TWI479432B - Abnormal detection method for a video camera - Google Patents

Description

Camera anomaly detection method

The invention relates to a surveillance photography system, and more particularly to a camera anomaly detection method.

At present, the surveillance photography system on the market can capture images of the surveillance scene and display them synchronously on the display, so that the person can immediately view the image of one or more scenes by the display. The image captured by the surveillance camera system can be further recorded on a storage medium such as a video tape or a computer hard disk to replay the image after a specific event (such as a theft event) to confirm the event.

However, in order to avoid being small, the camera was given a broken line/steering/defocus/painting/shadowing before the crime was committed, causing the surveillance system to record an invalid image. One of the existing prevention methods is to install a detection loop on the camera to detect whether the camera is connected to the monitoring system. However, this detection loop can only achieve the warning that the image transmission line of the camera and the video recorder is cut.

In order to detect whether the camera is turned or not, another prevention method is to install a displacement sensor (such as a three-axis gyroscope or a three-axis accelerometer) on the camera, and the displacement sensor detects whether the camera has displacement. However, this method can only support camera steering. If the camera is blocked or out of focus, this method will not detect this abnormal state.

Therefore, there is also a way to prevent images and backgrounds captured by the camera. The base map is compared to detect whether the camera has an abnormal event such as shadowing or out-of-focus. However, when there is a major change in ambient light or light (such as turning off the lights or turning on the lights), in order to avoid false positives caused by detection errors, it is necessary for the personnel to manually set a new background model sample, so that the inconvenient process will generate additional personnel. The burden of making images has no real benefit.

In view of the above problems, the present invention provides a camera anomaly detection method, which solves the problem that the prior art has to manually set a new background model sample for a light change person.

An embodiment of the present invention provides a method for detecting an abnormality of a camera. Among them, the camera has a daytime shooting mode and a nighttime shooting mode. The abnormality detecting method comprises: obtaining a captured image captured by the camera; detecting a color saturation of the captured image; determining that the camera enters when the color saturation of the captured image is lower than a first threshold a night shooting mode or a daytime shooting mode; selecting a corresponding background model according to the shooting mode of the camera; and determining whether the captured image is abnormal according to the background model.

According to the abnormality detecting method of the camera of the present invention, the edge feature and the scene structure can be combined as a background model for detecting an abnormal event (ie, an edge feature model and a scene structure model). Since the edge features are robust under different light sources, the edge features can be preserved through the infrared image even in a low light source environment. Therefore, the abnormality detecting method of the camera of the present invention can be applied to any light intensity environment and can resist severe Light changes to avoid false positives. Furthermore, the abnormality detecting method of the camera of the present invention further analyzes the types of abnormal events according to the edge feature model and the scene structure model, and can distinguish out of the out-of-focus event, the shadowing event, the steering event, and the shock event (micro-turn).

1 is a block diagram showing the structure of a surveillance photography system 100 in accordance with an embodiment of the present invention.

As shown in FIG. 1, the surveillance camera system 100 can include a monitoring host 110, a camera 130, a video recorder 150, and a display 170. The monitoring host 110 can be substantially a computer host (such as an x86-based computer system) or an embedded host (such as an embedded system based on an ARM, SoC or DSP architecture) for running an image analysis software and can receive images from the camera 130. And the video signals of the video recorder 150, and the video signals are output to the display 170 for display.

The camera 130 can be placed in the surveillance area and photographed in a specific direction, and one or more cameras 130 can be provided depending on the needs of use. The camera 130 can be a digital camera, and can be connected to the monitoring host 110 via an interface such as an image capture card or a network card of the monitoring host 110, so that the monitoring host 110 can receive the image captured by the camera 130 and display the image on the display. 170 shows the captured image. The camera 130 can also be an analog camera, and the captured image is output to the video recorder 150 by analog signal (for example, connected between the camera 130 and the video recorder 150 via a coaxial cable signal). The video recorder 150 can be a digital video recorder (DVR) for instant use. The captured image of the connected camera 130 is backed up, and the captured image is further converted into a digital signal and transmitted to the monitoring host 110 for display on the display 170. The display 170 can be a cathode ray tube display or a liquid crystal display or the like.

Here, the camera 130 referred to in the embodiment of the present invention may be an infrared camera and has an infrared photography function. Infrared photographic images or color photographic images can be obtained by enabling the infrared photography function. Moreover, the camera 130 has a photodetector that can detect the ambient light intensity to automatically enable the infrared photography function when the ambient brightness is insufficient, thereby overcoming the problem that the captured image is unclear due to insufficient brightness.

In other words, the camera 130 according to the embodiment of the present invention has a daytime shooting mode and a nighttime shooting mode, and can capture color images in the daytime shooting mode. When entering the nighttime shooting mode, the camera 130 will turn on the infrared device therein to shoot infrared images. .

FIG. 2 is a flow chart of an abnormality detecting method of the camera 130 according to an embodiment of the invention. The image detecting software run by the host 110 and the camera 130 are executed to execute the abnormality detecting method of the camera 130 shown in FIG.

Please refer to Figure 2. First, the captured image captured by the camera 130 is acquired (step S210). Next, the color saturation of the captured image is detected (step S220). When the color saturation of the captured image is lower than the first threshold, it is determined that the camera 130 enters the night shooting mode or the day shooting mode (step S230). Continuing, the corresponding background model is selected in accordance with the shooting mode of the camera 130 (step S240). If the camera 130 is in the daytime shooting mode, the background model corresponding to the daytime shooting mode is selected; if the camera 130 is in the nighttime shooting mode, the background mode corresponding to the nighttime shooting mode is selected. type. Finally, it is determined whether the captured image is abnormal according to the selected background model (step S250), and it is known whether the camera 130 is abnormal.

Here, as described above, the camera 130 can be an infrared camera, and can automatically detect the ambient light intensity. When the camera 130 detects that the ambient light intensity is lower than the second threshold value (such as 10 lux) in the daytime shooting mode, the camera 130 automatically switches to the infrared photography mode (ie, the night shooting mode) to obtain the infrared image. Capture images. Conversely, the night shooting mode is switched to the day shooting mode.

That is to say, when the light in the surveillance area is insufficient (such as when the light is turned off), the camera 130 switches to the infrared photography mode, and the captured image of the camera 130 is therefore an infrared image, and the color saturation of the captured image is also reduced. In order to further distinguish between camera daytime and nighttime switching or lens out of focus, shadowing and other anomalous events, in addition to the color saturation must be lower than the first threshold, the scene structure and edge information also need to meet certain conditions, in order to correctly distinguish the camera day and night Switch or lens out of focus, shadowing and other abnormal events.

In some embodiments, the foregoing step S240 may include the following steps: First, the background model is established according to the background image captured by the camera 130 in the daytime shooting mode or the nighttime shooting mode, respectively. Next, when the camera 130 enters the daytime shooting mode, the background model corresponding to the daytime shooting mode is selected; when the camera 130 enters the nighttime shooting mode, the background model corresponding to the nighttime shooting mode is selected.

FIG. 3 is an initialization flow chart of the abnormality detecting method of the camera 130 according to an embodiment of the present invention. At the beginning of the abnormality detecting method of the camera 130 of the embodiment (ie, between step S210 and step S220), the edge information needs to be established first. The scene structure model and the hybrid Gaussian mixture model are used to establish the edge feature model. As shown in Figure 3, the steps include: Step S301: The plurality of sample points 410 are selected on a background image captured by the camera 130 (ie, the background scene image), and the mixed Gaussian model of each sample point 410 is established as the edge feature model by using the edge intensity of the sample points 410. As shown in Figure 4A. 4A is a schematic diagram of an edge feature model in accordance with an embodiment of the present invention. The number and distribution of the sampling points 410 shown in FIG. 4A are only schematic, and the embodiment of the present invention is not limited thereto. The hybrid Gaussian model only updates the average selected sampling point 410, thereby reducing the amount of computation and accelerating the computational results.

Herein, the detection of the edge intensity of the sampling point 410 can be implemented by using the Sobel image edge detection method, but the embodiment of the present invention is not limited thereto, and may be other edge detection methods (such as the Robert operator). , Prewitt operator or Laplacian operator, etc.). After performing the image edge detection method, the edge intensity of each point in the background image may be binarized (such as Otsu algorithm) to determine which sample points 410 belong to the edge point. That is to say, via binarization, points in the sample points 410 where the edge intensity is greater than a specific value may be regarded as edge points.

Step S302: The background image 400 is cut into two-dimensionally distributed complex scene blocks 420 to form a scene structure model, wherein each scene block 420 partially overlaps with the adjacent scene block 420, as shown in FIG. 4B. 4B is a schematic diagram of a scene structure model in accordance with an embodiment of the present invention. The background image can be divided into m×n scene blocks 420 (m, n are positive integers), and the number of scene blocks 420 shown in FIG. 4B is only illustrative, and the embodiment of the present invention is not limited thereto. Here, due to the adjacent scene block 420 They overlap partially with each other, so that false alarms caused by shaking of the camera 130 can be reduced.

After the background image 400 is cut into a plurality of scene blocks 420, the regional features of each scene block 420 may be further established. The regional features may be the edge distribution and number in the scene block 420, and the scene block 420 is utilized. The regional features may constitute the aforementioned scene structure model.

Step S303: Combine the edge feature model and the scene structure model into a background model. That is to say, the background model includes the aforementioned edge feature model and scene structure model. That is, the background model is formed by selecting the edge feature of the plurality of sample points 410 from the background image 400 and cutting the background image 400 into the plurality of scene blocks 420, thereby identifying the abnormal event through the edge feature and the scene block 420. The camera is turned, blocked, out of focus, etc.

Referring to FIG. 5A, it is a schematic structural diagram of an image analysis software 500 according to an embodiment of the present invention. The image analysis software 500 can include an image capture module 510, a light change detection module 530, a camera anomaly detection module 550, and an abnormality reporting module 570.

The image capturing module 510 receives the captured image (such as the background image or the current image) of the camera 130 and performs image processing thereon. The image capturing module 510 includes a color conversion unit 512, an edge calculation unit 514, and a binarization unit 516. The color conversion unit 512 is configured to convert the captured image of the three primary color light (RGB) color space into a captured image of a color space (eg, HSV) separated by color and brightness to obtain a color saturation parameter of the captured image. The edge calculation unit 514 can calculate the gradient approximation of the image brightness function using a Sobel operator to detect the edge intensity of the sample point 410. The binarization unit 516 can utilize the Otsu algorithm as described above for the sampling point 410 The edge intensity is binarized to find the edge points.

The optical change detection module 530 and the camera abnormality detection module 550 receive the captured images processed by the image capturing module 510, and respectively perform light change detection and camera abnormality detection. When the infrared photography function of the camera 130 is not enabled (that is, when the ambient light is sufficient), the abnormality detection is directly performed by the camera abnormality detecting module 550. When the infrared photography function of the camera 130 is enabled (that is, when the ambient light is insufficient), The light change detection module 530 detects light changes according to the captured image and performs anomaly detection. As such, the abnormality detecting method of the camera 130 according to the embodiment of the present invention can be applied regardless of whether the light is sufficient or not. In particular, under the two different ambient brightness conditions, the aforementioned edge model and scene structure model can be separately established.

The optical change detection module 530 includes a color saturation detection unit 532, an edge number detection unit 534, a scene structure model detection unit 536, and a continuous block shadow detection unit 538.

The color saturation detecting unit 532 is configured to detect, according to the color saturation parameter obtained by the color converting unit 512, whether the color saturation of the captured image is lower than the first threshold, that is, the detecting camera 130 is converted by the daytime shooting mode. To night shooting mode. On the other hand, the color saturation detecting unit 532 can also detect that the color saturation is greatly increased from below the first threshold to a value higher than the first threshold, that is, the detecting camera 130 is switched from the night shooting mode to the daytime. Shooting mode.

The edge number detecting unit 534 is configured to identify the number of edge features according to the calculation result of the edge calculating unit 514. The scene structure model detecting unit 536 is configured to cut the captured image into the plurality of scene blocks 420, and detect each field of the currently captured image. Whether the scenic block 420 is similar to each of the scene blocks 420 corresponding to the background image 400.

The contiguous block occlusion detecting unit 538 is configured to determine whether the dissimilar scene block 420 changes along the adjacent scene block 420 in a continuous frame to determine whether the mask is blocked by the mask. event. In other words, the contiguous block occlusion detecting unit 538 is configured to detect whether the similarity of the consecutive scene blocks 420 is lower than a third threshold.

The camera abnormality detecting module 550 includes a scene structure model detecting unit 552 and an edge sampling model detecting unit 554. The scene structure model detecting unit 552 and the edge sampling model detecting unit 554 of the scene structure model detecting unit 536 and the camera anomaly detecting module 550 will be described later.

Here, the abnormality reporting module 570 includes an abnormality determining unit 572, an abnormality counting unit 574, and an abnormality reporting unit 576. The abnormality determining unit 572 is configured to determine whether an abnormal event occurs according to the detection result of the light change detecting module 530 and the camera abnormality detecting module 550, and determine the abnormal event. The abnormality counting unit 574 determines the cumulative number of occurrences as a result of the abnormality event based on the abnormality determining unit 572, and notifies the abnormality transmitting unit 576 to issue an alarm when it is accumulated to a specific number of times. By avoiding short-term unintentional changes that result in a huge change in the image, causing false alarms, such as headlights, lightning, etc. of the moving vehicle.

FIG. 5B is a schematic structural diagram of a scene structure model detecting unit 552/536 according to an embodiment of the invention.

As shown in FIG. 5B, the scene structure model detecting unit 552/536 includes a background block cutting subunit 5521, a fuzzy comparison subunit 5523, and a scene structure comparison subunit. 5525.

The background block cutting subunit 5521 is configured to cut the captured image (such as the background image 400 or the current image) into the foregoing plurality of scene blocks 420. Here, the current captured image is also divided into m×n scene blocks 420 (m, n are positive integers), that is, the number of scene blocks 420 is the same as the number of scene blocks 420 of the scene structure model.

The fuzzy comparison sub-unit 5523 uses the fuzzy-similarity algorithm to calculate the similarity between the scene structure model and each scene block 420 corresponding to the captured image, so that the scene structure comparison sub-unit 5524 The scene block 420 that identifies the captured image is not similar to the scene block 420 corresponding to the scene structure model. Here, the weight parameter is set according to the importance degree of the visual scene block 420, and the product of the similarity degree and the weight parameter is used as the basis for identifying the similarity or not. In this way, false positives can be avoided due to frequently changing areas such as corridors.

FIG. 5C is a schematic diagram of the architecture of the edge sampling model detecting unit 554 according to an embodiment of the invention.

As shown in FIG. 5C, the edge sampling model detecting unit 554 includes a foreground detecting subunit 5541, an unstable sampling point analyzing subunit 5543, and a Gaussian model building subunit 5545.

The foreground detection sub-unit 5541 is configured to identify an edge different from the edge feature in the edge feature model according to the calculation result of the edge calculation unit 514, and identify it as a foreground edge. The unstable sampling point analysis sub-unit 5543 analyzes the degree of edge stability. For example, whether the entropy value (Entopy) is the minimum value by the edge intensity of each sampling point 410 at different times (since the entropy value of the sampling point 410 is smaller) , indicating that the fetch The more stable the sample is, to confirm that this edge point is a stable edge. The Gaussian model building sub-unit 5554 is configured to perform the foregoing steps S301 to S303, and initially establish the aforementioned edge feature model.

FIG. 6 is a detailed flow chart of step S250 in the flowchart shown in FIG. 2, which is used to explain how to further distinguish whether an abnormal event or an abnormal event occurs after the color saturation is detected to be lower than the first threshold. As shown in FIG. 6, step S250 includes: step S251: comparing whether the color saturation of the captured image is lower than the first threshold. If the color saturation of the captured image is lower than the first threshold, it indicates that the captured image may be an infrared image or an abnormal event such as out-of-focus, shadowing, switching light or steering occurs, so the process proceeds to step S2511 to follow the night shooting mode. The background model determines whether an abnormal event has occurred. If not, it indicates that the captured image may be a color image captured during the day or an abnormal event such as out-of-focus, shadowing, switching lights, or steering occurs, and the process proceeds to step S2512. How to determine whether an abnormality has occurred will be explained in detail in Figure 7.

In step S2511, if an abnormal event is detected, the process proceeds to step S252, and if no, the process proceeds to step S260 to update the background model.

Similarly, in step S2512, if an abnormal event is detected, the process proceeds to step S252, and if not, the process proceeds to step S260 to update the background model. If the abnormality is not detected and the process proceeds to step S260, the edge feature of the sample point 410 of the background image in the background model and the scene block 420 are updated to the edge feature and the scene block of the sample point 410 in the current captured image. 420. Therefore, the edge feature model and the scene structure model (ie, the background model) will also gradually learn during the detection process to adapt to the scene change. (eg furniture shifting, slow light changes, etc.).

Step S252: determining whether the scene structure of the captured image is changed according to the scene structure model, and if so, determining that the steering event is; if not, proceeding to step S253 to continue determining whether another abnormal event has occurred (step S254).

Step S253: Determine whether an edge feature of the sampling point of the captured image exists according to the edge sampling model. If the edge features of the sampling point are not present, it can be determined that an out-of-focus event occurs, because if a light-off event occurs, all edge features of the infrared image will still exist (step S255). If the edge feature of the partial sampling point does not exist, it may be determined that a shadowing event may occur (step S256). If the edge features of the sampling points are present, it can be determined that a light-on/off event may occur (step S257). After detecting the switch light event (ie, after step S257), the process proceeds to step S260 to update the background model.

In step S260, if the light-off event is detected, the background model corresponding to the daytime shooting mode is replaced with the background model corresponding to the nighttime shooting mode; otherwise, the background model corresponding to the nighttime shooting mode is replaced with the corresponding daytime shooting mode. Background model. After detecting the occurrence of the occlusion event, the out-of-focus event or the steering event (ie, after steps S254, S255, and S256), the process proceeds to step S270 to issue an alarm.

In order to further confirm whether the occlusion event does occur, step S256 may further include the following steps: First, whether the scene blocks 420 corresponding to the plurality of consecutive frames of the captured image are similar. Then, when the corresponding positions of the dissimilar scene blocks 420 in the consecutive frames are continuously changed, it is determined that the shading event occurs.

In order to further distinguish the confirmation of the steering of the camera, step S254 may further comprise the following steps: first, comparing the scene block 420 in the frame of the captured image with the continuous Whether the scene block 420 adjacent to the scene block 420 is similar in the frame. Then, based on the similarity among the adjacent blocks, the steering of the camera is determined. For example, if each scene block 420 in the current captured image is displaced to the left compared to the similar and adjacent scene blocks 420 in the previous frame, the camera 130 may be determined to be in the opposite direction ( That is, turn to the right).

FIG. 7 is a flow chart of another abnormality detecting method of the camera 130 according to an embodiment of the present invention, which is used to explain how to use the image analyzing software and the camera 130 operated by the monitoring host 110 to detect an abnormal event and update the built-in device. Background model. Here, this flowchart omits the flow of initial construction of the background model and the process of distinguishing abnormal events. For the related flow, please refer to FIG. 3 and FIG.

As shown in FIG. 7, first, the user first inputs the state in which the infrared function of the camera 130 is turned on to the monitoring host 110 (step S710), and thereafter the monitoring photographing system 100 starts the automatic learning and determination, and does not need to be targeted. Manually change the background model sample if the ambient light is dim or not. Herein, the input of this step includes the captured image of the camera 130 in addition to the infrared function enabled state input by the user.

After step S710, the process proceeds to step S720, and the color saturation of the captured image is detected to determine whether the enabled state of the infrared function of the camera 130 is changed, that is, whether the ambient light produces a change in brightness or darkness (step S730). Thereby, different detection procedures can be performed according to whether the infrared activation state is changed. That is, if the state is changed (i.e., the state in which the infrared ray is not activated is changed to the state in which the infrared ray is enabled, or the state in which the infrared ray is enabled is changed to the state in which the infrared ray is not activated), the process proceeds to step S740; if not, Then, the process proceeds to step S750.

In step S740, continuous block dissimilarity analysis (step S741), edge number detection (step S743), and background structure comparison are performed (step S745).

In step S741, a continuous block dissimilarity analysis is performed, that is, whether the dissimilar dissimilar scene block 420 is changed along the adjacent scene block 420 in the continuous frame as described above. Further, in step S742, it is determined whether the number of consecutively different scene blocks 420 is less than a fourth threshold value, and a logical judgment result (ie, True/Yes or False) is output. If the logical judgment result is false, it means that a shadowing event or a steering event may occur.

After step S743, the process proceeds to step S744 to determine whether the number of edges is less than a fifth threshold value, and output a logical determination result. If the logical judgment result is false, it means that a shadowing event or an out-of-focus event may occur.

After step S745, the process proceeds to step S746 to determine whether the similarity between the background structure of the captured image and the scene structure model after the light change is greater than a sixth threshold, and output a logical determination result. If the logical judgment result is false, it means that a shadowing event, a defocusing event or a turning event may occur.

Here, if it is further determined that the abnormal event is a shadowing event, a defocusing event, or a turning event, the determination can be made through the flow shown in FIG. 6 described above.

After step S742, step S744 and step S746, the output result is subjected to "AND" logic operation. If the operation result is true (Ture), the results output in step S742, step S744 and step S746 are all true, representing When it is detected that the infrared device of the camera is switched and no abnormal event occurs, the process proceeds to step S761; The result is False, which indicates that an abnormal event has occurred, and proceeds to step S762.

In step S761, the background model is updated, that is, the edge feature of the sampling point 410 of the background image in the background model and the scene block 420 are updated to the edge feature of the sampling point 410 and the scene block 420 in the current captured image, and The edge features of all the sample points 410 are resampled, and the process returns to step S710 to analyze the next captured image.

In step S762, the number of abnormal events is accumulated, and it is determined whether the number of abnormal events exceeds a predetermined number of times (step S771), and if so, an alarm is transmitted (step S772). By avoiding short-term unintentional changes that result in a huge change in the image, causing false alarms, such as headlights, lightning, etc. of the moving vehicle.

In step S750, it is judged whether or not the resampling is completed. If yes, the process proceeds to step S781 and step S745 (dashed line); if not, the process proceeds to step S745 (dotted line).

Here, the flow when the resampling is not completed (dotted line) is explained. After step S745 is performed, the process proceeds to step S747 to determine whether the similarity between the background structure of the captured image and the scene structure model of the same ray state is greater than a seventh threshold value, and output a logical determination result. If yes, proceed to step S748 to update the background model, that is, to update the edge feature of the sample point 410 of the background image in the background model and the scene block 420 to the edge feature and the scene block of the sample point 410 in the current captured image. 420; if no, indicating that a shadowing event, a defocusing event, or a turning event may occur, proceeding to step S762, accumulating the number of abnormal events, and determining whether the number of abnormal events exceeds a predetermined number of times (step S771), and if the predetermined number of times is exceeded, sending an alarm (Step S772).

Next, the flow (dotted line) at the time of completion of resampling will be described, and step S745 and step S781 are simultaneously performed. For the step S745 and the subsequent step S747, please refer to the foregoing, and the detailed description is not repeated here. In step S781, foreground edge detection is performed based on the sampling point 410, and the flow proceeds to step S782. In step S782, it is determined whether the foreground ratio is less than an eighth threshold value, that is, whether part of the edge features disappear (as caused by a shadowing event).

According to the logical determination result of step S747 and step S782, an "AND" logic operation is performed. If the operation result is true (Ture), that is, the results output in steps S747 and S782 are all true, indicating that no abnormal event has occurred, then the process proceeds to step S748. If the result of the operation is false (False), it means that an abnormal event has occurred (step S748), then the process proceeds to step S762, the number of abnormal events is accumulated, and an alarm is issued when the predetermined number of times is exceeded (step S772).

In some embodiments, in combination with the processes shown in FIG. 6 and FIG. 7 above, in step S772, an alarm is issued for a specific abnormal event to enable the user to know what an abnormal event (such as a shadowing event or a defocus). An event or a turn event occurs.

According to the abnormality detecting method of the camera 130 of the present invention, it is automatically determined whether the infrared mode of the camera 130 is turned on according to the input image, and then the edge feature and the scene structure of the daytime or nighttime are compared as a background model for detecting an abnormal event (ie, Edge feature model and scene structure model). Since the edge features are robust under different light sources, the edge features can be preserved through the infrared image even in a low light source environment. Therefore, the abnormality detecting method of the camera 130 of the present invention can be applied to any light intensity environment and can resist Violent light changes to avoid false positives. Furthermore, the present invention The abnormality detecting method of the camera 130 further analyzes the types of abnormal events according to the edge feature model and the scene structure model, and can distinguish out of the out-of-focus event, the shadowing event, the steering event, and the shock event (micro-turn).

While the present invention has been described above in the foregoing embodiments, it is not intended to limit the invention, and the invention may be modified and modified without departing from the spirit and scope of the invention. The scope of patent protection shall be subject to the definition of the scope of the patent application attached to this specification.

100‧‧‧ surveillance photography system

110‧‧‧Monitoring host

130‧‧‧ camera

150‧‧‧Video recorder

170‧‧‧ display

400‧‧‧ background image

410‧‧‧ sampling points

420‧‧‧ Scene Blocks

500‧‧‧Image Analysis Software

510‧‧‧Image capture module

512‧‧‧Color Conversion Unit

514‧‧‧Edge calculation unit

516‧‧‧ Binarization unit

530‧‧‧Light Change Detection Module

532‧‧‧Color saturation detection unit

534‧‧‧Edge number detection unit

536‧‧‧Scenario structure model detection unit

538‧‧‧Continuous block mask detection unit

550‧‧‧Camera Anomaly Detection Module

552‧‧‧Scenario structure model detection unit

5521‧‧‧Background block cutting subunit

5523‧‧‧Fuzzy comparison subunit

5524‧‧‧Scenario structure comparison subunit

554‧‧‧Edge sampling model detection unit

5541‧‧‧ foreground detection subunit

5543‧‧‧Unstable sampling point analysis subunit

5545‧‧‧Gaussian model subunit

570‧‧‧Abnormal reporting module

572‧‧‧Abnormal judgment unit

574‧‧‧Abnormal counting unit

576‧‧‧Abnormal reporting unit

1 is a block diagram showing the structure of a surveillance photography system in accordance with an embodiment of the present invention.

FIG. 2 is a flow chart of an abnormality detecting method of a camera according to an embodiment of the invention.

FIG. 3 is an initial flowchart of an abnormality detecting method of a camera according to an embodiment of the invention.

4A is a schematic diagram of an edge feature model in accordance with an embodiment of the present invention.

4B is a schematic diagram of a scene structure model in accordance with an embodiment of the present invention.

FIG. 5A is a schematic structural diagram of an image analysis software according to an embodiment of the invention.

FIG. 5B is a schematic structural diagram of a scene structure model detecting unit according to an embodiment of the invention.

FIG. 5C is a schematic structural diagram of an edge sampling model detecting unit according to an embodiment of the invention.

Fig. 6 is a detailed flow chart of step S250 in the flowchart shown in Fig. 2.

FIG. 7 is a flow chart of another abnormality detecting method of the camera according to an embodiment of the invention.

Claims (10)

A camera abnormality detecting method, the camera has a daytime shooting mode and a nighttime shooting mode, the abnormality detecting method includes: obtaining a captured image captured by the camera; and detecting a color saturation of the captured image; When the color saturation of the captured image is lower than a first threshold, determining that the camera enters a night shooting mode or a daytime shooting mode; selecting a corresponding background model according to the shooting mode of the camera; and determining the background model according to the background model Whether the image is abnormal. The method for detecting an abnormality of the camera according to claim 1, wherein the selecting the corresponding background model according to the shooting mode of the camera comprises: according to a background image captured by the camera in the daytime shooting mode or the nighttime shooting mode, The background model is established; when the camera enters the daytime shooting mode, the background model corresponding to the daytime shooting mode is selected; and when the camera enters the nighttime shooting mode, the background model corresponding to the nighttime shooting mode is selected. The method for detecting an abnormality of the camera according to claim 1, further comprising: selecting a plurality of sampling points on a background image, and establishing an edge feature model by using the edge intensity of the sampling points; and cutting the background image into two dimensions. a plurality of scene blocks distributed to form a scene structure model; and The edge feature model and the scene structure model are merged into the background model. The camera abnormality detecting method of claim 3, wherein each of the scene blocks partially overlaps the adjacent scene block. The method for detecting an abnormality of the camera according to claim 3, wherein determining whether the captured image is abnormal according to the background model comprises: determining, according to the edge feature model, whether an edge feature of the sampling points of the captured image exists; And if none of the edge features of the sampling points exist, it is determined that an out-of-focus event occurs. The method for detecting an abnormality of the camera according to claim 3, wherein determining whether the captured image is abnormal according to the background model comprises: determining, according to the edge feature model, whether an edge feature of the sampling points of the captured image exists; And if some of the edge features of the sampling points do not exist, it is determined that a shadowing event occurs. The method for detecting an abnormality of the camera according to claim 6, wherein if the edge features of the sampling points do not exist, determining that the masking event occurs comprises: respectively corresponding to the plurality of consecutive frames of the captured image Whether the scene blocks are similar; and when the corresponding positions of the scene blocks in the different consecutive frames are continuously changed, it is determined that a shadow event occurs. The method for detecting an abnormality of the camera according to claim 3, wherein determining whether the captured image is abnormal according to the background model comprises: determining, according to the edge feature model, an edge of the sampling points of the captured image Whether the levy exists; and if the edge features of the sampling points are present, it is determined that a switching light event occurs. The method for detecting an abnormality of the camera according to claim 8, wherein determining whether the captured image is abnormal according to the background model comprises: determining, according to the scene structure model, whether a scene structure of the captured image is changed; if the captured image is captured If the scene structure changes, it is determined that a turn event has occurred. The method for detecting an abnormality of the camera according to claim 3, further comprising: if the captured image has no abnormality, updating the edge features of the sampling points of the background image in the background model and the scene blocks to the current The edge features of the sample points in the image and the scene blocks are captured.