CN110826378A

CN110826378A - Edge calculation dynamic soil scene recognition algorithm based on deep learning and intelligent terminal

Info

Publication number: CN110826378A
Application number: CN201810917866.7A
Authority: CN
Inventors: 景象
Original assignee: Nanjing Huicheng Technology Co Ltd
Current assignee: Nanjing Huicheng Technology Co Ltd
Priority date: 2018-08-13
Filing date: 2018-08-13
Publication date: 2020-02-21

Abstract

The invention provides an edge calculation dynamic soil scene recognition algorithm based on deep learning and an intelligent terminal. The invention obtains the picture of the front-end camera by timing; performing moving-soil scene detection on the acquired image, detecting whether a target scene exists in the image, and framing a target area; and transmitting the analysis result to the back end at regular time. The dynamic state of illegal moving soil and illegal construction can be detected in time, the spanning development of a government department from a traditional supervision mode to a modern supervision mode is promoted, and real-time and efficient comprehensive information and supervision basis are provided for moving soil supervision.

Description

Edge calculation dynamic soil scene recognition algorithm based on deep learning and intelligent terminal

Technical Field

The invention belongs to the technical field of informatization solutions, and particularly relates to a method for providing moving-earth scene recognition for government industry departments such as environmental protection and residential construction and a corresponding intelligent terminal.

Background

With the development of science and technology, especially the explosive growth of the internet of things and artificial intelligence in recent years, various industries in China face opportunities and challenges of industry upgrading. As a traditional industry in China, the construction industry generally has the problems of extensive production supervision modes, low production efficiency, light weight, safety, generally low quality of construction workers, backward supervision means of continuous education left in forms, field persons, machines, materials and the like, and the like.

The soil-working operation supervision is executed according to the chemical industry standard ' safety regulations for soil-working operation in plant areas ' of the people's republic of China. The "safety work certificate of soil movement" must be dealt with in the soil movement work. All above-ground operations affecting the safety of underground cables, pipelines and other facilities are included in the scope of earth moving operations. Such as earth excavation, pile driving, grounding electrode burying and the like, which are carried out more than a certain depth; greening plants, arranging large-scale placards, propagandizing galleries, discharging a large amount of sewage and the like, which affect the operation of underground facilities; filling soil or leveling the field by using construction machinery such as a bulldozer, a road roller and the like; when heavy objects are stacked in places other than the specified places or heavy materials are transported in the boundary areas of the places other than the regular roads, the earth moving operation is considered when the total weight of the objects including the transport vehicles carried by the objects is more than the limited value. The limit values of the heaps, loads and total carrying amount are determined according to the soil quality.

At present, the supervision of the domestic homework is still in a primary stage, a plurality of steps and links depend on a traditional manual supervision mode, all the links of the homework supervision are not effectively associated to realize a closed loop, the application of various technologies is in a state of independent operation, the sharing of data and the transmission of information among all the steps are not perfect, the supervision cost is high, and the working efficiency of the system is low.

Disclosure of Invention

The invention aims to provide an edge calculation soil-shifting scene recognition algorithm based on deep learning and an intelligent terminal, which are used for carrying out timing intelligent analysis and detection on real-time video information of a front-end camera based on deep learning, carrying out detection and recognition on soil-shifting events and scenes by analyzing the behavior of a moving target and transmitting picture results.

In order to achieve the purpose, the invention provides the following technical scheme:

in one aspect, the invention provides an edge calculation dynamic soil scene recognition algorithm based on deep learning, which comprises the following steps: acquiring a picture of a front-end camera at fixed time; carrying out moving soil scene detection on the obtained image; and transmitting the analysis result to the back end at regular time.

Further, the step of obtaining the picture of the front-end camera at regular time includes the following steps: the camera is directly butted with the cameras, video images of all the cameras are automatically obtained through timing polling, and the polling interval is adjustable within 2-10 minutes.

Further, the detection of the moving-soil scene of the acquired image comprises the following steps: and detecting whether a target scene exists in the image, namely whether a moving soil scene exists.

Further, the detection of the moving-soil scene of the acquired image comprises the following steps: selecting a frame of the target area, and selecting the position of the target object if the detection result is that the target object exists; and if the detection result is that no target object exists, uploading the original image.

Further, the timing of transmitting the analysis result to the back end includes the following steps: and uploading the identified result and the picture to a back-end application platform at a fixed time interval for display.

Further, the method also comprises the following steps of preventing repeated alarming: and comparing the picture identified as the target object with the last picture identified as the target object, and if the picture is the same target object, not repeatedly alarming.

On the other hand, the invention provides an edge computing dynamic scene recognition intelligent terminal based on deep learning, which comprises a shell and a control panel arranged in the shell, wherein a circuit is arranged on the control panel and comprises an ARM CPU, a FLASH, a memory and a TCP/IP interface; the ARM CPU is a core module of the intelligent terminal and is used for detecting and identifying a moving soil scene through a deep learning algorithm; the FLASH and the memory are used for storing and caching the picture of the front-end camera and the analysis intermediate data and the analysis result data of the ARMCPU, and are respectively connected with the ARM CPU to realize the read-write processing of the data; and the TCP/IP interface is respectively connected with the ARM CPU and the memory and used for acquiring pictures from the front-end camera and transmitting the analysis result to the back end.

Furthermore, the device also comprises a 1.25V generating circuit, which is used for converting the 3.3V voltage into the 1.25V voltage through voltage stabilization for the circuit.

Furthermore, the system also comprises a power line interface and a power indicator lamp which are respectively connected with the ARM CPU.

Due to the adoption of the scheme, the invention has the beneficial effects that:

the edge calculation moving soil scene recognition algorithm based on the deep learning and the intelligent terminal realize the edge calculation moving soil scene recognition based on the deep learning, are used for detecting the real-time image of the front-end camera and detecting whether a moving soil scene exists in the current image. And acquiring picture analysis at regular time, and transmitting the result to a back-end platform. The method mainly has the following three functions: 1. acquiring pictures of a front-end camera in a fixed time; 2. performing moving-soil scene detection on the acquired image, detecting whether a target scene exists in the image, and framing a target area; 3. and transmitting the analysis result to the back end at regular time.

The application of the invention can improve the safety level of the construction site, reduce the risk of breaking the moving soil, reduce the investment of manpower and capital, unify the standard, improve the supervision efficiency, promote the spanning development of enterprises and government departments from the traditional supervision mode to the modern supervision mode, and provide real-time and efficient comprehensive information and supervision basis for the moving soil supervision.

Drawings

FIG. 1 is a schematic block diagram of the control board circuitry of the present invention.

FIG. 2 is a schematic diagram of an ARM CPU related circuit of the present invention.

Fig. 3 is a schematic diagram of a 1.25V generation circuit in the present invention.

FIG. 4 is a schematic diagram of the FLASH and memory related circuits of the present invention.

Fig. 5 is a schematic diagram of an interface circuit of the present invention.

FIG. 6 is a diagram of an example of a real-time video of a functional display of the present invention.

FIG. 7 is a diagram of an example of the result of the functional display of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The concept of deep learning stems from the study of artificial neural networks. A multi-layer perceptron with multiple hidden layers is a deep learning structure. Deep learning forms a more abstract class or feature of high-level representation properties by combining low-level features to discover a distributed feature representation of the data.

Cloud computing and terminal computing are limited by bandwidth and computing power, respectively, and fall into bottlenecks. The concept of edge calculation finds a balance between the two. The edge calculation based on deep learning has important application and development prospects in the field of moving-earth work supervision.

The invention relates to an edge calculation dynamic soil scene recognition algorithm based on deep learning, which comprises the following steps:

s1, regularly acquiring the front-end camera picture;

s2, detecting the moving soil scene of the acquired image;

the method comprises the following two sub-steps: s2.1, detecting whether a target scene exists in the image; s2.2, selecting a frame of the target area;

and S3, periodically transmitting the analysis result to the back end.

The method is realized by an intelligent terminal for recognizing the dynamic scene based on the edge calculation of the deep learning, the intelligent terminal comprises a shell and a control panel arranged in the shell, the control panel is a main board which is independently developed, and the composition of a circuit on the control panel is shown in figure 1 and comprises an ARM CPU, a 1.25V generating circuit, a FLASH, a memory, a TCP/IP interface, a power line interface and a power indicator lamp.

Referring to fig. 2, the ARM CPU of the present invention is a core module of the intelligent terminal, and is used for detecting and identifying a soil-working scene through a deep learning algorithm; referring to fig. 3, the present invention further includes a 1.25V generating circuit for converting the 3.3V voltage into a 1.25V voltage for use by the circuit; referring to fig. 4, the FLASH and the memory of the invention are used for storing and caching the picture of the front-end camera and the analysis intermediate data and the analysis result data of the ARM CPU, and are respectively connected with the ARM CPU to realize the read-write processing of the data; referring to fig. 5, the TCP/IP interface of the present invention is respectively connected to the ARM CPU and the memory, and is configured to obtain a picture from the front-end camera and transmit an analysis result to the back end, and further includes a power line interface and a power indicator, which are respectively connected to the ARM CPU, and are configured to supply power to the terminal and indicate the power.

The intelligent terminal takes an ARM CPU as a core, is embedded with a deep learning algorithm, has 4 cores of 1.4G main frequency, 8GFlash, MAIL400GPU, a DDR 1G memory and DC 12V voltage, and comprises a network cable port interface, a power line interface and a power indicator lamp.

The working mode of the intelligent terminal is as follows:

real-time video image acquisition and round robin: the intelligent terminal is directly connected with the cameras in a butt joint mode, and the video images of all the cameras are obtained through automatic timing polling. The polling interval is adjustable within 2-10 minutes.

Detecting a characteristic algorithm and automatically identifying scenes: the intelligent terminal captures a front-end video image at a fixed time interval to perform target retrieval, and detects whether the image contains characteristics of illegal moving soil (such as an excavator and the like).

Uploading the recognition result: and uploading the identified result (whether illegal soil movement exists) and the picture (the frame selects the position of the target object if the result is that the target object exists; and the original picture is uploaded if the result is that the target object does not exist) to a rear-end application platform for display by the intelligent terminal at a fixed time interval.

And (3) repeated alarm prevention: the intelligent terminal compares the picture identified as the target object with the last picture identified as the target object, and if the pictures are the same target object, the intelligent terminal does not give an alarm repeatedly.

Fig. 6 and 7 illustrate the functionality of the present invention. Referring to fig. 6, at this time, a moving soil scene appears in the real-time video picture, that is, an excavator is set up to perform moving soil operation. Referring to fig. 7, the method and the intelligent terminal of the invention can identify the analysis result, i.e. identify the scene on the picture as the soil movement, and automatically frame the main excavator for soil movement operation.

The embodiments described above are described to facilitate understanding and application of the present patent to those of ordinary skill in the art. It will be readily apparent to those skilled in the art that various modifications to these embodiments may be made, and the generic principles described herein may be applied to other embodiments without the use of the inventive faculty. Therefore, the present invention is not limited to the embodiments described herein, and those skilled in the art should make improvements and modifications within the scope of the present invention based on the disclosure of the present invention.

Claims

1. An edge calculation dynamic soil scene recognition algorithm based on deep learning is characterized by comprising the following steps:

acquiring a picture of a front-end camera at fixed time;

carrying out moving soil scene detection on the obtained image;

and transmitting the analysis result to the back end at regular time.

2. The edge computing dynamic soil scene recognition algorithm based on deep learning of claim 1, wherein the periodically acquiring the front-end camera image comprises the following steps: the camera is directly butted with the cameras, video images of all the cameras are automatically obtained through timing polling, and the polling interval is adjustable within 2-10 minutes.

3. The algorithm for recognizing the edge calculation moving soil scene based on the deep learning as claimed in claim 1, wherein the moving soil scene detection on the taken image comprises the following steps: and detecting whether a target scene exists in the image, namely whether a moving soil scene exists.

4. The algorithm for recognizing the edge calculation moving soil scene based on the deep learning as claimed in claim 3, wherein the moving soil scene detection on the taken image comprises the following steps: selecting a frame of the target area, and selecting the position of the target object if the detection result is that the target object exists; and if the detection result is that no target object exists, uploading the original image.

5. The algorithm for recognizing the edge calculation dynamic soil scene based on the deep learning as claimed in claim 1, wherein the step of periodically transmitting the analysis result to the back end comprises the following steps: and uploading the identified result and the picture to a back-end application platform at a fixed time interval for display.

6. The edge computing dynamic soil scene recognition algorithm based on deep learning of claim 5, further comprising the anti-repeat alarm step of: and comparing the picture identified as the target object with the last picture identified as the target object, and if the picture is the same target object, not repeatedly alarming.

7. An edge calculation soil-shifting scene recognition intelligent terminal based on deep learning is characterized by comprising a shell and a control panel arranged in the shell, wherein a circuit is arranged on the control panel and comprises an ARM CPU, a FLASH, a memory and a TCP/IP interface; the ARM CPU is a core module of the intelligent terminal and is used for detecting and identifying a moving soil scene through a deep learning algorithm; the FLASH and the memory are used for storing and caching the pictures of the front-end camera and the analysis intermediate data and the analysis result data of the ARM CPU, and are respectively connected with the ARM CPU to realize the read-write processing of the data; and the TCP/IP interface is respectively connected with the ARMCPU and the memory and is used for acquiring pictures from the front-end camera and transmitting the analysis result to the back end.

8. The intelligent terminal for recognizing the moving earth scene based on the edge computing of the deep learning as claimed in claim 7, further comprising a 1.25V generating circuit for converting a 3.3V voltage into a 1.25V voltage through voltage stabilization for use by the circuit.

9. The intelligent edge computing soil-shifting scene recognition terminal based on deep learning of claim 7, further comprising a power line interface and a power indicator light, which are respectively connected to the ARM CPU.