CN112232108B - AI-based intelligent gate system - Google Patents

Abstract

The invention discloses an AI-based intelligent gate system, which comprises a camera arranged on a gate, wherein the camera is connected with an AI identification system, and the AI identification system comprises an image analysis system, a truck CNN model system, a container CNN model system and a character identification system. Adopt AI identification technology and the pier gate system originally to combine, discernment only needs to trigger through the video stream, need not external sensor equipment such as infrared trigger, time schedule controller, RFID, ground induction coil, and AI discernment is not influenced by external factors, as long as the people can see clearly all can discern, and as long as the camera, the cost reduces, maintains also corresponding reduction of the degree of difficulty.

Description

AI-based intelligent gate system

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to an AI-based intelligent gate system.

Background

The container terminal is a same collective storage place before and after the container is opened and closed, is used for loading, unloading, transferring, keeping and handing over of heavy containers or empty containers, and plays an important role in container transportation. When the containers are stacked in ports and docks, the containers are generally stacked according to freight ship numbers so as to be conveniently loaded on the ships in a unified way. The vehicles (trucks) transporting the containers need to pass through the container terminal gates when entering and exiting the yard. The gate is an important component of the logistics system of the container terminal.

The traditional OCR image recognition technology is adopted in the current gate system, and a camera, an infrared trigger, a time schedule controller, an RFID card reader, a ground induction coil and a vehicle detector are required to be installed as technical supports.

Shelter from infrared trigger through the container and trigger the camera and take a picture, carry out the sectional drawing then with the word stock in the system to the specific area of photo and compare and discern the case number. Because the image recognition effect is influenced by weather, light, vehicle speed and the like, the recognition effect is poor and unstable. And because the number of sensors is large, the number of fault nodes is correspondingly increased, and the system can frequently have faults, higher requirements on system maintenance are required.

Therefore, the traditional OCR image recognition technology cannot recognize the attributes on the container and cannot meet the requirements of the intelligent gate.

Disclosure of Invention

Aiming at the defects in the prior art, the invention provides an AI-based intelligent gate system, wherein the AI system is directly butted with a network camera, each frame of image information in a video stream is intercepted by a VLC decoding technology, each pixel is analyzed for each frame of image information, image characteristic points are extracted, and then a truck collection model and a container model are utilized for identification and judgment; and identifies the container number of the container through character recognition optimization.

In order to solve the technical problem, the invention is solved by the following technical scheme: AI-based intelligent gate system, including setting up the camera on the gate, the camera is connected with AI identification system, AI identification system includes image analysis system, collection truck CNN model system, container CNN model system and word identification system to including following detection step:

s1, shooting an object to be passed through the gate in real time by a camera and forming a video stream;

s2, the image analysis system decodes the video stream of the camera and intercepts each frame of image information, and carries out binarization calculation on each frame of image information to generate a plurality of pixels, analyzes each pixel, extracts image characteristic points, and judges whether the collection truck CNN model system is a collection truck; if not, jumping to S1;

s3: the image analysis system decodes the video stream of the camera, intercepts each frame of image information, performs binarization calculation on each frame of image information to generate a plurality of pixels, analyzes each pixel, extracts image characteristic points, and jumps to S1 if the container CNN model system is a container or not;

s4: and recognizing the letter box number information on the container by a letter recognition system, and if the letter box number information on the container is correct, opening the gate.

And further optimizing, wherein the character recognition system comprises three sets of recognition engines, and the optimal box number is selected according to the results of the three sets of recognition engines.

Further optimized, the container CNN model comprises 12 layers of neural networks, and the truck CNN model comprises 12 layers of neural networks.

Further preferably, the image analysis system further comprises a buffer pool size and a network delay time which can be set and adjusted.

Further optimizing, the frame rate of the video stream is 60-100.

Further optimization, the logic of the optimal box number is as follows:

s1, setting whether to receive switch parameters participating in calculation for each engine;

s2, receiving the identification information sent by each engine, recording the receiving time and the engine source, and associating with the current operation, facilitating the subsequent analysis of the performance and the identification rate of each engine;

s3, firstly, identifying whether a switch of the information source engine participating in calculation is opened, if so, then, carrying out ISO verification on the identification information at a server, if the identification information passes the verification, then, executing a brake opening instruction according to the previous logic, and after the vehicle is released from the brake, finishing the operation, and putting the identification information of the three engines into the history; if the verification fails to pass to S4; if not, jumping to S4;

s4, adopting second box identification information, firstly identifying whether a switch of the information source engine participating in calculation is opened, if yes, then detecting whether the box identification number meets ISO verification, if yes, passing the verification, and if not, jumping to S5; if not, jumping to S5;

s5, adopting the third box identification information, firstly identifying whether the switch of the information source engine participating in calculation is opened, if so, then detecting whether the box identification number meets ISO check, if so, then passing the verification; if the verification does not meet the ISO verification, waiting for manual correction of the input box number by the user; and if not, waiting for the user to manually correct the input box number.

The invention has the beneficial effects that: 1. adopt AI identification technology and the pier gate system originally to combine, discernment only needs to trigger through the video stream, need not external sensor equipment such as infrared trigger, time schedule controller, RFID, ground induction coil, and AI discernment is not influenced by external factors, as long as the people can see clearly all can discern, and as long as the camera, the cost reduces, maintains also corresponding reduction of the degree of difficulty. 2. Whether the objects are containers and trucks is identified by establishing a truck-collecting CNN model and a container CNN model, and the most preferable letter box number on the container can be selected by a multi-engine letter identification system, so that the identification rate is further improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be discussed below, it is obvious that the technical solutions described in conjunction with the drawings are only some embodiments of the present invention, and for those skilled in the art, other embodiments and drawings can be obtained according to the embodiments shown in the drawings without creative efforts.

FIG. 1 is a flow chart illustrating the detection steps of the AI-based intelligent gateway system of the present invention.

Fig. 2 is a flow chart of the optimal bin number selection of the present invention.

Fig. 3 is a schematic diagram of the container CNN model of the present invention.

Fig. 4 is a schematic diagram of the truck CNN model of the present invention.

Detailed Description

The technical solutions of the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings, and it is to be understood 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 described herein without the need for inventive work, are within the scope of the present invention.

The embodiment of the invention provides an AI-based intelligent gate system, which comprises a camera arranged on a gate, wherein the camera is connected with an AI identification system, the AI identification system comprises an image analysis system, a truck-collecting CNN model system, a container CNN model system and a character identification system, and the AI identification system comprises the following detection steps:

s1, shooting an object to be passed through the gate in real time by a camera and forming a video stream;

s2, the image analysis system decodes the video stream of the camera and intercepts each frame of image information, and carries out binarization calculation on each frame of image information to generate a plurality of pixels, analyzes each pixel, extracts image characteristic points, and judges whether the collection truck CNN model system is a collection truck; if not, jumping to S1;

s3: the image analysis system decodes the video stream of the camera, intercepts each frame of image information, performs binarization calculation on each frame of image information to generate a plurality of pixels, analyzes each pixel, extracts image characteristic points, and jumps to S1 if the container CNN model system is a container or not;

s4: and intercepting the interested area on the container, and identifying the letter box number information by using a letter identification system, and if the letter box number information on the container is correct, opening the gate.

The character recognition system comprises three sets of recognition engines, and the optimal box number is selected according to the results of the three sets of recognition engines.

The container CNN model comprises 12 layers of neural networks, and the truck CNN model comprises 12 layers of neural networks.

Wherein, AI model training: container CNN model: as shown in fig. 3:

neural network with 12 layers

Inputting: 1920 × 1080 size picture, 3 channels

The first layer of convolution: the convolution kernel size 11 × 11 is 96, 48 per GPU.

First layer max-pooling: 2 × 2 cores.

Second layer convolution: 256 of the 5 × 5 convolution kernels, 128 per GPU.

Second layer max-pooling: 2 × 2 cores.

And a third layer of convolution: with the previous layer being fully connected, 384 convolution kernels of 3 x 3 are present. And the number of the GPU is 192.

And a fourth layer of convolution: the 3 × 3 convolution kernels are 384, 192 for each of the two GPUs. This layer was connected to the previous layer without a pooling layer.

And (3) convolution of a fifth layer: 256 convolution kernels of 3 × 3, 128 on the two GPUs.

Fifth layer max-pooling: 2 × 2 cores.

The first layer is fully connected: 4096 dimensions, the outputs of the fifth layer max-firing are concatenated into a one-dimensional vector as the input to the layer.

The second layer is fully connected: 4096 dimension

Softmax layer: the output is 1000, and each dimension of the output is the probability that the picture belongs to that category.

Wherein, the collection truck CNN model: as shown in fig. 4:

neural network with 12 layers

Inputting: 1920 × 1080 size picture, 3 channels

The first layer of convolution: the convolution kernel size 11 × 11 is 96, 48 per GPU.

First layer max-pooling: 2 × 2 cores.

Second layer convolution: 256 convolution kernels, 5 × 5, 128 per GPU.

Second layer max-pooling: 2 × 2 cores.

And a third layer of convolution: with the previous layer being fully connected, 384 convolution kernels of 3 x 3 are present. And the number of the GPU is 192.

And a fourth layer of convolution: the 3 × 3 convolution kernels are 384, 192 for each of the two GPUs. This layer was connected to the previous layer without a pooling layer.

And (3) convolution of a fifth layer: 256 convolution kernels of 3 × 3, 128 on the two GPUs.

Fifth layer max-pooling: 2 × 2 cores.

The first layer is fully connected: 4096 dimensions, the outputs of the fifth layer max-firing are concatenated into a one-dimensional vector as the input to the layer.

The second layer is fully connected: 4096 dimension

Softmax layer: the output is 1000, and each dimension of the output is the probability that the picture belongs to that category.

The image analysis system also includes a buffer pool size and a network delay time that are settable to adjust. The value is set in 100, the video decoding playing delay is smaller than 300MS, communication is carried out through a heartbeat technology and the camera, when the camera is off-line and inflation is completed, the system can be reconnected quickly, and the fact that the video stream is read quickly and stably is guaranteed.

The frame rate of the video stream is 60. I.e. 60 pictures per second.

The optimal box number logic is as follows:

s1, setting whether to receive switch parameters participating in calculation for each engine;

s2, receiving the identification information sent by each engine, recording the receiving time and the engine source, and associating with the current operation, facilitating the subsequent analysis of the performance and the identification rate of each engine;

s3, firstly, identifying whether a switch of the information source engine participating in calculation is opened, if so, then, carrying out ISO verification on the identification information at a server, if the identification information passes the verification, then, executing a brake opening instruction according to the previous logic, and after the vehicle is released from the brake, finishing the operation, and putting the identification information of the three engines into the history; if the verification fails, jumping to S4; if not, jumping to S4;

s4, adopting second box identification information, firstly identifying whether a switch of the information source engine participating in calculation is opened, if yes, then detecting whether the box identification number meets ISO verification, if yes, passing the verification, and if not, jumping to S5; if not, jumping to S5;

s5, adopting the third box identification information, firstly identifying whether the switch of the information source engine participating in calculation is opened, if so, then detecting whether the box identification number meets ISO check, if so, then passing the verification; if the verification does not meet the ISO verification, waiting for manual correction of the input box number by the user; and if not, waiting for the user to manually correct the input box number.

In summary, the AI identification system is directly connected to the network camera, the video stream of the real-time RTSP protocol of the camera is read, the VLC decoding technology is used for decoding locally, the size of the buffer pool is set, the network delay time is shortened (the value is set to 100), the video decoding playing delay is less than 300 milliseconds, the heartbeat technology is used for communicating with the camera, when the camera is offline and restarted, the system can be used for quickly reconnecting, and the video stream can be quickly and stably read.

The method comprises the steps of carrying out binarization operation on image information, analyzing each pixel, extracting image characteristic points, carrying out object identification by using a truck container CNN model, judging whether the truck container is a truck container or not, carrying out object identification by using a container CNN model, judging whether the truck container is a container or not, intercepting an interested area in the image, carrying out box number identification by using a character identification model, identifying the box number of the container, and screening out an optimal result according to each identification result by a plurality of sets of identification engines.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive. The scope of the invention is indicated by the appended claims, rather than the foregoing description, and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims (6)

1. AI-based intelligent gate system comprises a camera arranged on a gate, and is characterized in that: the camera is connected with an AI identification system, the AI identification system comprises an image analysis system, a truck collection CNN model system, a container CNN model system and a character identification system, and the method comprises the following detection steps:

s1, shooting an object to be passed through the gate in real time by a camera and forming a video stream;

s2, the image analysis system decodes the video stream of the camera and intercepts each frame of image information, and carries out binarization calculation on each frame of image information to generate a plurality of pixels, analyzes each pixel, extracts image characteristic points, and judges whether the collection truck CNN model system is a collection truck; if not, jumping to S1;

s3: the image analysis system decodes the video stream of the camera, intercepts each frame of image information, performs binarization calculation on each frame of image information to generate a plurality of pixels, analyzes each pixel, extracts image characteristic points, and jumps to S1 if the container CNN model system is a container or not;

s4: and recognizing the letter box number information on the container by a letter recognition system, and if the letter box number information on the container is correct, opening the gate.

2. The AI-based intelligent gate system of claim 1, wherein: the character recognition system comprises three sets of recognition engines, and the optimal box number is selected according to the results of the three sets of recognition engines.

3. The AI-based intelligent gate system of claim 1, wherein: the container CNN model comprises 12 layers of neural networks, and the truck CNN model comprises 12 layers of neural networks.

4. The AI-based intelligent gate system according to claim 1, wherein: the image analysis system also includes a buffer pool size and a network delay time that are settable to adjust.

5. The AI-based intelligent gate system of claim 1, wherein: the frame rate of the video stream is 60-100.

6. The AI-based intelligent gate system of claim 2, wherein: the optimal box number logic is as follows:

s1, setting whether to receive switch parameters participating in calculation for each engine;

s2, receiving the identification information sent by each engine, recording the receiving time and the engine source, and associating with the current operation, facilitating the subsequent analysis of the performance and the identification rate of each engine;

s3, firstly, identifying whether a switch of the information source engine participating in calculation is opened, if so, then, carrying out ISO verification on the identification information at a server, if the identification information passes the verification, then, executing a brake opening instruction according to the previous logic, and after the vehicle is released from the brake, finishing the operation, and putting the identification information of the three engines into the history; if the verification fails, jumping to S4; if not, jumping to S4;

s4, adopting second box identification information, firstly identifying whether a switch of the information source engine participating in calculation is opened, if yes, then detecting whether the box identification number meets ISO verification, if yes, passing the verification, and if not, jumping to S5; if not, jumping to S5;

s5, adopting the third box identification information, firstly identifying whether the switch of the information source engine participating in calculation is opened, if so, then detecting whether the box identification number meets ISO check, if so, then passing the verification; if the verification does not meet the ISO verification, waiting for manual correction of the input box number by the user; and if not, waiting for the user to manually correct the input box number.

Images