CN-117315785-B - Fall behavior detection method, device, equipment and computer readable storage medium

CN117315785BCN 117315785 BCN117315785 BCN 117315785BCN-117315785-B

Abstract

The application discloses a method, a device and equipment for detecting falling behaviors and a computer-readable storage medium, and belongs to the technical field of machine vision. The falling behavior detection method comprises the steps of obtaining a video stream, sampling the video stream to obtain a video frame, identifying the video frame through a preset deep learning algorithm model to obtain characteristic information of an identification object, determining a historical behavior state of the identification object according to the historical characteristic information, determining a current behavior state of the identification object according to the historical behavior state and the current characteristic information, and analyzing based on the historical behavior state and the current behavior state to obtain a falling behavior detection result. According to the application, the target behavior state in the falling event is analyzed and identified through the preset deep learning algorithm model, so that the key behavior state in the falling event is identified under the condition of limited model accuracy, and the whole falling event process is identified.

Inventors

CAO XIAOLEI
CHAO SHAN
DAI CHAO
FU JINWU

Assignees

苏州汇川控制技术有限公司

Dates

Publication Date: 20260508
Application Date: 20231013

Claims (7)

1. A fall behavior detection method, characterized in that the fall behavior detection method comprises: Acquiring a video stream, and sampling the video stream to obtain a video frame; the video frame is identified through a preset deep learning algorithm model to obtain feature information of an identification object, wherein the feature information comprises historical feature information and current feature information, the feature information further comprises a current coordinate frame, a historical coordinate frame, a current behavior category, a historical behavior category, a current skeleton sequence and a historical skeleton sequence, and historical behavior states corresponding to the current behavior category and the historical feature category comprise a squat state, a falling state and a standing state; Mapping the historical coordinate frame with the largest overlapping degree with the current coordinate frame to the same identification object; acquiring a historical skeleton sequence corresponding to the historical coordinate frame from the historical characteristic information as a historical skeleton sequence of the identification object; Determining standard point coordinates according to head coordinates in the current skeleton sequence, wherein if the head coordinates are not all 0 and have no out-of-range phenomenon, the average value of the nose, eye and ear coordinates in the current skeleton sequence is used as the standard point coordinates, and if the head coordinates are not all 0 or have the out-of-range phenomenon, the average value of the left shoulder and right shoulder coordinates in the current skeleton sequence is used as the standard point coordinates; Taking the vector difference between the standard point coordinates and the leg coordinate sequence in the current skeleton sequence as a current leg offset distance sequence; taking the sum of squares of differences between the current leg offset distance sequence and a historical leg offset distance sequence in the historical skeleton sequence as a time domain leg offset distance sequence of the identification object; The method comprises the steps of obtaining historical behavior states of an identification object from historical characteristic information, determining the current behavior state of the identification object according to the historical behavior states and the current characteristic information, wherein when the current behavior type is a squatting state, the current behavior type is converted into a falling state or a standing state based on the historical behavior state corresponding to a historical video frame adjacent to the current video frame; Comparing the continuous action consisting of the historical behavior state and the current behavior state with a preset falling standard to obtain a falling behavior detection result, wherein the preset falling standard comprises that a previous frame is standing and a next frame is falling, the previous frame is standing and the next frame is falling, the previous frame is falling and the next frame is standing, and the falling behavior detection result comprises that the user stands up after falling, continuously falling and falling.
2. The method for detecting fall behavior according to claim 1, wherein the predetermined deep learning algorithm model is a human body posture estimation model, and the step of identifying the video frame by the predetermined deep learning algorithm model to obtain feature information of an identified object comprises: And identifying the video frame through the human body posture estimation model to obtain a coordinate frame, a behavior category and a skeleton sequence of an identification object.
3. A fall behavior detection method as claimed in claim 1, wherein the predetermined deep learning algorithm model comprises a target detection algorithm and a skeleton recognition algorithm, and the step of recognizing the video frame by the predetermined deep learning algorithm model to obtain feature information of the recognition object comprises: identifying the video frame through a target detection algorithm to obtain a coordinate frame and a behavior category of an identification object; clipping a coordinate area including the identification object from the video frame based on the coordinate frame; And identifying the coordinate region through a skeleton identification algorithm to obtain a skeleton sequence of the identification object.
4. A fall behavior detection method as claimed in claim 1, wherein the step of determining the current behavior state of the recognition object from the historical behavior state and the current feature information further comprises: And under the condition that the current behavior category is a falling state, determining the current behavior state of the identification object according to the continuous occurrence frequency of standing states in the historical behavior state.
5. A fall behavior detection device, characterized in that the fall behavior detection device comprises: the acquisition module is used for acquiring a video stream and sampling the video stream to obtain a video frame; The recognition module is used for recognizing the video frame through a preset deep learning algorithm model to obtain feature information of a recognition object, wherein the feature information comprises historical feature information and current feature information, the feature information further comprises a current coordinate frame, a historical coordinate frame, a current behavior category, a historical behavior category, a current skeleton sequence and a historical skeleton sequence, and the historical behavior states of the current behavior category and the historical feature category respectively comprise a squat state, a falling state and a standing state; the computing module is used for mapping the historical coordinate frame with the largest overlapping degree with the current coordinate frame to the same identification object; acquiring a historical skeleton sequence corresponding to the historical coordinate frame from the historical characteristic information as a historical skeleton sequence of the identification object; Determining standard point coordinates according to head coordinates in the current skeleton sequence, wherein if the head coordinates are not all 0 and have no out-of-range phenomenon, the average value of the nose, eye and ear coordinates in the current skeleton sequence is used as the standard point coordinates, and if the head coordinates are not all 0 or have the out-of-range phenomenon, the average value of the left shoulder and right shoulder coordinates in the current skeleton sequence is used as the standard point coordinates; Taking the vector difference between the standard point coordinates and the leg coordinate sequence in the current skeleton sequence as a current leg offset distance sequence; taking the sum of squares of differences between the current leg offset distance sequence and a historical leg offset distance sequence in the historical skeleton sequence as a time domain leg offset distance sequence of the identification object; The method comprises the steps of obtaining historical behavior states of an identification object from historical characteristic information, determining the current behavior state of the identification object according to the historical behavior states and the current characteristic information, wherein when the current behavior type is a squatting state, the current behavior type is converted into a falling state or a standing state based on the historical behavior state corresponding to a historical video frame adjacent to the current video frame; the analysis module is used for comparing continuous actions formed by the historical behavior state and the current behavior state with a preset falling standard to obtain a falling behavior detection result, wherein the preset falling standard comprises that a previous frame stands and a next frame stands, the previous frame stands and the next frame stands, and the falling behavior detection result comprises that the falling user falls continuously and stands up after falling.
6. A fall behavior detection device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed by the processor carries out the steps of a fall behavior detection method as claimed in any one of claims 1 to 4.
7. A computer-readable storage medium, having stored thereon a computer program which, when executed by a processor, implements the steps of a fall behaviour detection method as claimed in any one of claims 1 to 4.

Description

Fall behavior detection method, device, equipment and computer readable storage medium Technical Field The application relates to the technical field of machine vision, in particular to a method, a device, equipment and a computer readable storage medium for detecting falling behaviors. Background In recent years, with the development of behavior recognition technology, there has been a great deal of attention as fall behavior recognition with high heat in visual scenes. At present, most of falling behavior recognition algorithms in the related art mainly perform instantaneous abnormal behavior recognition based on a space point or a period of time, but in some scenes targeting the whole process of an abnormal behavior event, the algorithms cannot meet the target requirement, the partial scenes hope that the algorithms can continuously recognize the target behavior and further obtain the key state information of the event, so that the algorithm model can be required to maintain the detection capability on a reference line all the time, and few models at present can ensure that falling postures of various forms can be recognized with high accuracy under any background. Disclosure of Invention The application mainly aims to provide a method, a device, equipment and a computer readable storage medium for detecting falling behaviors, which aim to solve the technical problem of identifying key behavior states in falling events under the condition of limited model accuracy, thereby identifying the whole falling event process. To achieve the above object, the present application provides a fall behavior detection method, including: Acquiring a video stream, and sampling the video stream to obtain a video frame; Identifying the video frame through a preset deep learning algorithm model to obtain feature information of an identification object, wherein the feature information comprises historical feature information and current feature information; acquiring a historical behavior state of the identification object from the historical characteristic information, and determining a current behavior state of the identification object according to the historical behavior state and the current characteristic information; And comparing the continuous action formed by the historical behavior state and the current behavior state with a preset falling standard to obtain a falling behavior detection result. Optionally, the preset deep learning algorithm model is a human body posture estimation model, the characteristic information comprises a coordinate frame, a behavior category and a skeleton sequence, and the step of identifying the video frame through the preset deep learning algorithm model to obtain the characteristic information of the identification object comprises the following steps: And identifying the video frame through the human body posture estimation model to obtain a coordinate frame, a behavior category and a skeleton sequence of an identification object. Optionally, the preset deep learning algorithm model comprises a target detection algorithm and a skeleton recognition algorithm, the characteristic information comprises a coordinate frame, a behavior category and a skeleton sequence, and the step of recognizing the video frame through the preset deep learning algorithm model to obtain the characteristic information of the recognition object comprises the following steps: identifying the video frame through a target detection algorithm to obtain a coordinate frame and a behavior category of an identification object; clipping a coordinate area including the identification object from the video frame based on the coordinate frame; And identifying the coordinate region through a skeleton identification algorithm to obtain a skeleton sequence of the identification object. Optionally, the coordinate frames comprise a current coordinate frame and a historical coordinate frame, the behavior categories comprise a current behavior category and a historical behavior category, and before the step of acquiring the historical behavior state of the identification object from the historical characteristic information, the falling behavior detection method further comprises: Mapping the historical coordinate frame with the largest overlapping degree with the current coordinate frame to the same identification object; The step of obtaining the historical behavior state of the identification object from the historical characteristic information comprises the following steps: and acquiring a historical behavior category corresponding to the historical coordinate frame from the historical characteristic information, and taking the historical behavior category as the historical behavior state of the identification object. Optionally, the skeleton sequence comprises a current skeleton sequence and a historical skeleton sequence, and after the step of mapping the historical coordinate frame with the largest overlapping degree wi