Shipping Reviews
HangYunPingLun
In order to promote the construction of smart ports, the digital port collaborative research team of Shanghai Maritime University integrates a new generation of information technologies such as machine vision, edge computing, object detection, OCR text recognition, and semantic segmentation, develops and completes a series of port machine vision products, provides port full-time domain intelligent identification solutions, and finally realizes the full coverage of port operations from artificial intelligence recognition and perception to cognitive decision-making.
Port machine vision products cover all aspects of port business, including business operation identification, port security identification, port machinery automation identification and other scenarios, and provide image recognition and real-time video stream collection and recognition functions.
Figure 1 Haida Port machine vision series products
The product provides services in a containerized manner, supporting cloud deployment, server deployment, edge deployment and other deployment methods, reducing dependence on the network. At the same time, the team has developed and completed the edge computing terminal product - Miro Smart Box, which supports the configuration of port machine vision product service modules on demand, provides customized smart box products, and customers are plug-and-play, automatic configuration services, convenient and fast, and provide corresponding decision support.
Figure 2 Platform architecture
The recognition algorithm modules developed by the team recognize more than 100 million containers annually, with an average recognition rate of more than 98% for all modules, and close to 100% for some modules (container numbers, etc.) (barely distinguishable by the human eye).
Figure 3 Digital solution
01
Identification of business operations
Business operation identification integrates three major scenarios: intelligent tally identification, intelligent yard identification and intelligent crossing recognition, to achieve full coverage of the port business operation identification field.
（1）
Container information (container number, etc.) identification and container group identification
function
Identify the container numbers circulating in different operation scenarios of the port, and provide the identification results in real time. At the same time, it supports the identification of multiple boxes and box groups. Adapt to multiple scenarios such as bridge cranes, field bridges, and crossings.
algorithm
Independently developed image correction, positioning, recognition and other related algorithms, after comprehensive evaluation and large-scale inspection of the production environment, the system recognition rate has reached more than 99%.
advantage
Adaptability to complex conditions. It supports the recognition of weather conditions such as rain and snow, as well as low light at night.
Complete result validation. The verification is carried out through the check digit in the 11-digit box number and the multiple consecutive recognition results of the video stream to ensure the accuracy of the recognition results.
Support fouling optimization identification and deformation optimization identification. For cases such as blurred, defaced, and deformed box numbers, the self-developed positioning algorithm improves the recognition accuracy in cases where the text is skewed, unclear, and incomplete.
Figure 4 The identification result of the bridge gondola box number is displayed
Figure 5 The container number identification result of a container group
（2）
Truck number recognition for internal and external sets
function
Identify the number of the inner truck and the outer truck (including yellow card, green card, etc.), the system is based on the video stream and trigger sampling two ways to identify, in the video stream mode, the system regularly identifies the camera sampling picture and returns the result to the client, and the recognition accuracy has reached more than 99%.
advantage
Nighttime image optimization is supported. The self-developed image enhancement algorithm automatically enhances the images that are difficult to recognize, achieving the same effect as the normal situation during the day, and greatly improving the accuracy of recognition at night.
Data augmentation is supported. In view of the fact that there may be license plate interference recognition such as dangerous goods next to the truck number, a picture generation algorithm is developed to accurately identify the vehicle number.
Figure 6 The identification result of the container card outside the crossing is displayed
（3）
Identification of dangerous goods signs
function
It solves the problem of difficult identification and low efficiency in the case of inconspicuous color difference of dangerous goods signs and boxes, compression and deformation of dangerous goods due to the limitation of camera angles, and can quickly and accurately identify dangerous goods containers, so as to avoid various potential risks in a timely and effective manner.
effect
The recall rate of comprehensive dangerous goods marks is more than 97%. The detection rate of horizontal shooting angle can reach 98%, and the false alarm rate is less than 4%. For images with serious deformation of shooting angles, the comprehensive recognition rate can reach 87%, and the false alarm rate is less than 7%.
apply
In the link of quay cranes and crossings, it is supported to identify whether ordinary containers and tank containers contain dangerous goods signs.
Figure 7 Tank box dangerous goods mark identification result display
Figure 8 The identification result of the dangerous goods mark of the ordinary box is displayed
（4）
Door identification
algorithm
A three-dimensional positioning algorithm for containers was developed to realize the pixel-level classification of coordinates on different surfaces of containers.
effect
Based on this algorithm, the door side can be effectively identified, and the accuracy of seal identification and container disability inspection can be greatly improved. The comprehensive recognition accuracy rate is more than 97%. The recognition accuracy of the door side of the horizontal shooting box can reach more than 99%, and the recognition accuracy of the oblique overhead shooting can reach more than 95%.
Fig.9. Schematic diagram of door identification
（5）
Seal identification
function
According to the photographs taken, based on the door recognition, the door seal is positioned and marked, and the type of seal of the container is accurately and real-time identified, which meets the real-time requirements of seal identification.
effect
The recognition accuracy rate is generally more than 95%.
apply
In the quay crane and crossing links, it supports the identification of whether the container lock button is locked and whether the container is hung with a seal.
Figure 10 The identification result of the seal is displayed
（6）
Container damage identification
function
The customized algorithm developed for identifying container damage will effectively filter the interference of non-top factors, support the identification of damage such as top holes, scratches, rust, deformation, and depressions, and unify the evaluation standard of container damage.
Container top damage identification: It supports the identification of container damage such as box top holes, scratches, deformations, rust, and oil stains.
Container side damage identification: Support the identification of container damage such as holes, scratches, and deformation on the side of the container.
effect
The effective detection rate of box damage can reach more than 95%, and the false alarm rate is less than 13%.
Figure 11 The damage identification result of the box top is displayed
Figure 12 The damage identification result on the box side is displayed
Port online machine vision API platform trial version
Access link: http://47.103.112.75:8501/user/login
Username: cv_test1
Password: 123456
02
Port Security Identification
Port security identification covers port facilities, ship operations, cargo security and personnel safety, and monitors port operations through machine vision technology