JP7201554B2 - Person detection device, method and program - Google Patents

Description

The present invention relates to a person detection device, method, and program, and in particular, in person detection targeting a camera image in which a person is photographed with a shadow, such as a camera image of a person photographed from above, the person is affected by the shadow. a person detection device, method, and program capable of accurately detecting a person without

For security and crime prevention purposes, research is being conducted to extract a person's region based on a camera image of a person taken outdoors from above, and to analyze the movement of each person. Conventionally, such outdoor cameras were fixedly installed in high places on structures such as buildings, but in recent years, it has become possible to mount small cameras on so-called drones. This makes it easy to shoot at any angle and orientation, and is expected to be applied to various fields.

Non-Patent Document 1 discloses a technique for identifying an object based on the color characteristics of its image by machine learning. Non-Patent Document 2 discloses a technique for extracting a person's skeleton from only an image.

Patent Literature 1 discloses a technique for quickly and accurately tracking an unauthorized intruder to a detection area for detecting an intruder in a monitored site. Patent Document 2 discloses an image that can efficiently monitor a suspicious person who may have carried out a highly suspicious delivery in a monitoring area where an unspecified number of people gather and contact between people occurs exclusively to greet them. A monitoring device is disclosed.

Patent application No. 2017-216477 Patent application No. 2017-238178

SSD: Single Shot Multi Detector (https://arxiv.org/pdf/1512.02325.pdf) Z. Cao, T. Simon, S. Wei and Y. Sheikh, "Realtime Multi-person 2D Pose Estimation Using Part Affinity Fields," 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, 2017, pp 1302-1310.

FIG. 8A is a diagram showing an example of a camera image of a person photographed obliquely from above under fine weather, in which the shadow of the person is projected on the ground in addition to the person. If an existing human region extraction algorithm, such as that disclosed in Non-Patent Document 2, is applied to such a camera image, as shown in FIG. In addition to being set in the person area, as shown in FIG. 8C, the shadow area may also be erroneously set.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above technical problems, and to detect a person accurately without being affected by the shadow in the person detection using a camera image in which a person is captured with the shadow. to provide the program.

In order to achieve the above object, the present invention provides a person detection method for registering a person area in a current frame based on a person area registered in a previous frame and a person area candidate extracted from the current frame between frames of a moving image. The device is characterized by having the following configuration.

(1) Means for initial registration of all human regions in a frame, means for extracting human region candidates from subsequent frames after the initial registration of the human regions and acquiring their reliability, and human regions registered in the previous frame. a means for calculating the degree of similarity with each person region candidate extracted from the current frame; and means for registering the person area in the frame.

(2) means for registering a human region registers human region candidates whose reliability is equal to or greater than a predetermined threshold in the current frame and whose similarity to the human region in the previous frame is equal to or greater than a predetermined threshold; Among them, the person area candidate with the largest similarity is registered as the person area in the current frame.

(3) The means for registering a person area is to detect a person area candidate whose reliability is equal to or higher than a predetermined threshold in the current frame and whose similarity to the person area in the previous frame is equal to or higher than a predetermined threshold. If no candidate exists, the person area in the previous frame is registered as the person area in the current frame.

(4) The means for registering the human area registers, if there is no human area candidate whose reliability is equal to or higher than a predetermined threshold value in the current frame, the human area candidate having the maximum similarity with the human area in the previous frame. A person area candidate is registered as a person area in the current frame.

(5) The means for registering the person area registers the person area in the previous frame as the person area in the current frame if the person area candidate does not exist in the current frame.

According to the present invention, the following effects are achieved.
(1) After correctly initially registering the human region in the frame, the human region is registered in the current frame based on the degree of similarity and reliability with each human region candidate extracted from the current frame for each human region registered in the previous frame. be registered. Therefore, assuming that the initial registration has been correctly performed, the person area can be accurately registered in the current frame based on the person area already registered in the previous frame and the person area candidate extracted from the current frame.

(2) If there is a human region candidate whose reliability is equal to or higher than a predetermined threshold in the current frame, the human region candidate whose degree of similarity to the human region in the previous frame is equal to or higher than a predetermined threshold is the candidate with the highest similarity. The person area candidate is registered as the person area of the current frame. Therefore, if the person area in the previous frame is correct, the person area can be correctly registered in the current frame as well.

(3) If a human region candidate whose reliability is equal to or higher than a predetermined threshold exists in the current frame, but there is no human region candidate whose similarity to the human region in the previous frame is equal to or higher than a predetermined threshold, The person area of the frame is registered as the person area of the current frame. Therefore, if the person area in the previous frame is correct, the person area can be correctly registered in the current frame even if the person area candidate cannot be correctly extracted from the current frame.

(4) If there is no human region candidate whose reliability is equal to or higher than a predetermined threshold in the current frame, the human region candidate having the highest degree of similarity with the human region in the previous frame is the human region candidate in the current frame. Registered as an area. Therefore, if the person area in the previous frame is correct, the person area can be correctly registered in the current frame even if the person area candidate cannot be correctly extracted from the current frame.

(5) If there is no person area candidate in the current frame, the person area in the previous frame is registered as the person area in the current frame. Therefore, if the person area in the previous frame is correct, the person area can be correctly registered in the current frame even if the person area candidate cannot be correctly extracted from the current frame.

1 is a functional block diagram showing the configuration of a human detection device according to one embodiment of the present invention; FIG. FIG. 11 is a diagram (part 1) showing a method of registering a Boundinb Box in the current frame; FIG. 10 is a diagram (part 2) showing a method of registering a Boundinb Box in the current frame; FIG. 13 is a diagram (part 3) showing a method of registering the Boundin Box of the current frame; FIG. 4 is a diagram (4) showing a method of registering the BoundinBBox of the current frame; FIG. 10 is a flow chart showing a procedure for registering a Boundinb Box in the current frame; FIG. FIG. 11 is a diagram (part 1) showing a method of registering a Boundinb Box in the current frame; FIG. 10 is a diagram (part 2) showing a method of registering a Boundinb Box in the current frame; FIG. 13 is a diagram (part 3) showing a method of registering a Boundinb Box in the current frame; FIG. 4 is a diagram (part 4) showing a method of registering a Boundinb Box in the current frame; It is a figure showing an example of a camera image which photographed a person from oblique information. FIG. 10 is a diagram showing an example in which a Boundinb Box is correctly set in a camera image; FIG. 10 is a diagram showing an example in which a BoundinB Box is erroneously set in a camera image;

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a functional block diagram showing the configuration of the main part of a human detection device 1 according to one embodiment of the present invention. A person area is registered in the current frame based on the person area candidate. In this embodiment, an example of human detection in an environment in which there is no change in people within the imaging range of the camera and no change in the number of people will be described.

Such a person detection device 1 can be configured by installing an application (program) that realizes each function described later on a general-purpose computer or server equipped with a CPU, a memory, an interface, and a bus connecting them. Alternatively, a part of the application can be configured as a dedicated machine or a single-function machine that is hardware or programmed.

In the human detection device 1, the frame image acquisition unit 101 acquires a frame image from a camera image of a person. The initial registration unit 102 initially registers all of the person areas in the frame at correct positions. In the present embodiment, the operation of manually setting a rectangular bounding box (BB) that circumscribes each person area on the frame by the operator can be accepted, and the BB can be initially registered in each person area. Alternatively, if an existing Boundinb Box extraction algorithm is applied and it is confirmed that all human regions have been extracted from the frame and the reliability of each human region is equal to or higher than a predetermined threshold, the extracted human regions can be extracted. Initial registration may be performed. As described above, in this embodiment, all the BBs initially registered by the initial registration unit 102 are assumed to be correct, and each frame after the next frame is processed.

After the initial registration of BB by the initial registration unit 102, the human region candidate extraction unit 103 extracts a human region candidate (candidate BB) from each subsequent frame using a known algorithm such as that disclosed in Non-Patent Document 2. Extract it and get its confidence T.

The similarity calculator 104 calculates the similarity Q between each BB registered in the previous frame and each candidate BB extracted from the current frame. In this embodiment, the similarity Q is obtained by the following equation (1). When BB is a rectangle, its size is represented by the sum of its long side and short side. The distance between each BB is represented by the distance between the center positions of the BBs.

Q=1／(Bounding box size difference + Bounding box distance) ...(1)

The human area registration unit 105 matches the BB registered in the previous frame with the candidate BB extracted from the current frame, and selects either the registered BB or the candidate BB based on the reliability T and the similarity Q. Register as BB of the person area in the current frame.

2A to 2D are diagrams conceptually showing the registration method of BB by the human area registration unit 105. As shown in FIG. If the candidate BB extracted from the frame is also correct, the candidate BB extracted from the current frame is determined and registered as the BB of the current frame. As shown in FIG. 2B, if the candidate BB extracted from the current frame is not correct (broken line frame), the BB already registered in the previous frame is determined and registered as the BB of the current frame.

Whether or not the BB registered in the previous frame or the candidate BB extracted from the current frame is correct depends on the reliability when each BB/candidate BB is extracted. If it is equal to or greater than the threshold value, it is determined to be correct, and if it is less than a predetermined threshold value, it is determined to be incorrect (the same applies hereinafter).

On the other hand, as shown in FIG. 2C, if the registered BB in the previous frame is incorrect and the candidate BB extracted from the current frame is correct, the candidate BB is determined and registered as the BB of the current frame. . As shown in FIG. 2D, if the candidate BB extracted from the current frame is also incorrect, the BB already registered in the previous frame is determined and registered as the BB of the current frame. However, in this case the current frame's BB remains incorrect.

Returning to FIG. 1, in the person area registration unit 105, the first registration unit 105a selects candidate BBs whose reliability T is equal to or greater than a predetermined threshold value Tref in the current frame, as will be described later with reference to FIGS. exists, and there exists a candidate BB whose similarity Q to the registered BB of the previous frame is equal to or greater than a predetermined threshold value Qref. Register as

As will be described in detail later with reference to FIGS. 3 and 5, the second registration unit 105b detects that the current frame has a candidate BB whose reliability T is equal to or greater than a predetermined threshold value Tref, but the previous frame has already been registered. If there is no candidate BB whose similarity Q to the BB is equal to or greater than the threshold Qref, the registered BB of the previous frame is registered as the corresponding BB of the current frame.

As will be described later in detail with reference to FIGS. The candidate BB with the largest similarity Q with the registered BB is registered as the corresponding BB of the current frame.

As will be described in detail later with reference to FIGS. 3 and 7, the fourth registration unit 105d replaces the registered BB of the previous frame with the current frame if no candidate BB corresponding to the registered BB of the previous frame exists in the current frame. Register as the corresponding BB of the frame.

FIG. 3 is a flow chart showing a procedure for matching each BB registered in the previous frame with each candidate BB extracted from the current frame and registering the BB in each person area of the current frame.

When extraction of all candidate BBs from the current frame is completed, in step S1, one of the unexamined registered BBs from the previous frame is selected as the current examination target (examination BB). In step S2, it is determined whether or not the candidate BB remains in the current frame. Since all candidate BBs are left at first, the process proceeds to step S3.

In step S3, it is determined whether or not there is a candidate BB whose reliability T is greater than or equal to the reference reliability Tref in the current frame. If at least one such candidate BB exists, the process proceeds to step S4, and it is determined whether or not there exists a candidate BB whose similarity Q to the BB under consideration this time is equal to or greater than the reference similarity Qref.

If there is a candidate BB whose similarity Q with the BB under consideration is equal to or higher than the reference similarity Qref, the process proceeds to step S5, and the candidate BB with the highest similarity Q with the BB under consideration is determined as the BB of the current frame. In step S6, the current consideration BB is regarded as having been examined and is excluded from subsequent examination targets.

For example, as shown in FIG. 4, the BB1 of the previous frame is the BB under consideration this time, the reliability T of all the candidate BBs of the current frame is equal to or higher than the reference reliability Tref, and the BB under consideration and the candidate BB1 , BB2, and BB3 are equal to or greater than the reference similarity Qref, and if the similarity Q to the candidate BB1 is the largest among them, the candidate BB1 of the current frame becomes the current frame in step S5. BB (BB1) is determined and registered.

Thus, according to the present embodiment, if there is a candidate BB with high reliability T in the current frame, and if there is a BB whose similarity Q with the BB in the previous frame is equal to or greater than a predetermined value, the The candidate BB with the highest degree of similarity is registered as the BB of the current frame, so if the BB of the previous frame is correct, the BB of the current frame can also be registered correctly.

In step S7, it is determined whether or not there is an unexamined BB remaining in the previous frame. If unexamined BBs remain, the process returns to step S1, a new BB under consideration is selected from the unexamined BBs, and the above processes are repeated.

In step S4, if it is determined that there is no candidate BB whose similarity Q to the BB under consideration this time is equal to or greater than the reference similarity Qref in the current frame, the process proceeds to step S9, where the BB under consideration this time is determined to be the BB of the current frame. registered.

For example, as shown in FIG. 5, BB2 of the previous frame is the BB under consideration this time, and although the reliability T of the candidates BB3 and BB4 of the current frame is equal to or higher than the reference reliability Tref, the similarity Q with the BB2 under consideration is is equal to or greater than the reference similarity Qref, the BB under consideration this time is determined and registered as BB2 of the current frame.

As described above, according to the present embodiment, although there are candidate BBs with high reliability T in the current frame, if there is no candidate BB with high similarity Q to the BB of the previous frame, the previous frame BB is registered instead of the candidate BB. Therefore, if the BB of the previous frame is correct, even if the candidate BB cannot be extracted correctly from the current frame, the BB of the size that would exist in the current frame at a position that is not significantly different from the actual person region in the current frame. can be registered based on the position and size of the BB in the previous frame.

In step S3, if it is determined that there is no candidate BB whose reliability T with respect to the BB under consideration this time is equal to or higher than the reference reliability Tref, the process proceeds to step S10. In step S10, the candidate BB having the highest similarity Q with the BB under consideration this time is determined and registered as the BB of the current frame.

For example, as shown in FIG. 6, BB3 of the previous frame is the BB under consideration this time, and there is no candidate BB whose reliability T with the BB under consideration is greater than or equal to the reference reliability Tref in the current frame. If the degree of similarity is the highest, candidate BB3 is determined and registered as BB3 of the current frame.

As described above, according to the present embodiment, if there is no candidate BB with a high reliability T in the current frame, the candidate BB with the highest similarity Q to the BB in the previous frame is selected from among the candidate BBs in the current frame. Therefore, if the BB of the previous frame is correct, the BB can be correctly registered in the current frame even if the candidate BB cannot be correctly extracted from the current frame.

In step S2, if there are no candidate BBs left in the current frame even though unexamined BBs remain in the previous frame, the process proceeds to step S11. In step S11, the BB under consideration this time (BB4) is determined as the BB of the current frame and registered.

For example, as shown in FIG. 7, since the number of candidate BBs detected in the current frame is smaller than the original number of people, there may be a shortage of candidate BBs when BB4 in the previous frame becomes the target BB. be. In such a case, the BB4 under consideration this time is determined and registered as the BB4 of the current frame.

Thus, according to the present embodiment, if the candidate BB does not exist in the current frame, the BB of the previous frame is registered as the BB of the current frame. BBs can be correctly registered in the current frame even if they have not been extracted, and the same number of BBs as registered in the previous frame can be registered in the current frame as well.

1... person detection device, 101... frame image acquisition unit, 102... initial registration unit, 103... person area candidate extraction unit, 104... similarity calculation unit, 105... person area Registration unit, 105a...first registration unit, 105b...second registration unit, 105c...third registration unit, 105d...fourth registration unit

Claims (20)

A person detection device that registers a person region in a current frame between frames of a moving image based on a person region registered in the previous frame and a person region candidate extracted from the current frame,
means for initially registering all of the person regions in the frame;
means for extracting a human region candidate from each subsequent frame after the initial registration of the human region and obtaining the reliability thereof;
means for calculating a degree of similarity between each human region registered in the previous frame and each human region candidate extracted from the current frame;
means for registering each person area registered in the previous frame in the current frame based on the degree of similarity and reliability with each person area candidate in the current frame;
The means for registering the human region registers, when there are human region candidates whose reliability is equal to or higher than a predetermined threshold in the current frame, 1. A person detection device characterized by registering a person area candidate with the highest degree of similarity as a person area in a current frame. The means for registering the human region is configured such that although there are human region candidates whose reliability is greater than or equal to a predetermined threshold in the current frame, there is a human region candidate whose similarity to the human region of the previous frame is greater than or equal to a predetermined threshold. 2. The person detection device according to claim 1, wherein if the person area does not exist, the person area in the previous frame is registered as the person area in the current frame. The means for registering the human region registers a human region having the highest degree of similarity with the human region of the previous frame among the human region candidates if there is no human region candidate whose reliability is equal to or higher than a predetermined threshold value in the current frame. 3. The person detection device according to claim 1, wherein the candidate is registered as a person area in the current frame. 4. The person area registering means according to any one of claims 1 to 3, wherein if the person area candidate does not exist in the current frame, the means for registering the person area registers the person area in the previous frame as the person area in the current frame. Person detection device. A person detection device that registers a person region in a current frame between frames of a moving image based on a person region registered in the previous frame and a person region candidate extracted from the current frame,
means for initially registering all of the person regions in the frame;
means for extracting a human region candidate from each subsequent frame after the initial registration of the human region and obtaining the reliability thereof;
means for calculating a degree of similarity between each human region registered in the previous frame and each human region candidate extracted from the current frame;
means for registering each person area registered in the previous frame in the current frame based on the degree of similarity and reliability with each person area candidate in the current frame;
The means for registering the human region is configured such that although there are human region candidates whose reliability is greater than or equal to a predetermined threshold in the current frame, there is a human region candidate whose similarity to the human region of the previous frame is greater than or equal to a predetermined threshold. 1. A person detection device, characterized in that, if the person does not exist, the person area in the previous frame is registered as the person area in the current frame. The means for registering the human region registers a human region having the highest degree of similarity with the human region of the previous frame among the human region candidates if there is no human region candidate whose reliability is equal to or higher than a predetermined threshold value in the current frame. 6. The person detection device according to claim 5, wherein the candidate is registered as a person area in the current frame. 7. The human detection device according to claim 5, wherein said means for registering said human region registers said human region in said previous frame as said human region in said current frame when said human region candidate does not exist in said current frame. . A person detection device that registers a person region in a current frame between frames of a moving image based on a person region registered in the previous frame and a person region candidate extracted from the current frame,
means for initially registering all of the person regions in the frame;
means for extracting a human region candidate from each subsequent frame after the initial registration of the human region and obtaining the reliability thereof;
means for calculating a degree of similarity between each human region registered in the previous frame and each human region candidate extracted from the current frame;
means for registering each person area registered in the previous frame in the current frame based on the degree of similarity and reliability with each person area candidate in the current frame;
The means for registering the human region registers a human region having the highest degree of similarity with the human region of the previous frame among the human region candidates if there is no human region candidate whose reliability is equal to or higher than a predetermined threshold value in the current frame. A person detection device characterized by registering a candidate as a person area in a current frame. 9. The person detection apparatus according to claim 8, wherein said means for registering the person area registers the person area of the previous frame as the person area in the current frame if the person area candidate does not exist in the current frame. A person detection device that registers a person region in a current frame between frames of a moving image based on a person region registered in the previous frame and a person region candidate extracted from the current frame,
means for initially registering all of the person regions in the frame;
means for extracting a human region candidate from each subsequent frame after the initial registration of the human region and obtaining the reliability thereof;
means for calculating a degree of similarity between each human region registered in the previous frame and each human region candidate extracted from the current frame;
means for registering each person area registered in the previous frame in the current frame based on the degree of similarity and reliability with each person area candidate in the current frame;
The person detecting device, wherein the means for registering the person area registers the person area in the previous frame as the person area in the current frame if the person area candidate does not exist in the current frame. 11. The person according to any one of claims 1 to 10, wherein said means for initial registration accepts a manual operation for setting a person area on said frame and initially registers said person area. detection device. The means for initial registration performs initial registration of the extracted human regions when all the human regions have been extracted from the frame and the reliability of each human region is equal to or higher than a predetermined threshold. The person detection device according to any one of claims 1 to 10. In a person detection method executed by a computer, between frames of a moving image, a person region is registered in a current frame based on the person region registered in the previous frame and a person region candidate extracted from the current frame,
Initial registration of all human areas in the frame,
After the initial registration of the person region, extracting the person region candidate from each subsequent frame and obtaining its reliability,
For each human region registered in the previous frame, the degree of similarity with each human region candidate extracted from the current frame is calculated,
For each human region registered in the previous frame, the human region is registered in the current frame based on the degree of similarity and reliability with each human region candidate in the current frame,
When registering the human region in the current frame, if there is a human region candidate whose reliability is equal to or higher than a predetermined threshold in the current frame, a human region among them whose similarity to the human region in the previous frame is equal to or higher than a predetermined threshold. A person detection method, comprising: registering a person area candidate having the highest degree of similarity among candidates as a person area in a current frame. In a person detection method executed by a computer, between frames of a moving image, a person region is registered in a current frame based on the person region registered in the previous frame and a person region candidate extracted from the current frame,
Initial registration of all human areas in the frame,
After the initial registration of the person region, extracting the person region candidate from each subsequent frame and obtaining its reliability,
For each human region registered in the previous frame, the degree of similarity with each human region candidate extracted from the current frame is calculated,
For each human region registered in the previous frame, the human region is registered in the current frame based on the degree of similarity and reliability with each human region candidate in the current frame,
When registering a human region in the current frame, there are human region candidates whose reliability is greater than or equal to a predetermined threshold in the current frame, and among them, a person whose degree of similarity to the human region in the previous frame is greater than or equal to a predetermined threshold. A person detection method comprising registering a person area in a previous frame as a person area in a current frame if no area candidate exists. In a person detection method executed by a computer, between frames of a moving image, a person region is registered in a current frame based on the person region registered in the previous frame and a person region candidate extracted from the current frame,
Initial registration of all human areas in the frame,
After the initial registration of the person region, extracting the person region candidate from each subsequent frame and obtaining its reliability,
For each human region registered in the previous frame, the degree of similarity with each human region candidate extracted from the current frame is calculated,
For each human region registered in the previous frame, the human region is registered in the current frame based on the degree of similarity and reliability with each human region candidate in the current frame,
When registering a human region in the current frame, if there is no human region candidate whose reliability is equal to or higher than a predetermined threshold value in the current frame, the similarity to the human region in the previous frame is maximized among the human region candidates. A human detection method characterized by registering a human region candidate of the current frame as a human region. In a person detection method executed by a computer, between frames of a moving image, a person region is registered in a current frame based on the person region registered in the previous frame and a person region candidate extracted from the current frame,
Initial registration of all human areas in the frame,
After the initial registration of the person region, extracting the person region candidate from each subsequent frame and obtaining its reliability,
For each human region registered in the previous frame, the degree of similarity with each human region candidate extracted from the current frame is calculated,
For each human region registered in the previous frame, the human region is registered in the current frame based on the degree of similarity and reliability with each human region candidate in the current frame,
A person detection method, comprising registering a person area in a previous frame as a person area in the current frame if no person area candidate exists in the current frame when registering the person area in the current frame. In a person detection program for registering a person region in a current frame between frames of a moving image based on a person region registered in a previous frame and a person region candidate extracted from the current frame,
a procedure for initial registration of all person regions in a frame;
a procedure for extracting a human region candidate from each subsequent frame after the initial registration of the human region and obtaining its reliability;
a procedure for calculating the degree of similarity between each human region registered in the previous frame and each human region candidate extracted from the current frame;
causing a computer to execute a procedure for registering a human region in the current frame based on the degree of similarity and reliability with each human region candidate in the current frame for each human region registered in the previous frame;
In the procedure for registering the human region, if there are human region candidates whose reliability is equal to or higher than a predetermined threshold in the current frame, among the human region candidates whose similarity to the human region in the previous frame is equal to or higher than a predetermined threshold, A human detection program characterized by registering a human region candidate with the highest degree of similarity as a human region in a current frame. In a person detection program for registering a person region in a current frame between frames of a moving image based on a person region registered in a previous frame and a person region candidate extracted from the current frame,
a procedure for initial registration of all person regions in a frame;
a procedure for extracting a human region candidate from each subsequent frame after the initial registration of the human region and obtaining its reliability;
a procedure for calculating the degree of similarity between each human region registered in the previous frame and each human region candidate extracted from the current frame;
causing a computer to execute a procedure for registering a human region in the current frame based on the degree of similarity and reliability with each human region candidate in the current frame for each human region registered in the previous frame;
In the procedure for registering the human region, there are human region candidates whose reliability is equal to or higher than a predetermined threshold in the current frame, and among them, there are human region candidates whose similarity to the human region in the previous frame is equal to or higher than the predetermined threshold. A person detection program characterized by registering a person area in a previous frame as a person area in a current frame if the person area does not exist. In a person detection program for registering a person region in a current frame between frames of a moving image based on a person region registered in a previous frame and a person region candidate extracted from the current frame,
a procedure for initial registration of all person regions in a frame;
a procedure for extracting a human region candidate from each subsequent frame after the initial registration of the human region and obtaining its reliability;
a procedure for calculating the degree of similarity between each human region registered in the previous frame and each human region candidate extracted from the current frame;
causing a computer to execute a procedure for registering a human region in the current frame based on the degree of similarity and reliability with each human region candidate in the current frame for each human region registered in the previous frame;
In the procedure for registering the human region, if there is no human region candidate whose reliability is equal to or higher than a predetermined threshold value in the current frame, the human region having the maximum similarity to the human region in the previous frame among the human region candidates is registered. A person detection program characterized by registering a candidate as a person area in a current frame. In a person detection program for registering a person region in a current frame between frames of a moving image based on a person region registered in a previous frame and a person region candidate extracted from the current frame,
a procedure for initial registration of all person regions in a frame;
a procedure for extracting a human region candidate from each subsequent frame after the initial registration of the human region and obtaining its reliability;
a procedure for calculating the degree of similarity between each human region registered in the previous frame and each human region candidate extracted from the current frame;
causing a computer to execute a procedure for registering a human region in the current frame based on the degree of similarity and reliability with each human region candidate in the current frame for each human region registered in the previous frame;
A human detection program, wherein, in the procedure for registering the human region, if a human region candidate does not exist in the current frame, the human region in the previous frame is registered as the human region in the current frame.

Images