Button Text


AI for tracking Shipping Containers

More than 130 Million containers are shipped every year – transporting goods from all corners of the earth. Tracking these containers through logistics facilities such as ports, warehouses, ICDs, CFS, etc. is still a major challenge.
Using AI Services For Shipping Containers

AI for tracking Shipping Containers

We believe that “AI will drive insane automation” and are excited by how it is shaping industries such as manufacturing, logistics and retail.

More than 130 Million containers are shipped every year – transporting goods from all corners of the earth. Tracking these containers through logistics facilities such as ports, warehouses, ICDs, CFS, etc. is still a major challenge. We at Silversparro, have recently worked with few logistics players - solving this problem of tracking containers while entering & exiting a facility.

Shipping Container OCR using AI Video Analytics
Gate for manually surveying entry and exit of Shipping Container trucks

Each container has a unique Container Number. This container number has to be noted down every-time the truck enters or exits the facility to keep a record. This ‘Surveying’ of container number, happens manually and inefficiently:

  • Manual effort: As there are multiple entry & exit gates which need to be manned 24*7 in various shifts, surveying becomes manpower heavy task. Not all gates are fully engaged at all times, so the productivity of each of these surveyors is very low.
  • Errors: Though there are multiple checks involved, manual data punching is prone to errors and single error can lead to huge costs. Every logistic company has a horror story of how a shipping container meant for Brisbane reached Qingdao.
  • Speed: Often there is a long queue of trucks, just waiting for the manual surveyor to punch down the container number - in spite of the manual surveyor doing his best. This wait time - costs fuel and time of the trucks - and when this loss is multiplied by million container trucks it becomes a significant overhead.
  • No proof/record: Manual survey also lacks a ‘record’ of which container actually passed through the gate and any error assignment becomes a huge investigative exercise.

There are solutions such as GPS tracking devices or RFID tags, which are often expensive and are not inter-operable across multiple logistic partners. We are solving it using AI-powered OCR - built on Sparrosense platform and are extremely excited by initial results.

How the solution works:

Five CCTV cameras are installed at the specified angles at the entry and exit gate. This ensures that all five faces of the container are captured when a truck carrying the container passes through.

The feed is picked from the NVRs by a GPU server which runs AI algorithms which does the following:

  • Identifying right frames: Each frame is essentially classified as ‘a frame with container number’ or ‘a frame without a container number. The trick here is to do it at the very low processing power and with 100% recall.
  • OCR of the Container Number: Once the right frame is identified the entire image is analyzed for identifying text and data is extracted. The algorithms are ‘smart’ to focus only on areas with the highest probability of having the container number.
  • Verifying Container Number: The model goes through all the text and picks the string most likely to be a container number. What helps is the fact that container numbers have a standard format.

Reaching Six Sigma Accuracy:

Many technology companies have so far not been very successful in achieving extremely high accuracies required to make the solution work. Smallest of error can lead to huge loss — but certain factors and our approach together are helping us reach near Six Sigma levels of accuracies.

  • Multiple Container Faces: The container number is written on all 5 faces and this gives us 5 sources of data.
  • Multiple Frames: As the source data is a video file that has the same number, we get up-to 10 shots for each container face i.e. 50 shots in total.
  • CheckSum: There is an added check built into the container number itself. The last digit can be extracted from applying a simple formula to all previous digit. Another check is against the facility’s ERP – where it is verified whether the extracted Container Number was indeed supposed to enter or leave the facility.
  • Secret Sauce: The secret sauce is that we do not use standard OCR solutions but an approach and model that we have built over the years. Instead of working on detecting each alphabet boxes our models OCR the entire ‘text box’ in one go reducing process time and significantly increasing accuracy as well.

This is not to say that there are no challenges. Rains and High Winds make our task tricky and non-standard container sizes (or container numbers) still cause plenty of trouble. But this only means that we need to get more creative in solving the edge cases.

Business value

This solution drives huge manpower savings, reduces errors and saves trucks time & fuel. The solution also helps create a record of all containers that ever passed through the facility in an easily searchable interface.

This can be extended to many more scenarios creatively for e.g. a simple mobile app that can scan a container yard and search through all the containers in real-time to assist staff.

What’s more

We are keen to extend the solution to adjacent use cases:

  • Rail OCR – OCR of Container and Wagon Numbers: Every ICD (Inland Container Depot) and CFS (Container Freight Station) need to maintain a log of which container is loaded into which wagon. This is still done manually. If cameras are placed on the entry and exit gate of the platform and record both the numbers, the current solution can be extended to automate this process.
Using AI Video Analytics for OCR of Wagon Numbers and Container Numbers.
Capturing Wagon Number's using AI Video Analytics
  • Monitoring loading during loading & unloading on Ships: The sequence of loading and unloading a container ship is a carefully planned orchestra, with even a small error throwing off entire process. If a camera can be placed suitably at the crane itself — it can be automatically ensured that the crane has picked up the right container eliminating the chances of errors.

We are excited to be working on a problem of global scale with relevance across logistics industry. We would be happy to partner with more clients to ensure that the solution further matures and benefits the industry at large.

If you're juts as excited as we are and constantly looking for ways to improve the results and increase the ROI, we have just the thing for you. AI Video Analytics is the next big thing and people have already started using it.

 You might do it as well. We are here to help you!

About the Author

Abhinav is the founder and CEO at Sparrosense. He graduated as an Engineer from IIT Delhi and has 10+ years of experience. He has worked as a Management Consultant with BCG and KPMG working with clients across Finance, Banking Healthcare, Government, Technology.