TW201432633A - Falling down detection method - Google Patents

Abstract

A falling down detection method, implemented by a monitoring system, comprises following steps: based on images transferred from a camera, identifying a person entering a target area; in a current image, locating and categorizing a plurality of edges of the person, and calculating a vertical-edge-direction proportion based on a number of the edges with a vertical direction; determining if the vertical-edge-direction proportion is less than a preset threshold proportion, if so, adding one to a number of times of an abnormal condition and recording corresponding time of the current image; determining if an interval between the corresponding time of the current image and that of the first abnormal condition or that of the last abnormal condition is not more than a preset length of time, if not so, zeroing the number of times of the abnormal condition, and if so, determining if the number of times of abnormal condition is greater than a preset number of times, if greater, than judging that a person is falling down and proceeding related outputs, so as to precisely and rapidly detect if a person is falling down.

Description

Fall detection method

The invention relates to a method for detecting abnormal conditions of personal safety, in particular to a method for detecting a fall.

Fall is the main cause of accidental death in the elderly. Because the elderly have a low degree of coordination due to aging, and often fail to get help in time after the fall, it causes more harm. Even if it is an average person, it also needs timely in case of serious falls. assist.

A method for detecting a fall is to allow the object to be detected to carry a sensor. In addition to causing inconvenience to the carrier, the object to be detected is also very limited, and only the high-risk group that falls will carry the sensor in advance. And, when the user forgets to carry the sensor, it has no effect.

It also uses image recognition technology to detect the abnormal situation of people falling down and issue an alarm based on the general surveillance video. However, most of the information processing methods are to distinguish the limbs in detail and define various fall movements in advance. Posture, and then identify the action in a way that compares features. However, in real life, the human body is tall and thin, and the body movements are very diverse. It is not easy to make a precise and universal definition, so that the prior art has poor recognition effect on abnormal conditions, often false alarms or unable to detect abnormal conditions. Failure to issue an alert in time.

Therefore, the object of the present invention is to provide a fall detection method capable of accurately recognizing a person falling down and performing related output in time using an image.

Thus, the fall detection method of the present invention is performed by a monitoring system including a camera that takes an image toward a target area, and a processing unit that receives images from the camera, the method comprising the following steps performed by the processing unit : (A) according to the image transmitted by the camera, it is recognized that a person enters the target area, and an abnormal number of times is set to 0; (B) the plural edge of the person is determined for a current image, and the edges are determined according to a predetermined plural The edge direction is classified, and the ratio of the number of edges whose direction is vertical to the total edge direction of the sum of all the edge numbers is calculated; (C) determining whether the vertical edge direction ratio is less than a preset ratio threshold, and if yes, performing step (D); (D) adding 1 to the abnormal number and recording the time corresponding to the current image; and (E) determining whether the time corresponding to the current image and the interval between the first abnormal time or the last abnormal time does not exceed a preset time length, If yes, proceed to step (F), if not, return to step (A); (F) determine whether the abnormal number is greater than a preset number of times, if If yes, go to step (G), if not, go back to step (B); and (G) determine that a fall has occurred and the relevant output is made.

Preferably, the step (B) further calculates a person's body angle representing the degree of inclination of the person, and the step (C) further determines whether the body angle of the person is greater than a predetermined angle threshold, and if the two determination conditions are met This step (D) is performed.

Preferably, in the step (B), the edges are classified into four edge directions of 0, 45, 90 and 135 degrees with respect to the vertical direction.

The invention has the advantages of comprehensive logic judgment using parameters such as vertical edge direction ratio, human body angle and time, and can accurately detect the fall of a person without consuming large calculation resources.

S11~S20‧‧‧Steps

S141~S143‧‧‧Steps

S151, S152‧‧‧ steps

Other features and advantages of the present invention will be apparent from the following description of the embodiments of the invention. FIG. 1 is a flow chart illustrating a preferred embodiment of the fall detection method of the present invention.

2 is an image diagram showing the case where no fall occurs; FIG. 3 is an image diagram showing the occurrence of a fall; FIG. 4 is a sequence image showing the detection result of a fall occurrence; and FIG. It is a sequence of image diagrams showing the results of detections without a fall.

Before the present invention is described in detail, it should be noted that in the following description, similar elements are denoted by the same reference numerals.

Referring to FIG. 1, a fall detection method of the present invention is executed by a monitoring system, and when a person enters a target area, it can start detecting whether a person has an abnormal condition of falling. The monitoring system includes a camera that images an image of a target area, and a processing unit that receives images from the camera. The method includes the following steps: Step S11: The camera continues to image the target area to obtain a sequence image.

The following steps are processed by the processing unit in accordance with the image transmitted by the camera.

Step S12 detects whether a person has entered, that is, whether a single or plural person enters the target area. This step is not the main technical feature of the present invention. The specific technical means is not limited to specific means. For example, foreground tracking or region-based tracking (Region-Based Tracking) can be used to detect the changed region by subtracting the current image from the background image. Further set the rules for screening; or use Contour-Based Tracking to find the contours and track them according to the changes in the contours; or use Feature-Based Tracking to capture the objects to be tracked first. Features such as center of gravity, area, etc., to track objects in comparison to features between successive images; or Model-Based Tracking, first to create an object model, build a motion model, and then search for Find objects in the continuous image. The person tracking of this step will generate a tracking box for the area where the person is located (as shown in the box in Figure 2).

In step S13, the number of abnormal times is set to 0, and the count parameter i is 0. Next, the vertical edge direction proportional condition determination in step S14 and the human body angle condition determination in step S15 are performed, and the condition judgment result is comprehensively judged in step S16. Step S14 includes steps S141 to S143, and step S14 includes steps S151 and S152.

Step S141 determines a contour line composed of a plurality of edges of the person for a current image. The specific calculation method is detailed below. The tracking frame generated by the personnel tracking executed in step S12 is used as the sampling range, and then the sampling area is divided into a plurality of small sampling areas, and Sobel gradient direction calculation is performed on each small sampling area, and each can be obtained. The distribution of the edge direction of the small sampling area, each pixel point is classified into four types of directional, 0°, 45°, 90°, 135°, and the like.

Then, the proportion of each of the four edge directions in each small sampling area is calculated, and the category with the highest proportion is the characteristic representing the small sampling area. If it belongs to the directional category, the small sampling area is an edge. Otherwise, it belongs to the non-directional category, not the edge. The distribution set of each edge is the contour.

Step S142 calculates a vertical edge number belonging to 0° among the edges, and a vertical edge direction ratio which is the sum of all the edge numbers.

Step S143: determining whether the vertical edge direction ratio is smaller than A preset ratio threshold is recorded, and the judgment result of this step, that is, the judgment result of the vertical edge direction proportional condition is recorded. The preset ratio threshold can be adjusted according to different scenes or image quality. For example, if a person in a scene has a vertical edge direction ratio of 1.0 to 0.5 if standing, then we can set a preset ratio threshold. 0.45, as long as the vertical edge direction ratio is less than the preset ratio threshold, the vertical edge direction proportional condition is achieved.

Step S151 - calculating a person's body angle representing the degree of inclination of the person. In this embodiment, the human body angle calculation is to approximate the foreground area where the person is located in an elliptical manner, and the angle of the ellipse is calculated to obtain the body angle of the person.

Step S152: determining whether the body angle of the person is greater than a predetermined angle threshold, and recording the judgment result of the step, that is, the judgment result of the human body angle condition. Similarly, the preset angle threshold can also be adjusted according to different scenes or image quality. For example, if the body angle of a scene is 0~40 when standing, then we can set 45 as the preset. The angle threshold of the person is achieved as long as the angle is greater than the preset angle threshold.

Step S16 - For the vertical edge direction proportional condition judgment result of step S143 and the human body angle condition judgment result of step S152, it is analyzed whether both conditions are satisfied. If so, it means that the current image capture time has occurred, so step S17 is performed. If not, the process returns to step S13, and the next image is taken to perform the condition determination of steps S14 and S15.

Step S17 - Adding the number of abnormalities by 1, incrementing the counting parameter i by 1, and recording the time T _i as the time corresponding to the current image.

Step S18: determining whether the time T _i corresponding to the current image and the interval between the first abnormal time T ₁ or the last abnormal time T _i-1 does not exceed a preset time length, and if so, the state in which the person falls may be continuous. Instead of the sporadic transient, the process proceeds to step S18. If not, the process returns to step S13 to return to zero, and the number of abnormalities is restarted. This embodiment is exemplified by T _i -T ₁ , and the preset time length is 3 seconds, but the invention is not limited thereto.

Step S19 - determining whether the abnormal number of times is greater than a preset number of times, if yes, indicating that the state of the current image capture is continuous, which may be a more severe fall state, so proceeding to step S20, if not, returning to step S13 Thereafter, the next image is taken and the condition determinations of steps S14 and S15 are performed.

Step S20 - determining that a person falls and performing related output, for example, causing an alarm of the target area to sound a beep, or displaying an abnormal condition on the display screen monitored by the coordinated medical staff.

Using the above calculation technique, the fall detection is performed on the images as shown in FIG. 2 and FIG. 3, and the three numbers above the square tracking frame in the figure represent the original value of the human body angle, the vertical edge direction ratio, and the abnormal number.

In terms of the human body, in Figure 2, the original value of the human body angle is -18°. Since the inclination of the person only needs to consider the angle between the human body and the vertical direction, the sign is not considered, in the sense of calculation. Yes Taking the absolute value, so the body angle of the person in FIG. 2 is 18°, and in FIG. 3, the human body angle is 75°, so the 75° of FIG. 3 is larger than the 18° of FIG. 2, and the person in FIG. 3 is more than the person in FIG. 2 It also approaches the state of falling.

In terms of the vertical edge direction ratio, 0.37 of FIG. 3 is smaller than 0.52 of FIG. 2, and the vertical edge direction ratio of the person in FIG. 3 is lower, and the person in FIG. 2 is also closer to the falling state.

Considering again for the sequence image shown in FIG. 4, each time the flow proceeds to step S17 during the process, the number of abnormalities gradually accumulates, and when the preset number of times is accumulated, an alarm is issued. For the sequence image shown in FIG. 5, since the behavior of the normal walking of the person does not fall, whenever the flow proceeds to step S17, the number of abnormalities does not accumulate, and therefore no alarm is issued.

In summary, the preferred embodiment of the fall detection method of the present invention utilizes parameters such as vertical edge direction ratio, human body angle and time to perform comprehensive logical judgment, and can accurately discover personnel without consuming large calculation resources. In the case of a fall, it is indeed possible to achieve the object of the present invention.

The above is only the preferred embodiment of the present invention, and the scope of the present invention is not limited thereto, that is, the simple equivalent changes and modifications made by the patent application scope and patent specification content of the present invention, All remain within the scope of the invention patent.

S11~S20‧‧‧Steps

S141~S143‧‧‧Steps

S151, S152‧‧‧ steps

Claims (3)

A fall detection method is performed by a monitoring system including a camera that takes an image toward a target area, and a processing unit that receives images from the camera, the method comprising the following steps performed by the processing unit: A) according to the image transmitted by the camera, it is recognized that a person enters the target area, and an abnormal number of times is set to 0; (B) determining the plural edge of the person for a current image, and the edges are in accordance with a predetermined complex edge direction To classify, calculate the ratio of the number of edges whose direction is vertical to the total edge direction of the sum of all the edges; (C) determine whether the vertical edge direction ratio is less than a preset ratio threshold, and if yes, perform step (D); Adding 1 to the abnormal number and recording the time corresponding to the current image; and (E) determining whether the time corresponding to the current image and the interval between the first abnormal time or the last abnormal time does not exceed a preset time length, and if so Step (F) is performed, if otherwise, return to step (A); (F) determine whether the abnormal number is greater than a predetermined number of times, and if yes, proceed to step (G), if The process returns to step (B); and (G) is determined for falls and the correlation output. The fall detection method according to claim 1, wherein the step (B) further calculates a person's body angle representing the degree of inclination of the person, the step (C) And determining whether the body angle of the person is greater than a predetermined angle threshold, and if the two determination conditions are all true, the step (D) is performed. The fall detection method according to claim 1, wherein the step (B) classifies the edges into four edge directions of 0 degrees, 45 degrees, 90 degrees, and 135 degrees from the vertical direction.