CN117749986A - GNSS reference station personnel entry monitoring alarm method - Google Patents

Abstract

The invention discloses a method for monitoring and alarming personnel entering a GNSS reference station, which belongs to the technical field of monitoring and comprises the following specific steps: the invention has the advantages that when people enter the reference station, the intelligent level of the remote management reference station can be effectively improved, and the effect of practical verification is good.

Description

GNSS reference station personnel entry monitoring alarm method

Technical Field

The invention relates to the technical field of monitoring, in particular to a method for monitoring and alarming personnel entering a GNSS reference station.

Background

The integrated service system (HLJCORS) for continuous operation of the satellite positioning in Heilongjiang province consists of 148 Beidou satellite navigation positioning reference stations (simply referred to as reference stations) and 1 data center, is an important infrastructure for high-precision navigation positioning and coordinate frame maintenance, is built in 2016 and is in an unattended automatic operation state, 5000 differential service accounts are provided for more than 20 industries in the whole province, all-weather three-dimensional integrated centimeter-level real-time network RTK service is continuously provided, and remarkable social benefits are achieved.

In order to ensure long-term stable continuous operation of the HLJCORS system, the satellite navigation positioning service center of the first survey engineering institute of Heilongjiang continuously monitors and maintains the system. The reference station is equipped with a monitoring camera, but because most of the time the reference station is secure and unmanned, the center is usually not monitored by a dedicated person in real time and only looks back at the historical video if a problem is found with the reference station. The video review mode is time-consuming and low in efficiency, abnormal pictures are not easy to find, the safety risk of the reference station is increased to a certain extent, and the workload of personnel is increased. Therefore, the invention discloses a GNSS reference station personnel entering monitoring alarm method.

Disclosure of Invention

The present invention has been made in view of the above and/or problems associated with the entry of personnel into a monitoring and alerting method by a GNSS reference station.

Therefore, the present invention aims to provide a method for monitoring and alarming personnel entering a GNSS reference station, which can solve the above-mentioned existing problems.

In order to solve the technical problems, according to one aspect of the present invention, the following technical solutions are provided:

a GNSS reference station personnel entering monitoring alarm method comprises the following specific steps:

step one: the monitoring camera of the reference station adopts movement detection to carry out image detection, judges whether an image change exists or not, and uploads the image change to a designated folder on a server through FTP if the image change exists;

step two: after uploading, judging whether pedestrians move or not by adopting YOLO to perform image detection;

step three: if no pedestrian exists, deleting the picture without the pedestrian;

step four: if the pedestrians exist, cutting the pictures containing the pedestrians into other folders;

step five: and sending an alarm mail.

As a preferred embodiment of the present invention, a GNSS reference station personnel entry monitoring alarm method, wherein: the camera in the first step can send pictures and can perform movement detection.

As a preferred embodiment of the present invention, a GNSS reference station personnel entry monitoring alarm method, wherein: the YOLO uses Python language, based on PyTorch, openCV framework algorithm, and uses a simple and uncomplicated object detection model, and the analysis of the picture requires only a glance to predict the target object and its position on the input picture, the algorithm defines the target detection as a single regression problem, and each given image is divided into an S x S grid system, which is a subset or part of the image, and each grid identifies the object by predicting the number of bounding boxes of the object within the grid.

Compared with the prior art:

the invention can automatically remind the manager when the personnel enter the reference station, effectively improve the intelligent level of the remote management reference station and has good effect through actual verification.

Drawings

FIG. 1 is a schematic flow chart of the present invention;

FIG. 2 is a diagram illustrating a camera movement detection function according to the present invention;

FIG. 3 is a schematic diagram of the function of FTP image uploading of the camera according to the present invention;

FIG. 4 is a schematic diagram of the relationship among artificial intelligence, machine learning and deep learning according to the present invention;

FIG. 5 is a schematic diagram of a conventional machine learning algorithm according to the present invention;

FIG. 6 is a schematic diagram of a shallow network and a deep network according to the present invention;

FIG. 7 is a schematic diagram of a target detection algorithm based on deep learning according to the present invention;

FIG. 8 is a diagram showing the detection results of YOLO personnel according to the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

The invention provides a GNSS reference station personnel entering monitoring alarm method, referring to fig. 1-8, comprising the following specific steps:

step one: the monitoring camera of the reference station adopts movement detection to detect images, judges whether the images change, and uploads the images to a designated folder on a server through FTP if the images change, wherein the camera can send pictures and can perform movement detection;

step two: after uploading, judging whether pedestrians move or not by adopting YOLO for image detection, wherein YOLO adopts Python language, based on PyTorch, openCV frame algorithm, a simple and uncomplicated object detection model is adopted for analyzing pictures, a target object and the position of the target object on an input picture can be predicted only by looking at one eye, the algorithm defines the target detection as a single regression problem, each given image is divided into an S-S grid system, the grid system is a subset or a part of the image, and each grid identifies the object by predicting the number of boundary frames of the objects in the grid;

step three: if no pedestrian exists, deleting the picture without the pedestrian;

step four: if the pedestrians exist, cutting pictures containing the pedestrians into other folders, wherein the directory structure of the other folders is SendMail, monitoring ip, year and year accumulation;

step five: an alert mail is sent, titled station name discoverer activity.

The core key codes are as follows:

the method comprises the following steps:

regarding the camera:

taking the example of a sea-Kangwei video camera on a reference station of a satellite positioning continuous operation integrated service system in Heilongjiang province, the camera has a movement detection function, as shown in fig. 2. In addition, the camera was found to have a function of transmitting a picture through analysis, as shown in fig. 3. Based on the function, firstly starting a mobile detection option, transmitting all found moving pictures to a server of a data center through FTP, and then completing a personnel detection function of the pictures at the server side.

Person detection with respect to images:

no matter how complex the computer is, the basic logic of the current computer is always bit operations of 0 and 1, and by virtue of the hundreds of millions of times per second of the basic logic, the computer can rapidly complete complex mathematical operations, which is incomparable to human beings. However, if high semantic understanding, judgment, or even judgment is involved, the task of a few people in an image that is easy for humans is difficult for a computer to deal with, because this problem does not allow explicit mathematical rules to be established. In order to give computers human comprehensiveness and logic thinking, artificial intelligence disciplines are created, and among many algorithms for realizing artificial intelligence, machine learning is a fast developing one. The idea of machine learning is to let the machine automatically learn a rule from a large amount of data and use the rule to make predictions for unknown data. Deep learning is one of the technical branches of machine learning, and the relation of the three is shown in fig. 4.

Machine learning is an important approach to implementing artificial intelligence and is also the earliest developed artificial intelligence algorithm. Unlike conventional rule-based design algorithms, the key to machine learning is to find rules from a large amount of data and automatically learn parameters required by the algorithm. Machine learning was first seen in bayesian analysis of 1783, which is one type of machine learning, and derives the likelihood of occurrence from historical data of similar events. The most important of the machine learning algorithms is data, which can be classified into three main categories according to the data format used: supervised learning, unsupervised learning, and reinforcement learning.

The machine learning common algorithm is shown in fig. 5, while the deep learning is mainly to learn knowledge by building a deep artificial neural network, and the input data is usually complex, large in scale and high in dimension. The deep learning model can be divided into a convolutional neural network, a cyclic neural network, and a generative countermeasure network according to the difference of network structures.

Deep learning refers to a set of machine learning algorithms for applying various problems such as multi-layer neural network images and texts, as shown in fig. 6, the expression and processing capacity of the deep network are stronger, and the function effect of the shallow network can be obtained by fitting with few neurons, which can save more resources, but has the disadvantage of complex training, and a great amount of data and skills are required to train an excellent deep network.

In solving the problem of deep learning, it is more common to identify the problem non-chronology and analyze or predict the problem of chronology. To facilitate the structure, work training, and inference flow of the expression Model (Net/Model), a framework is created for correctly expressing the deep learning Model. The mainstream deep learning frameworks include TensorFlow, keras, caffe, pyTorch, CNTK, MXNet, DL, J, theano, torch7, caffe2, paddle, DSSTNE, tiny-dnn, chainer, neon, ONNX, bigDL, dyNet, brainstorm, coreML, and the like. These deep learning frameworks have different or characteristic features in terms of programming language, execution platform, model support, etc., and a suitable framework needs to be selected according to requirements.

The target detection algorithm based on deep learning is mainly divided into a double-stage detection algorithm and a single-stage detection algorithm, and as shown in fig. 7, the single-stage detection algorithm is a target detection algorithm based on regression analysis represented by YOLO (You Only Look Once).

YOLO adopts Python language, based on PyTorch, openCV and other frame algorithms, adopts a simple and uncomplicated object detection model, and can predict a target object and the position of the target object on an input picture only by 'seeing at a glance' when analyzing the picture. The algorithm defines object detection as a single regression problem, divides each given image into an S x S grid system that is a subset or portion of the image, each grid identifies objects by predicting the number of bounding boxes of objects within the grid, and through continued development, the current YOLO version is V5. The installation of YOLOv5 requires a Python version above 3.7.0 and a PyTorch version above 1.7, and other dependent packages are as follows:

YOLOv5 already contains a trained detection model, can be used for personnel detection, and can be used for executing a 'detect. Py' script under an installation directory, wherein a parameter_opt function in the script is mainly used for setting operation parameters, and specific codes are as follows:

/>

/>

the main parameters in the above codes function:

weight is a trained model path, a default official network model yolv5 s.

source, namely detecting an image path to be detected, and defaulting data/images;

data, configuring a classification name file path which comprises information such as image/label/class and the like;

imgsz, input picture size, default size is 640;

conf-thres, confidence threshold, default to 0.25;

IOU-thres, the IOU threshold of NMS, defaults to 0.45.NMS represents Non-maximum suppression (Non-Maximum Suppression), IOU represents the cross-over ratio (Intersection over Union), and NMS is typically used to remove boxes with IOU values above a certain threshold;

max-det, the number of the maximum detection frames reserved, and the number of detection targets in each picture is 1000 at most;

device, setting device CPU/CUDA, without setting;

view-img, whether showing the predicted picture, defaulting to False;

save-txt, whether to save predicted frame coordinates in txt file form, default False. If so, generating a txt file of each picture prediction under a path run/detect/exp/labels/. Txt;

save-conf, whether the confidence conf is also saved in txt, default False;

save-crop whether to save the picture of the cutting prediction frame, defaulting to False;

nosave, not saving the picture, defaulting to False;

setting and only reserving a certain part of categories, namely 0 or 0 2 3, and using-categories=n, wherein the picture stored under the path run/detect/exp is the category corresponding to n, and data needs to be set at the moment;

performing NMS to remove frames among different categories, and defaulting False;

predicting whether data enhanced TTA is also adopted, and defaulting to False;

visual, whether the network layer output characteristics are visualized, defaulting False;

and updating, namely if True, carrying out strip_optimizer operation on all models, and removing information such as optimizers in the pt file. Default to False;

project, namely saving a folder path of the detection result;

name, namely saving the name of the detection result folder;

exists_ok whether to re-create the folder, and re-create the folder when False;

line-thickness, line thickness of picture frame, defaulting to 3;

hide the predicted category in visualization, default False;

hide confidence in visualization, default False;

half, whether half precision Float16 is used for reasoning, so that the reasoning time can be shortened, and False is defaulted;

DNN predicted with OpenCV DNN.

After performing YOLO person detection, an image of the person is found as shown in fig. 8.

Although the invention has been described hereinabove with reference to embodiments, various modifications thereof may be made and equivalents may be substituted for elements thereof without departing from the scope of the invention. In particular, the features of the disclosed embodiments may be combined with each other in any manner as long as there is no structural conflict, and the exhaustive description of these combinations is not given in this specification merely for the sake of omitting the descriptions and saving resources. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed, but that the invention will include all embodiments falling within the scope of the appended claims.

Claims (3)

1. A GNSS reference station personnel entering monitoring alarm method is characterized by comprising the following specific steps:

step one: the monitoring camera of the reference station adopts movement detection to carry out image detection, judges whether an image change exists or not, and uploads the image change to a designated folder on a server through FTP if the image change exists;

step two: after uploading, judging whether pedestrians move or not by adopting YOLO to perform image detection;

step three: if no pedestrian exists, deleting the picture without the pedestrian;

step four: if the pedestrians exist, cutting the pictures containing the pedestrians into other folders;

step five: and sending an alarm mail.

2. The method of claim 1, wherein the camera in the first step is capable of transmitting pictures and performing motion detection.

3. The method of claim 1, wherein YOLO uses Python language, and based on PyTorch, openCV frame algorithm, uses a simple and uncomplicated object detection model, and the analysis of the picture requires only a single glance to predict the target object and its position on the input picture, and the algorithm defines the target detection as a single regression problem, and divides each given image into an S x S grid system, which is a subset or part of the image, and each grid identifies the object by predicting the number of bounding boxes of the object in the grid.