Video surveillance systems and servers

Abstract

To enable an object to be highly accurately detected while suppressing a processing load related to object detection.SOLUTION: A server 20 executes: a specific region cut-out step (S2) of generating a frame image after cut-out by cutting out a portion of a cut-out region set for each frame from an original frame image with each frame image included video data as an original frame image; and an object detection processing step (S3) of detecting an object by analyzing the images after reducing the total pixel number of the original frame image after contraction and the frame image after cut-out in comparison to the pixel number of the original frame image before contraction by contracting at least the original frame image with the resolution reduction.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a video surveillance system that detects an object from a captured video of a surveillance area. [Background technology]

[0002] Artificial intelligence (AI) based on machine learning is increasingly being applied to object detection tasks that use time-sequential image data (hereinafter simply referred to as "video") such as video or live footage as input (see, for example, Non-Patent Documents 1 and 2). By incorporating AI into products and systems, it is possible to replace tasks that were previously performed manually, thereby reducing labor costs and improving product value. This contributes to promoting sales of products and systems, and many companies are therefore highly interested in AI.

[0003] When applying AI to live video, methods such as edge AI computing, which completes AI analysis processing at the site where the video is shot, and cloud AI computing, which sends the video to an external cloud server via a network for analysis processing, are used.The advantages of edge AI computing include immediate analysis due to low video transmission delay, and the ability to analyze video in high quality without video compression because there is no need to consider transmission to an external network.The advantages of cloud AI computing include the ability to use large-scale AI servers without being restricted by the site where the video is shot, and the ability to use public cloud services provided by cloud vendors. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Joseph Redmon, Santosh Divvala, Ross Girshick, Ali Farhadi, “You Only Look Once: Unified, Real-Time Object Detection,” June 8, 2015, [online], https: / / arxiv.org / abs / 1506.02640. [Non-patent document 2] Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg, “SSD: Single Shot MultiBox Detector”, December 8, 2015, [online], https: / / arxiv.org / abs / 1512.02325. Summary of the Invention [Problem to be solved by the invention]

[0005] Generally, when performing object detection using an AI server, there is a trade-off between processing speed (number of frames processed per second) and detection accuracy. When edge AI computing is adopted for video surveillance applications, the AI ​​server's computing performance is often limited, and in order to achieve real-time object detection processing, a method is often adopted to reduce the amount of calculations by reducing the image resolution. However, when the image resolution is reduced, the pixel representation of the fine details of the image is lost, which can lead to the problem that the AI ​​cannot detect small objects in the image. Another problem is that when displaying the object detection results by AI in real time, even if a small object is detected in the image, it can be difficult to distinguish it when viewed with the human eye.

[0006] The present invention has been made in consideration of the above-described conventional circumstances, and aims to enable highly accurate object detection while suppressing the processing load associated with object detection. [Means for solving the problem]

[0007] In order to achieve the above object, a video monitoring system according to one aspect of the present invention is configured as follows. In other words, the video surveillance system of the present invention is a video surveillance system comprising a server that processes video data captured of a surveillance area and a monitor that displays the processing results by the server, wherein the server performs a specific area cut-out process for each frame image included in the video data, in which a portion of a cut-out area set for each frame is cut out from the frame image to generate a partial frame image, and an object detection process that reduces at least the frame images by lowering their resolution, thereby reducing the total number of pixels of the reduced frame image and the partial frame image below the number of pixels of the frame image before reduction, and then analyzes these images to detect objects.

[0008] Here, in the object detection process, the server may reduce both the frame image and the partial frame image by lowering the resolution, and detect an object by analyzing each of the reduced images.

[0009] The reduced frame image and the partial frame image may each have 25% or less of the number of pixels of the unreduced frame image.

[0010] In addition, the system has as its internal states a scanning state in which no object is detected by the object detection process, and a tracking state in which an object is detected by the object detection process, and the server may cyclically set, as the cut-out area for each frame, each of a plurality of divided areas that divide the entire image area into a predetermined number of areas in the scanning state, and may set, as the cut-out area for the next frame, a detected object area that includes the position of the detected object in the current frame, in the tracking state.

[0011] In addition, the server may exclude, in the scanning state, a divided area at a predetermined position among the plurality of divided areas from targets for setting the cropping area.

[0012] The server may also have a first division pattern that divides the entire image area by a first division number, and a second division pattern that divides the entire image area by a second division number different from the first division number, and in the specific area cut-out process in the scanning state, generate a first partial frame image cut out according to the first division pattern and a second partial frame image cut out according to the second division pattern, and provide them to the object detection process.

[0013] In addition, in the scanning state, the server may cyclically set each of the multiple divided areas to the cut-out area, and each time the operations of performing the specific area cut-out process and the object detection process are completed, determine whether an object has been detected by the object detection process during that process, and transition to the tracking state on the condition that an object has been detected by the object detection process.

[0014] The monitor may also display the frame image in the scanning state and the partial frame image in the tracking state.

[0015] In addition, the monitor may display, in the tracking state, an image obtained by reducing the frame image so as to be superimposed on the partial frame image.

[0016] In addition, in the tracking state, the monitor may display a detection frame indicating the position of the detected object superimposed on each of the partial frame image and an image obtained by reducing the frame image.

[0017] In addition, in the tracking state, the monitor may display a clipping frame indicating the range of the partial frame image in the frame image, superimposed on an image obtained by reducing the frame image.

[0018] Furthermore, when a plurality of objects are detected by the object detection process in the tracking state, the monitor may display a partial frame image including a detected object selected from among the detected objects.

[0019] Furthermore, the server may limit movement of the clipped region between frames in the tracking state, thereby suppressing fluctuations in the partial frame image displayed on the monitor.

[0020] A server according to another aspect of the present invention is configured as follows. That is, in a server that processes video data captured of a surveillance area, a specific area cut-out process is performed for each frame image included in the video data, in which a portion of a cut-out area set for each frame is cut out from the frame image to generate a partial frame image, and an object detection process is performed in which at least the frame images are reduced in resolution to reduce the total number of pixels of the reduced frame image and the partial frame image below the number of pixels of the frame image before reduction, and then these images are analyzed to detect objects. [Effects of the Invention]

[0021] According to the present invention, it is possible to detect an object with high accuracy while suppressing the processing load related to object detection. [Brief explanation of the drawings]

[0022] [Figure 1] 1 is a diagram illustrating an example of the configuration of a video monitoring system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a diagram showing an overview of processing steps of a video monitoring system. [Figure 3] FIG. 2 is a diagram showing the transition of the internal state of the video monitoring system. [Figure 4] 10A and 10B are diagrams illustrating a method for determining a cutout region in a scanning state. [Figure 5] 10A and 10B are diagrams illustrating a method for determining a clipping region in a tracking state. [Figure 6A]10A and 10B are diagrams illustrating a method for displaying the object detection processing result in a scanning state. [Figure 6B] 10A and 10B are diagrams illustrating a method for displaying the object detection processing result in a tracking state. DETAILED DESCRIPTION OF THE INVENTION

[0023] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The following description is an example of the present invention, and the present invention is not limited to this example. An example of the configuration of a video surveillance system according to one embodiment of the present invention is shown in Figure 1. The video surveillance system of this example includes a camera 10 that captures images of a surveillance area, a server 20 that processes the images captured by camera 10 and performs object detection, and a monitor 30 that displays the images captured by camera 10 and the results of object detection by server 20.

[0024] Figure 2 shows an overview of the processing steps in the video monitoring system shown in Figure 1. As shown in Figure 2, the video monitoring system of this example has a video input step (S1), a specific area cutout step (S2), an object detection processing step (S3), a next frame cutout area determination step (S4), an object detection processing result display step (S5), and a state transition processing step (S6).

[0025] In a video input step (S1), the server 20 acquires video data of a monitoring area from the camera 10. Alternatively, the server 20 may acquire streaming video data instead of video data from the camera 10. The acquired video data may be video captured or distributed in real time, or may be video recorded as a video file or a set of temporally consecutive still images. The video data acquired in this step is divided into frames and then output to the subsequent specific area extraction step (S2) and object detection processing step (S3).

[0026] In the specific area cutout step (S2), the server 20 cuts out a specific area from the frame image input from the video input step (S1) and outputs it to the subsequent object detection processing step (S3). The area cut out in this step depends on the internal state of the system and the processing result of the cutout area determination step (S4) for the previous frame, as will be described later.

[0027] In the object detection processing step (S3), the server 20 performs downsampling (reduction processing) on ​​the frame image input from the video input step (S1) (hereinafter referred to as the "original frame image") and the frame image input from the specific area cut-out step (S2) (hereinafter referred to as the "cut-out frame image") so that the number of pixels in each image reaches the target number of pixels, and then analyzes each image using object detection AI to perform object detection.

[0028] The target pixel count is set to, for example, less than 50% of the pixel count of the original frame image before reduction, and equal to or smaller than the pixel count of the cut-out frame image. If the pixel count of the cut-out frame image matches the target pixel count, only the original frame image needs to be reduced in resolution. On the other hand, if the pixel count of the cut-out frame image is greater than the target pixel count, both the original frame image and the cut-out frame image are reduced in resolution.

[0029] That is, in the object detection processing step (S3), the original frame image (and the cut-out frame image) is reduced to a target pixel count, and the total number of pixels in the reduced original frame image and partial frame image is reduced below the number of pixels in the original frame image before reduction, and then object detection is performed from these images. Preferably, the target pixel count is set to 25% or less of the pixel count of the original frame image, and then the total number of pixels in the reduced original frame image and cut-out frame image is set to half or less of the pixel count of the original frame image before reduction, and then object detection is performed from these images.

[0030] Any known method or algorithm can be selected as the method or algorithm for detecting a target object from an image. Examples include YOLO (You Only Look Once; see Non-Patent Document 1) and SSD (Single Shot Multibox Detector; see Non-Patent Document 2). As the output of this step, the object detection processing result for the cropped frame image is output to the subsequent crop region determination step (S4). In addition, the cropped frame image, the original frame image, and the object detection processing result for the original frame image are output to the subsequent object detection processing result display step (S5).

[0031] In the cut-out area determination step (S4), the server 20 refers to the internal state of the system and the object detection processing result for the cut-out frame image to determine the area to be cut out in the specific area cut-out step (S2) in the next frame. As the output of this step, the object detection processing result for the input cut-out frame image is output as is to the subsequent object detection processing result display step (S5).

[0032] In an object detection processing result display step (S5), the server 20 refers to the input object detection processing result and draws a detection frame indicating the position of the detected object for the input original frame image and the cut-out frame image. The server 20 also resizes and superimposes both frame images after the detection frame has been drawn, and transmits them to the monitor 30 for display. Here, the resizing and superimposing processing method is determined depending on the internal state of the system. As the output of this step, the object detection processing result for the cut-out frame image is output to the subsequent state transition processing step (S6).

[0033] In the state transition processing step (S6), the server 20 refers to the object detection processing result for the input extracted frame, and if the transition condition is satisfied, causes the internal state of the system to transition. The above steps are repeatedly executed for each frame image that constitutes the video data.

[0034] Next, we will explain the internal states defined in this system. As shown in Figure 3, this system has a scanning state (St1) and a tracking state (St2), and is always in one of these states. The scanning state (St1) is a state in which no object is detected, and the tracking state (St2) is a state in which an object is detected. The initial internal state of the system is the scanning state (St1).

[0035] In the state transition processing step (S6), if the internal state of the system is in the scanning state (St1), the object detection processing result for the input clipped frame image is referenced, and if the number of detected objects is 1 or more, the system transitions to the tracking state (St2). Also, in the state transition processing step (S6), if the internal state of the system is in the tracking state (St2), the object detection processing result for the input clipped frame image is referenced, and if the number of detected objects is 0, the system transitions to the scanning state (St1).

[0036] Next, a method for determining the cut-out region in the specific region cut-out step (S2) will be described. The method for determining the cut-out region differs depending on whether the internal state of the system is the scanning state (St1) or the tracking state (St2).

[0037] (When the system's internal state is scanning) A method for determining a cut-out area in the scan state (St1) will be described with reference to Fig. 4. In the specific area cut-out step (S2), the server 20 divides the original frame image input from the video input step (S1) into a predetermined number of divisions N, generating N cut-out frame images. Fig. 4 shows an example of division of an original frame image. Fig. 4 shows an example of division when the number of divisions N=4, in which the input original frame image is divided equally into four parts: upper left (division area 1), upper right (division area 2), lower left (division area 3), and lower right (division area 4).

[0038] Here, let us define i as an integer between 1 and N, and j as an integer equal to or greater than 0. In the cropping region determination step (S4), if the current frame is the (i+j×N)th frame in the video input to the system, divided region i is set as the cropping region, and the cropped frame image obtained by cropping that region from the (1+j×N)th original frame image is output to the subsequent object detection processing step (S3). That is, the (1+j×N)th original frame image is divided into N equal parts to generate N cropped frame images, and object detection processing is performed on each cropped frame image sequentially over the time required for N frames. Also, in the cropping region determination step (S4), server 20 references the results of the object detection processing on the input cropped frame image, and if the transition condition that the number of detected objects is 1 or greater is met, the cropping region for the specific region cropping step (S2) of the next frame is set according to the cropping region determination method described in the next section.

[0039] (When the system's internal state is tracking) A method for determining a cutout area in the tracking state (St2) (or in the scanning state and when the transition conditions defined in the previous section are satisfied) will be described with reference to Fig. 5. The upper part of Fig. 5 shows how the specific area cutout step (S2), object detection processing step (S3), and cutout area determination step (S4) are performed in order for the nth frame image in the video input to the system. As shown in the upper part of Fig. 5, if an object is detected in the object detection processing step (S3), the difference D between the center coordinate C1 of the post-cutout frame image 52 and the center coordinate C2 of the detection frame 50 of the detected object is calculated in the subsequent cutout area determination step (S4).

[0040] The lower part of Fig. 5 shows how the specific area cutout step (S2) is performed on the (n+1)th frame of the video input to the system. Here, the difference D calculated in the nth frame is added to the coordinates of the cutout area 52 in the nth frame to determine the cutout area 54 in the (n+1)th frame. In other words, the detected object area set to include the position of the detected object in the current frame is determined as the cutout area for the next frame.

[0041] 5 shows an example of calculating coordinates within the post-extraction frame image, but when many frames are processed in succession, errors may accumulate. Therefore, as an alternative method, the coordinate values ​​within the post-extraction frame image may be converted into a coordinate system within the original frame image, and then the process of determining the extraction area may be carried out.

[0042] Next, a method for displaying the object detection processing result in the object detection processing result display step (S5) will be described with reference to Figures 6A and 6B. When the internal state of the system is the scanning state (St1), of the original frame image and the cropped frame image that are input to the object detection processing result display step (S5), the original frame image is displayed in the entire window of the monitor 30, as shown as display example 60 in Figure 6A. Note that display example 60 in Figure 6A is a display example immediately after an object has been detected, and a cropping frame 61 that indicates the range of the cropped frame image on the original frame image is superimposed.

[0043] On the other hand, when the internal state of the system is the tracking state (St2), as shown as display example 70 in FIG. 6B, of the original frame image and the cut-out frame image that are input to the object detection processing result display step (S5), the cut-out frame image is displayed in the entire window of monitor 30. Furthermore, the original frame image is reduced and then displayed superimposed on a portion (75) of the window (cut-out frame image). Note that in display example 70 in FIG. 6B, a detection frame 71 that indicates the position of the object in the cut-out frame image is superimposed on a cut-out frame 76 that indicates the range of the cut-out frame image in the original frame image, and a detection frame 77 that indicates the position of the object. Here, the reduction ratio and display position of the original frame image that is displayed superimposed on a portion of the cut-out frame image may be determined arbitrarily.

[0044] As described above, in the video surveillance system of this example, the server 20 performs a specific area cutout step (S2) in which each frame image included in the video data is used as an original frame image and a portion of the cutout area set for each frame is cut out from the original frame image to generate a cutout frame image, and an object detection processing step (S3) in which the server reduces at least the original frame image by lowering the resolution so that the total number of pixels of the reduced original frame image and the cutout frame image is lower than the number of pixels of the original frame image before reduction, and then analyzes these images to detect objects. This makes it possible to detect objects with high accuracy while reducing the processing load related to object detection.

[0045] In addition, in a scanning state (St1) in which no object is detected, the server 20 cyclically sets each of a plurality of divided regions, which divide the entire image area into a predetermined number of regions, as the cutout region for each frame. In a tracking state (St2) in which an object is detected, the server 20 sets a detected object region including the position of the detected object in the current frame as the cutout region for the next frame. In other words, the server 20 performs object detection processing by scanning each divided region of the captured image, and if it detects the object to be detected in any divided region, it acts to track that object. In addition, if the object being tracked moves outside the range of the image or if the object being tracked is lost due to insufficient performance of the object detection AI, it can search for a new detection target or search again for a detection target that was once lost by terminating tracking and restarting scanning.

[0046] Furthermore, in this example's video surveillance system, regardless of whether the internal state is in scanning or tracking mode, the system always performs object detection processing on the entire captured video and displays the object detection processing results, allowing system users to visually check the entire captured video at any time.Furthermore, when the internal state of the video surveillance system is in tracking mode, as shown in Figure 6, compared to the conventional method of displaying the original frame image in the entire window, the cropped frame image that captures the detected object area is displayed larger, allowing system users to clearly visually check the object being detected.

[0047] In the above description, in the tracking state, object detection processing is performed on the original frame image and the clipped frame image for tracking, which is obtained by clipping a detected object area from the original frame image. However, object detection processing on the clipped frame image for search may also be added. That is, in the tracking state, a search operation is executed in the background, in which multiple divided areas are cyclically clipped from the original frame image to generate clipped frame images for search, and object detection processing is performed. This allows object detection to be performed in the tracking state at the same level as in the search state, making it possible to reduce missed detections. However, reduction processing must be performed so that the total number of pixels in the reduced original frame image and the two clipped frame images is smaller than the number of pixels in the original frame image before reduction.

[0048] An example of the present system will be described below, but the present system is not limited to the example shown below. (First Example) If the number of pixels in the original frame image obtained from the input video in the video input step (S1) is 100%, then the number of pixels in each image processed by the object detection AI in the object detection processing step (S3) is set to 25%. In other words, the object detection AI processes an image with a resolution that is half the height and half the width of the original frame image. Furthermore, when the system's internal state is in a scanning state, the number and method of dividing the frame image in the specific area extraction step (S2) is set to N=4 (i.e., divided into two equal parts vertically and horizontally), as shown in Figure 4. Furthermore, it is assumed that the calculation time of the object detection AI used in the object detection processing step (S3) is proportional to the number of pixels in the image to be detected.

[0049] In this case, in the object detection processing step (S3), the object detection AI processes two images with 25% of the pixel count compared to the original frame image, so the processing volume of the object detection AI is 50% of the processing volume of the original frame image per frame. Here, the original frame image is processed by the object detection AI after its pixel count has been reduced from 100% to 25%. On the other hand, the post-cropping frame image has 25% of the pixel count when it is cropped from the original frame image, so it is processed by the object detection AI at the same size without being reduced.

[0050] If this system were not applied and the original frame images were processed by the object detection AI at life-size, the processing volume of the object detection AI would be 100% of the processing volume per frame for the original frame images. In contrast, with this system, the processing volume of the object detection AI would be 50% of the processing volume per frame for the original frame images. In this way, it can be seen that by applying this system, object detection can be achieved at essentially life-size while halving the object detection processing time per frame.

[0051] (Second Example) In the specific area cutout step (S2), the number of divisions N into which the original frame image is divided can be set arbitrarily, and the shape and location of the divided areas can also be set arbitrarily. As an example, if there is an area that does not require object detection processing, such as when there is a known obstruction in the input image to the system, that area is excluded from the allocation of divided areas. In other words, when determining the cutout area in the scanning state, a divided area at a predetermined position among the multiple divided areas is excluded from the cutout area setting targets. This reduces the target area for object detection processing, thereby further shortening the object detection processing time.

[0052] (Third Example) In the explanation so far, the number of parallel processes for object detection processing in the object detection processing step (S3) has been set to two (original frame image and cropped frame image). However, by preparing multiple division patterns for the original frame image, object detection processing may be performed in parallel using a wider variety of patterns. For example, a first division pattern that divides the original frame image into four equal parts and a second division pattern that divides the original frame image into nine equal parts are defined. Then, in the specific area cropping step (S2) in the scanned state, a first cropped frame image cropped according to the first division pattern and a second cropped frame image cropped according to the second division pattern are generated, and these images are output to the object detection processing step (S3). In the object detection processing step (S3), object detection processing is performed in parallel on three images: the original frame image, the first cropped frame image, and the second cropped frame image. This makes it possible to vary the reduction ratio of each image that is the target of object detection processing. This makes it possible to improve the detection performance of the entire system in cases where the size of the object to be detected changes drastically in the input video to the system, or where the object detection AI is not robust against changes in the size of the object to be detected.

[0053] (Fourth Example) In the above description of the method for determining the clipping region for the next frame in the clipping region determination step (S4), a case where one object is detected in the object detection processing step (S3) has been used as an example. However, two or more objects may be detected in the object detection processing step (S3). Therefore, as a countermeasure for the case where two or more objects are detected, a mechanism for selecting an object to be tracked may be provided. For example, multiple detection frames are set to surround each detected object, and the object with the smallest detection frame area is selected as the tracking target. This allows the monitor 30 to display a clipped frame image including the smallest detected object, making it easier for the system user to track it. As another example, multiple detection frames surrounding each detected object are superimposed on the original frame image and displayed on the monitor 30, and an object selected by the user from among them is selected as the tracking target.

[0054] (Fifth Example) In the explanation so far, the method for determining the clipping region for the next frame in the clipping region determination step (S4) has been described as using the difference D between the center coordinate C1 of the clipped frame image and the center coordinate C2 of the detection frame, as shown in Fig. 5. However, this method may impair visibility and comprehensibility when the clipped frame image is displayed on the monitor 30 in cases such as when the detection target is moving rapidly. Therefore, as a countermeasure to this, fluctuations in the clipped frame image displayed on the monitor 30 may be suppressed by setting a limit on the movement of the clipping region between frames in the tracking state.

[0055] The movement of the cutout region can be restricted by various methods. As one example, the movement of the cutout region between frames is made to be a constant speed. As another example, the movement of the cutout region between frames is made to be an accelerated movement. As yet another example, the cutout region is moved only when the center coordinates of the object to be tracked are outside a certain range from the center coordinates of the cutout region, that is, when the distance between the center coordinates of the object to be tracked and the center coordinates of the cutout region exceeds a predetermined threshold. Note that these methods are merely examples, and the movement of the cutout region may be restricted by other methods.

[0056] (Sixth Example) In the above description of the state transition processing step (S6), when the internal state of the system is in the scanning state, the system immediately transitions to the tracking state when the transition condition is met. However, with this method, there is a possibility that the state transition occurs before object detection processing is performed on all divided areas in the original frame image. In this case, object detection processing will not be performed on some divided areas. To address this issue, the transition condition from the scanning state to the tracking state may be changed so that the state transition waits until object detection processing is completed on all divided areas. Specifically, each of divided areas 1 to N is cyclically set as an extraction area, and each time the specific area extraction step (S2) and object detection processing step (S3) are performed, it is determined whether an object has been detected in the object detection processing step (S3) during that process, and the system transitions to the tracking state if an object has been detected.

[0057] (Seventh Example) Regarding the object detection processing result display step (S5), when the internal state of the system is in a tracking state, in the explanation up to this point, the original frame image and the extracted frame image are displayed in a single window as shown in Fig. 6, but this is merely an example. For example, the original frame image and the extracted frame image may be displayed in separate windows.

[0058] Although the embodiments of the present invention have been described above, the above embodiments are merely illustrative and do not limit the technical scope of the present invention. The present invention can take various other embodiments, and various modifications such as omissions and substitutions can be made without departing from the spirit of the present invention. These embodiments and modifications are included in the scope and spirit of the invention described in this specification, etc., and are included in the invention described in the claims and their equivalents.

[0059] Furthermore, the present invention can be provided not only as devices such as those described above or as systems composed of these devices, but also as methods executed by these devices, programs for realizing the functions of these devices using a processor, and storage media for storing such programs in a computer-readable manner. [Industrial Applicability]

[0060] The present invention can be used in a video monitoring system that detects an object from a captured video of a monitored area. [Explanation of symbols]

[0061] 10: Camera, 20: Server, 30: Monitor

Claims

1. A video surveillance system including a server that processes video data captured in a surveillance area and a monitor that displays the results of the processing by the server, The server performs the following for each frame image included in the video data: a specific area cutout process for cutting out a portion of the cutout area set for each frame from the frame image to generate a partial frame image; a function of performing an object detection process of detecting an object by analyzing the images after reducing the total number of pixels of the reduced frame image and the partial frame image by reducing the resolution of at least the frame image so that the total number of pixels of the reduced frame image and the partial frame image is less than the number of pixels of the frame image before reduction, The internal state includes a scanning state in which an object is not detected by the object detection process, and a tracking state in which an object is detected by the object detection process, A video surveillance system characterized in that, in the scanning state, the server cyclically sets each of a plurality of divided areas that divide the entire image area into a predetermined number of areas as the cut-out area for each frame, and in the tracking state, sets a detected object area that includes the position of the detected object in the current frame as the cut-out area for the next frame.

2. 2. The video monitoring system according to claim 1, The video surveillance system is characterized in that, in the object detection process, the server reduces the resolution of both the frame image and the partial frame image, and analyzes each of the reduced images to detect objects.

3. 3. The video monitoring system according to claim 1, The video monitoring system is characterized in that, in the scanning state, the server excludes a divided area at a predetermined position among the plurality of divided areas from the setting targets for the cut-out area.

4. 3. The video monitoring system according to claim 1, The server has a first division pattern that divides the entire image area by a first division number, and a second division pattern that divides the entire image area by a second division number different from the first division number, and in the specific area cut-out process in the scanning state, generates a first partial frame image cut out according to the first division pattern and a second partial frame image cut out according to the second division pattern, and provides them to the object detection process.

5. 3. The video monitoring system according to claim 1, In the video surveillance system, the server, in the scanning state, cyclically sets each of the multiple divided areas to the cut-out area, and each time the specific area cut-out process and the object detection process are performed once, determines whether an object has been detected by the object detection process during that process, and transitions to the tracking state if an object has been detected by the object detection process.

6. 3. The video monitoring system according to claim 1, The video surveillance system is characterized in that the monitor displays the frame image in the scanning state and the partial frame image in the tracking state.

7. In a server that processes video data captured in a monitoring area, For each frame image included in the video data, a specific area cutout process for cutting out a portion of the cutout area set for each frame from the frame image to generate a partial frame image; a function of performing an object detection process of detecting an object by analyzing the images after reducing the total number of pixels of the reduced frame image and the partial frame image by reducing the resolution of at least the frame image so that the total number of pixels of the reduced frame image and the partial frame image is less than the number of pixels of the frame image before reduction, The internal state includes a scanning state in which an object is not detected by the object detection process, and a tracking state in which an object is detected by the object detection process, In the scanning state, the cut-out area for each frame is cyclically set to each of a plurality of divided areas that divide the entire image area into a predetermined number of areas, and in the tracking state, the cut-out area for the next frame is set to a detected object area that includes the position of the detected object in the current frame.

Images