Pedestrian tumble detection system and method based on intelligent lamp pole
By using cameras on smart light poles and an improved YOLOv3 convolutional neural network, combined with deep learning algorithms, pedestrian fall detection has been achieved without requiring pedestrians to wear sensors. This solves the problems of inaccurate identification and high hardware costs in existing systems, and provides fast and accurate fall recognition and real-time alarm functions.
Patent Information
- Application Number
- CN202511114140.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-11
- Publication Date
- 2025-11-25
AI Technical Summary
Existing smart street light systems struggle to quickly and accurately identify pedestrian falls, leading to delays in rescue efforts. Furthermore, some systems require pedestrians to wear sensors, limiting their applicability to various scenarios.
A pedestrian fall detection system based on smart light poles is adopted. It collects video data through cameras and uses an improved YOLOv3 convolutional neural network and deep learning algorithm, combined with a threshold method, to detect and judge whether pedestrians have fallen in real time. The system includes data acquisition, preprocessing, target detection, behavior calculation and alarm modules, realizing fall detection without requiring pedestrians to wear sensors.
It enables rapid and accurate identification of pedestrian falls in smart cities, providing safety assurance for multiple scenarios, reducing hardware costs, meeting the needs of urban safety management, and providing real-time and accurate technical support.
Smart Images
Figure CN121010945A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of intelligent monitoring and behavior recognition, and particularly relates to a pedestrian fall detection system and method based on a smart lamp pole. BACKGROUND
[0002] In the prior art, most of the fall detection researches are based on sensors, but this technology usually requires installing sensor nodes on the human body, which causes certain inconvenience to the life of the human body, and the detection range is small.
[0003] Based on the process of the vigorous development of smart cities, the smart lamp pole as an important part of urban infrastructure, its functions continue to expand. Specifically, the current smart lamp pole not only bears the basic lighting function, but also integrates high-definition cameras, various sensors and other intelligent devices, and has multiple monitoring capabilities.
[0004] However, in terms of pedestrian fall detection, the existing smart lamp pole system has obvious deficiencies, and it is difficult to quickly and accurately identify fall events, which leads to delayed rescue and cannot meet the needs of smart cities for pedestrian safety protection. At the same time, part of the pedestrian detection system requires the pedestrian to wear corresponding sensors, and the scene applicability is not high, so the existing technology has certain limitations. SUMMARY
[0005] The present application is to solve the above problems, and aims to provide a pedestrian fall detection system and method based on a smart lamp pole.
[0006] The present application provides a pedestrian fall detection system based on a smart lamp pole, which has the following characteristics: a data acquisition module, a data processing module, a target detection module, a behavior calculation module, and a judgment alarm module, the data acquisition module and the data processing module are in communication connection, the data processing module and the target detection module are in communication connection, the target detection module and the behavior calculation module are in communication connection, and the behavior calculation module and the judgment alarm module are in communication connection,
[0007] The data acquisition and processing module is used for acquiring monitoring video data through the camera on the smart lamp pole, and transmitting the monitoring video data to the data processing module,
[0008] The data processing module pre-processes the monitoring video data,
[0009] The target detection module is used for detecting and tracking the human target in the pre-processed monitoring video data through a convolutional neural network, and obtaining the bounding box information of the human target,
[0010] The behavior calculation module is used for calculating the width-height ratio, effective area ratio, and center point change of the human target according to the bounding box information, analyzing the posture and action state of the human target, and obtaining the behavior characteristics of the human target,
[0011] The alarm judgment module is used for judging whether the human target falls down according to the behavior characteristics of the human target and the set threshold value, and if the human target is detected to fall down, an alarm signal is immediately sent out, and data is uploaded to a supervision data center.
[0012] The application further provides a pedestrian fall detection method based on a smart lamp pole, which is suitable for the pedestrian fall detection system based on the smart lamp pole and has the following characteristics.
[0013] S1, video data of a monitoring area is collected in real time through a camera on the smart lamp pole, and the video data is subjected to gray scale and denoising processing to obtain high-quality video data;
[0014] S2, the human target is quickly located in the video frames of the high-quality video data by using an improved YOLOv3 convolutional neural network, and the human target is continuously tracked in the continuous video frames by using a tracking algorithm, so that the bounding box information of the human target is accurately obtained;
[0015] S3, the bounding box information of the human target is calculated and analyzed to obtain the width-height ratio, the effective area ratio and the center point change of the human target in each video frame, and the behavior characteristics of the human target are obtained;
[0016] S4, the behavior characteristics of the human target obtained by calculation and analysis are compared and analyzed with the pre-set threshold value, and whether the human target falls down is judged;
[0017] S5, whether an alarm signal is sent out is determined according to the judgment result, if the human target is detected to fall down, an alarm signal is immediately sent out, and data is uploaded to a supervision data center.
[0018] Further, in step S1, the video data is subjected to denoising processing by using a median filter algorithm or a Gaussian filter algorithm.
[0019] Further, in step S2, the bounding box information is a minimum circumscribed rectangular frame range formed according to the position of the human target in each video frame.
[0020] Further, in step S3, the effective area ratio is the ratio of the pixel area of the human target to the pixel area of the minimum circumscribed rectangular frame.
[0021] Further, in step S3, the center point change is the distance and the slope k=(y2-y1) / (x2-x1) of the center coordinates (x1, y1) and (x2, y2) of the human target in the front and rear two continuous video frames.
[0022] Effects of the application
[0023] The pedestrian fall detection system and method based on the intelligent lamp pole provided by the application can determine whether a pedestrian falls without the pedestrian wearing a sensor, is suitable for various scenes of pedestrian safety guarantee in a smart city, and provides effective technical support for city safety management. Moreover, the improved YOLOv3 convolutional neural network is used to extract the target frame of the pedestrian, and the deep learning algorithm and the threshold method are combined to meet the real-time requirement and have good real-time performance. In addition, the pedestrian fall detection system and method based on the existing intelligent lamp pole has the advantages of low deployment cost without additional hardware cost. BRIEF DESCRIPTION OF DRAWINGS
[0024] Figure 1 is a structure diagram of the improved YOLOv3 convolutional neural network involved in the application;
[0025] Figure 2 is a feature comparison diagram of the width-height ratio of the human body when standing and falling;
[0026] Figure 3 is an effective area comparison diagram of the pixel area and the area of the circumscribed rectangle of the human body under different actions;
[0027] Figure 4 is a feature comparison diagram of the change of the center coordinates of the human body in the front and rear frames;
[0028] Figure 5 is a fall judgment flowchart based on the comparison and analysis of the behavior features and the threshold value. DETAILED DESCRIPTION
[0029] In order to make the technical means, creative features, purposes and effects achieved by the application easy to understand, the application is specifically described below in combination with the drawings.
[0030] EMBODIMENT
[0031] The embodiment provides a pedestrian fall detection system based on an intelligent lamp pole, which comprises a data acquisition module, a data processing module, a target detection module, a behavior calculation module and a judgment alarm module, the data acquisition module is in communication connection with the data processing module, the data processing module is in communication connection with the target detection module, the target detection module is in communication connection with the behavior calculation module, the behavior calculation module is in communication connection with the judgment alarm module,
[0032] The data acquisition and processing module is used for acquiring monitoring video data through a camera on the intelligent lamp pole and transmitting the monitoring video data to the data processing module,
[0033] The data processing module is used for preprocessing the monitoring video data,
[0034] The target detection module is configured to detect and track human targets in the preprocessed monitoring video data through a convolutional neural network, to obtain bounding box information of the human targets,
[0035] The behavior calculation module is configured to calculate the width-height ratio, effective area ratio, and center point change of the human targets according to the bounding box information, to analyze the postures and action states of the human targets, and to obtain behavior features of the human targets,
[0036] The judgment and alarm module is configured to judge whether the human targets fall according to the behavior features of the human targets in combination with a set threshold, to immediately issue an alarm signal if the human targets are detected to fall, and to upload data to a supervision data center.
[0037] The embodiment also provides a pedestrian fall detection method based on a smart lamp pole, which is suitable for the pedestrian fall detection system based on the smart lamp pole and includes the following steps:
[0038] S1. Video data of a monitoring area is collected in real time through a camera on a smart lamp pole, and the video data is subjected to grayscale and denoising processing to obtain high-quality video data.
[0039] S2. A human target is quickly located in a video frame of the high-quality video data by using an improved YOLOv3 convolutional neural network, and the human target is continuously tracked in consecutive video frames by using a tracking algorithm, so as to accurately obtain bounding box information of the human target.
[0040] S3. The bounding box information of the human target is calculated and analyzed to obtain the width-height ratio, effective area ratio, and center point change of the human target in each video frame, so as to obtain behavior features of the human target.
[0041] S4. The behavior features of the human target obtained by calculation and analysis are compared and analyzed with a pre-set threshold, to judge whether the human target falls.
[0042] S5. Whether to issue an alarm signal is determined according to a judgment result, and an alarm signal is immediately issued if the human target is detected to fall, and data is uploaded to a supervision data center.
[0043] In step S1, the high-definition camera mounted on the intelligent lamp pole is used to collect video data of the monitoring area in real time, and then the collected video data is preprocessed. Specifically, the grayscale processing technology is used to convert the color video into a grayscale image, reduce the data dimension, and improve the subsequent processing efficiency; at the same time, advanced de-noising algorithms such as median filtering and Gaussian filtering are used to remove noise interference in the video frame, improve the image definition and quality, and provide high-quality data for subsequent human target detection and tracking. In this embodiment, the median filtering algorithm or the Gaussian filtering algorithm is preferably used for de-noising the video data.
[0044] In step S2 of this embodiment, the bounding box information is a circumscribed rectangular frame range formed according to the position of the human target in each video frame.
[0045] In step S3 of this embodiment, the calculated and analyzed human width-height ratio is the width-height ratio of the bounding box of the target human in each video frame. Specifically, as shown in Figure 3 , when the human target falls, the body posture changes significantly, and the width-height ratio will be much greater than 1, which is one of the important bases for judging the fall.
[0046] In step S3 of this embodiment, the calculated and analyzed effective area ratio is the ratio of the pixel area of the human target to the pixel area of the smallest circumscribed rectangular frame. Specifically, as shown in Figure 4 , it can be seen that this ratio can effectively distinguish between stretching and falling actions of the human body, because the form and area change of the human body in contact with the outside during the falling process are different from those of the simple stretching action, and the calculation of the effective area ratio can avoid misjudgment caused by similar actions.
[0047] In step S3 of this embodiment, the calculated and analyzed center point change is the distance and slope k = (y2-y1) / (x2-x1) of the center coordinates (x1, y1) and (x2, y2) of the human target in the previous and subsequent two consecutive video frames. Specifically, as shown in Figure 5 , when the human target falls, the body moves quickly, the center point distance increases, and the slope becomes larger; while the squatting action may have a larger distance |AB| between the two points, but the slope k tends to infinity, and through these accurate calculations and feature differences, the squatting and falling actions can be accurately distinguished.
[0048] Based on the above, in step S4 of this embodiment, the pre-set threshold refers to the human width-height ratio, effective area, and center point position of the target human in the falling state.
[0049] In the embodiment, step S5 is to determine whether to trigger the alarm mechanism based on the obtained fall determination result, generate an alarm signal, and transmit the alarm information to the supervision data center in real time through the communication module built in the smart lamp pole, such as a 4G / 5G communication module, a Wi-Fi module, etc. Specifically, if the human target is detected to fall, the alarm mechanism is triggered immediately, an alarm signal is sent, and data is uploaded to the supervision data center; if the human target is not detected to fall, it is skipped. In the embodiment, in order to ensure the accuracy and integrity of information transmission, data encryption technology can also be used to encrypt the alarm information to prevent information from being stolen or tampered with during transmission, and to provide reliable information support for the smooth development of rescue work.
[0050] Effects of the embodiments
[0051] The pedestrian fall detection system and method based on the smart lamp pole according to the present application can determine whether a pedestrian falls without the pedestrian wearing a sensor, is suitable for various scenes of pedestrian safety guarantee in a smart city, and provides effective technical support for city safety management. Moreover, the present application extracts the pedestrian target frame using the YOLOv3 convolutional neural network, and combines the deep learning algorithm and the threshold method to meet the real-time demand while having good real-time performance. In addition, the present application is a pedestrian fall detection system and method based on the existing smart lamp pole, so it does not require additional hardware costs and has the advantage of low deployment cost.
[0052] The above embodiments are preferred cases of the present application and do not limit the protection scope of the present application.
Claims
1. A smart pole-based pedestrian fall detection system, characterized in that, Comprise: A data acquisition module, a data processing module, a target detection module, a behavior calculation module, and a judgment alarm module, the data acquisition module is connected with the data processing module, the data processing module is connected with the target detection module, the target detection module is connected with the behavior calculation module, the behavior calculation module is connected with the judgment alarm module, The data acquisition processing module is used for collecting monitoring video data through the camera on the intelligent lamp pole, and transmitting the monitoring video data to the data processing module, The data processing module pre-processes the monitoring video data, The target detection module is used for detecting and tracking the human target in the pre-processed monitoring video data through a convolutional neural network, and obtaining the bounding box information of the human target, The behavior calculation module is used for calculating the aspect ratio, effective area ratio, and center point change of the human target according to the bounding box information, analyzing the posture and action state of the human target, and obtaining the behavior characteristics of the human target, The judgment alarm module is used for judging whether the human target falls according to the behavior characteristics of the human target combined with the set threshold value, and if the human target falls is detected, an alarm signal is immediately sent, and the data is uploaded to the supervision data center.
2. A pedestrian fall detection method based on a smart lamp pole, applicable to the pedestrian fall detection system based on the smart lamp pole in claim 1, characterized in that, Comprise the following steps: S1, through the camera on the intelligent lamp pole, real-time collection of video data of monitoring area, and the video data is grayed, denoising to obtain high-quality video data; S2, using an improved YOLOv3 convolutional neural network to quickly locate the human target in the video frames of high-quality video data, and realizing the continuous tracking of the human target in the continuous video frames through a tracking algorithm, and accurately obtaining the bounding box information of the human target; S3, calculating and analyzing the bounding box information of the human target to obtain the aspect ratio, effective area ratio, and center point change of the human target under each video frame, and obtaining the behavior characteristics of the human target; S4, comparing and analyzing the behavior characteristics of the human target obtained by calculation and analysis with the pre-set threshold value to judge whether the human target falls; S5, according to the judgment result, it is decided whether to send an alarm signal, if the human target falls is detected, an alarm signal is immediately sent, and the data is uploaded to the supervision data center.
3. The pedestrian fall detection method based on intelligent lamp pole according to claim 2, characterized in that: In step S1, the video data is denoised by using a median filter algorithm or a Gaussian filter algorithm.
4. The pedestrian fall detection method based on intelligent lamp pole according to claim 2, characterized in that: In step S2, the bounding box information is an external rectangular frame range formed according to the position of the human target in each video frame.
5. The pedestrian fall detection method based on intelligent lamp pole according to claim 2, characterized in that: In step S3, the effective area ratio is the ratio of the pixel area of the human target to the pixel area of the minimum external rectangular frame.
6. The pedestrian fall detection method based on intelligent lamp pole according to claim 2, characterized in that: In step S3, the center point is changed to the distance between the center coordinates (x1, y1) and (x2, y2) of the human target in the two consecutive video frames before and after, and the slope k = (y2 - y1) / (x2 - x1).