CN-121999408-A - Deep learning-based platform safety line grading early warning method and system

CN121999408ACN 121999408 ACN121999408 ACN 121999408ACN-121999408-A

Abstract

The invention discloses a platform safety line grading early warning method and system based on deep learning, and belongs to the technical field of computer vision and public safety. The method comprises the steps of initializing a system, acquiring video stream data of a platform monitoring area in real time, preprocessing images, utilizing a target detection and segmentation model to identify and position pedestrians in video frames, judging whether the pedestrians are overtime based on the foot position information of the identified pedestrians, utilizing a multi-task attribute identification model to identify the attributes of the overtime pedestrians, calculating the minimum overtime distance of the overtime pedestrians and the target density factor of the overtime area, generating comprehensive threat scores, generating a priority queue, and generating and executing self-adaptive multi-mode safety reminding. The system and the method realize the automatic, accurate and hierarchical management of the station line crossing behavior, can effectively improve the timeliness, accuracy and humanization level of early warning, and are suitable for the safety protection of public traffic scenes such as high-speed rail stations, subway stations and the like.

Inventors

LV SHUAISHUAI
ZHAI YIFEI
LU YIXUE
CHEN JI
NI HONGJUN
ZHU YU
WANG XINGXING
YAO JIANNAN
ZHANG FUBAO

Assignees

南通大学

Dates

Publication Date: 20260508
Application Date: 20251230

Claims (10)

1. The platform safety line grading early warning method based on deep learning is characterized by comprising the following steps of: s1, completing system initialization through camera calibration and warning line calibration; S2, video stream data of a platform monitoring area are obtained in real time and image preprocessing is carried out; S3, identifying and positioning pedestrians in the video frame by using a deep learning target detection and segmentation model; S4, judging whether to cross a preset safety warning line or not based on the identified foot position information of the pedestrian; s5, identifying attributes of the pedestrians beyond the line by utilizing a multi-task attribute identification model, wherein the attributes comprise age, gender and nationality; s6, calculating the minimum line crossing distance of the line crossing pedestrians and the target density factor of the line crossing region, generating comprehensive threat scores based on a dynamic priority algorithm, and sequencing all the line crossing pedestrians according to the comprehensive threat scores to generate a priority queue; And S7, generating and executing self-adaptive multi-mode safety reminding according to the priority queue and the attribute of the pedestrian.
2. The method of claim 1, wherein the object detection and segmentation model in step S3 employs a YOLOv-Seg model to obtain a bounding box and a pixel level mask for the pedestrian.
3. The method according to claim 1, wherein the step S4 of determining whether to cross the preset safety guard line specifically includes extracting a mask of a foot region of a pedestrian by using an example segmentation model, or obtaining ankle coordinates by using a human body key point detection model, and determining a positional relationship between the ankle coordinates and a polygon of the preset guard line.
4. The method according to claim 1, wherein the dynamic priority algorithm in step S6 specifically comprises: ranking risk according to minimum line-crossing distance Calculating the target number in the unit area of the line crossing area as a density factor By the formula Calculating comprehensive threat scores Generating a priority queue in descending threat score order, wherein And Is an adjustable weight coefficient.
5. The method according to claim 1, wherein the adaptive multi-modal alert in step S7 specifically comprises: Selecting a Chinese or English voice library according to the nationality attribute; generating personalized titles according to the age and sex attribute combination; selecting voice emergency degree and sound-light alarm intensity according to the threat score; And if the target density factor of the line crossing area is higher than a set threshold value, triggering group warning broadcasting.
6. The utility model provides a platform safety line hierarchical early warning system based on degree of depth study which characterized in that includes: the image acquisition module is used for acquiring a monitoring video stream of the platform area; the preprocessing module is connected with the image acquisition module and is used for carrying out image correction and enhancement processing on the video stream; The core analysis module is connected with the preprocessing module and is used for carrying out pedestrian target detection and instance segmentation on the processed video frames, judging whether pedestrians cross lines or not based on segmentation results, and carrying out attribute identification on the pedestrians crossing lines; the decision module is connected with the core analysis module and is used for calculating the comprehensive threat score and generating a priority queue according to the attribute of the line crossing pedestrians, the minimum line crossing distance and the target density of the line crossing area; And the execution module is connected with the decision module and used for generating and executing self-adaptive multi-mode safety reminding according to the priority queue and the pedestrian attribute.
7. The system of claim 6, wherein the core analysis module integrates a plurality of deep learning models including YOLOv-Seg models for object detection, openPose models for keypoint detection, and ResNet models for attribute classification.
8. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any one of claims 1 to 5.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the computer program is executed to implement the steps of the method according to any of claims 1 to 5.
10. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method of any one of claims 1 to 5.

Description

Deep learning-based platform safety line grading early warning method and system Technical Field The invention relates to the technical field of computer vision and public safety, in particular to a platform safety line grading early warning method and system based on deep learning. Background In scenes such as a high-speed rail station, a railway station and the like, a yellow safety guard line at the edge of the station is an important facility for guaranteeing the life safety of passengers. The method has the core effects of defining a safety area for a waiting passenger to prevent two main risks, namely that the passenger crosses a line due to crowding, playing or negligence during the waiting period of the train which is not yet in the station, the direct hidden danger of unexpected falling off a track exists, and the high-speed train can generate strong adsorption air flow and has extremely small gap with the station in the whole process of the train entering the station, so that the crossing passenger is involved in the air flow or has fatal danger of scratch with a car body. Currently, passengers are prevented from crossing the line mainly by means of visual observation and verbal reminding of platform staff. However, during peak traffic hours, it is difficult for staff to monitor all areas simultaneously, and leakage tends to occur. Some stations are provided with simple infrared or laser induction alarm devices, but the technologies cannot distinguish the attribute of an intrusion target, such as adults or children, domestic or foreign passengers, and cannot intelligently judge the priority of reminding when a plurality of people cross lines at the same time, so that the reminding efficiency is low, and passengers are easy to dislike due to false alarm (such as luggage touch). Disclosure of Invention The technical problem to be solved by the invention is to overcome the defects of the prior art, and provide the platform safety early warning method and the system which can automatically detect the line crossing behavior in real time and perform differentiation and priority voice prompt according to the target attribute, so that the efficiency and humanization level of platform safety management are obviously improved. In order to solve the problems, the invention adopts the following technical scheme: firstly, the invention provides a platform safety line grading early warning method based on deep learning, which comprises the following steps: s1, completing system initialization through camera calibration and warning line calibration; S2, video stream data of a platform monitoring area are obtained in real time and image preprocessing is carried out; S3, identifying and positioning pedestrians in the video frame by using a deep learning target detection and segmentation model; S4, judging whether to cross a preset safety warning line or not based on the identified foot position information of the pedestrian; s5, identifying attributes of the pedestrians beyond the line by utilizing a multi-task attribute identification model, wherein the attributes comprise age, gender and nationality; s6, calculating the minimum line crossing distance of the line crossing pedestrians and the target density factor of the line crossing region, generating comprehensive threat scores based on a dynamic priority algorithm, and sequencing all the line crossing pedestrians according to the comprehensive threat scores to generate a priority queue; And S7, generating and executing self-adaptive multi-mode safety reminding according to the priority queue and the attribute of the pedestrian. Preferably, the object detection and segmentation model in step S3 employs a YOLOv-Seg model to obtain a bounding box and a pixel-level mask for the pedestrian. Preferably, in the step S4, whether the preset safety guard line is crossed or not is judged, and specifically, the method comprises the steps of extracting a mask of a foot area of a pedestrian through an example segmentation model or obtaining ankle joint coordinates through a human key point detection model, and judging the position relation between the ankle joint coordinates and a polygon of the preset guard line. Preferably, the dynamic priority algorithm in step S6 specifically includes: ranking risk according to minimum line-crossing distance Calculating the target number in the unit area of the line crossing area as a density factorBy the formulaCalculating comprehensive threat scoresGenerating a priority queue in descending threat score order, whereinAndIs an adjustable weight coefficient. Preferably, the adaptive multi-modal alert in step S7 specifically includes: Selecting a Chinese or English voice library according to the nationality attribute; generating personalized titles according to the age and sex attribute combination; selecting voice emergency degree and sound-light alarm intensity according to the threat score; And if the target density factor of the line