and third comparisons of respective head-like areas (which can subsequently be identified as a second and third overlapping person 62, 63) with respective expected head size pixel areas x at respective second and third vertical positions y.

[0043] ​The predefinable conformity threshold can be set to any value that is considered suitable and / or feasible, and e.g. indicates a level, degree and / or percentage of match between the area of the grouped pixel area 40 and the expected head size x that needs to be exceeded in order for the at least first group 4 to be considered to comprise an overlapping person 6. The conformity threshold can accordingly e.g. be represented by a match of at least 50%, at least 70% or at least 90%. The phrase "determining that the at least first group comprises" can refer to "estimating and / or considering that the at least first group comprises", and "overlapping person" can refer to "at least partially overlapping person" and / or "overlapping person in the image". On the other hand, "when" can refer to "should" and / or "assuming", and "comparing" can refer to "comparing activity". Furthermore, the phrase "predefinable conformity threshold" can refer to "predefinable similarity threshold", "predefinable head match threshold" and / or just "conformity threshold".

[0044] Optionally, the person distinguishing system 1 can e.g. by means of the optional number estimation unit 107 (shown in and further described in Fig. 3 be adapted and / or configured for estimating a number of overlapping persons 6 in the at least first group 4 based on the number of comparisons exceeding the conformity threshold. Thereby, by counting each time the conformity threshold is exceeded (which can be equivalent to determining how many overlapping persons 6 are comprised in the at least first group 4), the number of overlapping persons 6 in the at least first group 4 can be estimated. In the exemplary Fig. 2 example, three overlapping persons 61, 62, 63 are depicted in an exemplary manner. The phrase "estimating the number" can refer to "determining the number" and / or "calculating the number", and "based on the number of comparisons exceeding the conformity threshold" can refer to "by counting the number of comparisons exceeding the conformity threshold".

[0045] Further optionally, the person distinguishing system 1 can e.g. by means of the optional total number estimation unit 108 (shown in and further described in Fig. 3 be adapted and / or configured for estimating a total number of persons in the image 3 by adding the number of overlapping persons 6 to the one or more detected persons 2. Thereby, a method is provided to support estimating a total number of persons in a crowded image 3. In the exemplary Fig. 1-2 example, two detected persons 21, 22 and three overlapping persons 61, 62, 63 are depicted in an exemplary manner, thus adding up to an exemplary total number of five persons. The phrase "estimating the total number" can refer to "determining the total number" and / or "calculating the total number".

[0046] Further as Fig. 3 shown, Fig. 3is a schematic block diagram illustrating an exemplary person differentiation system 1 according to embodiments of the present disclosure, which can comprise a person identification unit 101, a group identification unit 102, a zone determination unit 103, an expected zone determination unit 104, a comparison unit 105, a compliance determination unit 106, an optional number estimation unit 107 and / or an optional total number estimation unit 108, all of which are described in more detail above. Furthermore, the embodiments herein for differentiating between persons in a crowd (in image 3) can be implemented by one or more processors, such as processor 109, denoted CPU herein, and computer program code for performing the functions and actions of the embodiments herein. Said program code can also be provided as a computer program product, for example in the form of a data carrier carrying computer program code for performing the embodiments herein when being loaded into the person differentiation system 1. One such carrier can be in the form of a CD ROM disc and / or a hard disk drive, but other data carriers, such as an optical disk, e.g. a DVD, or a memory stick, can also be feasible. Furthermore, the computer program code can be provided as pure program code on a server and downloaded to the person differentiation system 1. The person differentiation system 1 can further comprise a memory 110 comprising one or more memory units. The memory 110 can be arranged for storing, for example, information, and further storing data, configurations, schedules and applications for performing the methods herein when being executed in the person differentiation system 1. For example, the computer program code can be implemented in firmware of the embedded processor 109 stored in the flash memory 110, and / or downloaded wirelessly, for example from an off-board server. Furthermore, said units 101-108, the optional processor 109 and / or the optional memory 110 can at least partly be comprised in, associated with and / or connected to the thermal camera 10 and / or a surveillance system, for example optionally comprising the thermal camera 10. The skilled person will also realize that said units 101-108 described above can refer to a combination of analog and digital circuits and / or one or more processors configured with software and / or firmware, for example stored in a memory such as the memory 110, that when executed by the one or more processors, perform as described herein. One or more of these processors, as well as the other digital hardware, can be comprised in a single application-specific integrated circuit (ASIC), or several processors and various digital hardware can be distributed among several separate components, whether individually packaged or assembled into a system-on-a-chip (SoC).

[0047] Fig. 4 is a flowchart illustrating an exemplary method performed by the person differentiation system 1 according to embodiments of the present disclosure. The method is for differentiating between persons in a crowd in an image. The exemplary method, which can be repeated continuously, comprises in a first step S101Fig. 1-3 One or more of the following actions discussed below can be taken with the support of the person differentiation system 1. Furthermore, actions and / or one or more actions can be taken in any suitable order and / or can be performed simultaneously and / or in an alternating order, where applicable. For example, the action 1004 can be performed simultaneously with and / or before the action 1002 and / or the action 1003.

[0048] Action 1001

[0049] In the action 1001, the person differentiation system 1 identifies one or more detected objects classified as persons 2 in the image 3 derived from the thermal camera 10 adapted to capture a scene in a substantially front-looking angle, e.g. with the support of the person identification unit 101.

[0050] Action 1002

[0051] In the action 1002, the person differentiation system 1 identifies at least a first group 4 of adjacent pixels in the image 3 having an intensity in a predefinable intensity range, which is not included in the one or more detected persons 2, e.g. with the support of the group identification unit 102.

[0052] Action 1003

[0053] In the action 1003, the person differentiation system 1 determines a group pixel area 40 of the at least first group 4 in the image 3, e.g. with the support of the area determination unit 103.

[0054] Action 1004

[0055] In the action 1004, the person differentiation system 1 determines an expected pixel area x expected of a person head at the at least first vertical position y expected based on the head size reference data 5 for the at least first vertical position y expected .

[0056] Action 1005

[0057] In the action 1005, the person differentiation system 1 compares at least a portion of the group pixel area 40 with the expected head pixel area x expected for the at least first vertical position y expected , e.g. with the support of the comparison unit 105.

[0058] Action 1006

[0059] In action 1006, the person differentiation system 1 determines that the at least first group 4 comprises the at least first overlapping person 6 when the at least first comparison resulting from the comparison of action 1005 exceeds a predefinable conformity threshold, e.g. with the support of the conformity determination unit 106.

[0060] Action 1007

[0061] In optional action 1007, the person differentiation system 1 can estimate the number of overlapping persons 6 in the at least first group 4 based on the number of comparisons exceeding the conformity threshold, e.g. with the support of the optional number estimation unit 107.

[0062] Action 1008

[0063] In optional action 1008, the person differentiation system 1 can estimate the total number of persons in the image 3 by adding the number of overlapping persons 6 to the one or more detected persons 2, e.g. with the support of the optional total number estimation unit 108.

[0064] The person skilled in the art realizes that the present disclosure is in no way limited to the preferred embodiments described above. Rather, a large number of modifications and variations are possible within the scope of the attached claims. Moreover, it should be noted that the drawings were not necessarily drawn to scale and that the dimensions of certain features can have been exaggerated for the sake of clarity. Rather, emphasis is placed on the principles of the embodiments described herein. In addition, in the claims the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality.

Claims

1. A method performed by a person differentiation system (1) for differentiating persons in a crowd in an image, the method comprising: identifying (1001) one or more detected objects classified as persons (2) in an image (3) derived from a thermal camera (10) adapted to capture a scene at a substantially forward viewing angle considered to be at or slightly above a human eye level; identifying (1002) at least a first group (4) of adjacent pixels in the image (3) having intensities within a predefinable intensity range not comprised in the one or more detected persons (2); determining (1003) a group pixel area (40) of the at least first group (4) in the image (3); determining (1004), based on head size reference data (5), for at least a first vertical position of the group pixel area (40) in the image (3), an expected pixel area size and form of a human head at the at least first vertical position, wherein the head size reference data (5) is represented by data indicative of a relationship between vertical positions (y) in the image (3) and corresponding expected pixel areas (x) of human heads, i.e. sizes and forms; comparing (1005) at least a portion of the group pixel area (40) resembling a human head to the expected pixel area for the at least first vertical position; and determining (1006) that the at least first group (4) comprises at least a first overlapping person (6) when at least a first comparison resulting from the comparison exceeds a predefinable compliance threshold.

2. The method according to claim 1, further comprising: estimating (1007) a number of overlapping persons (6) in the at least first group (4) based on a number of comparisons exceeding the compliance threshold.

3. The method according to claim 2, further comprising: estimating (1008) a total number of persons in the image (3) by adding the number of overlapping persons (6) to the one or more detected persons (2).

4. The method of any one of claims 1-3, wherein, The identifying (1002) at least a first group (4) comprises the intensity range is based on intensity measurements of at least one (21, 22) of the one or more detected persons (2).

5. The method of any one of claims 1-3, wherein, The head size reference data (5) is based on mapping two (21, 22) or more of the detected persons (2) to respective vertical positions (yl, y2) and pixel areas (xl, x2) of heads (212, 222) of the detected persons (21, 22) in the image (3).

6. The method of claim 5, wherein, The head size reference data (5) is based on interpolation from the mappings.

7. A person differentiation system (1) for differentiating persons in a crowd in an image, the person differentiation system (1) comprising: a person identification unit (101) for identifying (1001) one or more detected objects classified as persons (2) in an image (3) derived from a thermal camera (10) adapted to capture a scene at a substantially forward viewing angle considered to be at or slightly above a human eye level; a group identification unit (102) configured to identify (1002) at least a first group (4) of contiguous pixels in the image (3) having intensities within a predefinable intensity range not comprised in one or more detected persons (2); a region determination unit (103) adapted to determine (1003) a group pixel region (40) of the at least first group (4) in the image (3); an expected region determination unit (104) configured to determine (1004), based on head size reference data (5), for at least a first vertical position of the group pixel region (40) in the image (3), an expected pixel region size and form of a human head at the at least first vertical position, wherein the head size reference data (5) is represented by data indicative of a relationship between vertical positions (y) in the image (3) and corresponding expected pixel regions (x) of human heads, i.e. sizes and forms; a comparison unit (105) configured to compare (1005) at least a portion of a human head at least similar to the group pixel region (40) with the expected pixel region for the at least first vertical position; and a compliance determination unit (106) configured to determine (1006) that the at least first group (4) comprises at least a first overlapping person (6) when at least a first comparison resulting from the comparison exceeds a predefinable compliance threshold.

8. The person differentiation system (1) according to claim 7, further comprising: a number estimation unit (107) configured to estimate (1007) a number of overlapping persons (6) in the at least first group (4) based on a number of comparisons exceeding the compliance threshold.

9. The person differentiation system (1) according to claim 7 or 8, further comprising: a total number estimation unit (108) configured to estimate (1008) a total number of persons in the image (3) by adding the number of overlapping persons (6) to the one or more detected persons (2).

10. The human distinguishing system (1) according to any one of claims 7-8, wherein The group identification unit (102) is adapted such that the intensity range is based on intensity measurements of at least one (21, 22) of the one or more detected persons (2).

11. The human differentiation system (1) according to any one of claims 7-8, wherein, The head size reference data (5) is based on two (21, 22) or more mappings of respective vertical positions (yl, y2) and pixel regions (xl, x2) of heads (212, 222) of the detected persons (21, 22) in the image (3) for the detected persons (2).

12. The human distinguishing system (1) according to claim 11, wherein The head size reference data (5) is based on interpolation from the mappings.

13. A thermal camera (10) comprising the person differentiation system (1) according to any one of claims 7-12.

14. A computer program product comprising a computer program stored on a computer readable medium, the computer program comprising computer program code means adapted to cause a computer or a processor to perform the steps of the method according to any one of claims 1-6.

15. A non-transitory computer readable storage medium having stored thereon the computer program product according to claim 14.

Images