AI-enabled Number Plates for Commercial vehicles
Published on : Sunday 07-02-2021
Sureshbabu Chigurupalli presents a case study on AI-enabled Automatic Number Plate Recognition (ANPR) of commercial vehicles.
This is a summary of a pilot project conducted at the plant of Balasore Alloys Limited in Odisha, to validate 3rd party trucks and vehicle license plates using AI-enabled Automatic Number Plate Recognition (ANPR) for comparison against system entries and logging by other existing enterprise applications.
Background and context
It has been observed that SAP gate entries may be generated for certain 3rd party vehicles at the main plant gate even though the same ones do not turn up to the weighbridge (later identified through security footage). Moreover, valid material slips for them had also been created based on manual inspection and as a result the material receipt transaction was completed and no actual raw materials were delivered to the plant. To address the above issue, we wanted to test an ANPR based number plate matching solution. This was expected to circumvent human errors and wilful negligence and lead to reduced raw material procurement-related leakages. We, therefore, partnered with a service provider for the implementation of the pilot project. Based on the outcome's success, we may consider rolling out the solution at various 3rd party vehicle touchpoints such as weighbridges and entry/exits of its other manufacturing units.
What exactly is ANPR?
ANPR or Automatic Number Plate Recognition is a technology that uses optical character recognition to read vehicle registration plates. Initially, ANPR was used by police forces around the world for law enforcement purposes.
However, it soon entered the security market and industry surveillance. More and more providers saw the benefits of utilising this electronic toll collection solution, parking management, and smart parking. The development of this technology has helped industries enormously in monitoring outward and inward logistics in a significantly better and efficient manner.
Scope of the study
i. Custom development of ANPR algorithm, including testing, training, and on-site deployment for a live trial of 40 trucks for license plate recognition.
ii. Time frame –2 Weeks (3 days in the final week of activities)
iii. Completion criteria – Successful test of 40 consecutive trucks and corresponding file generation in the designated format in near real-time (less than 10 seconds).
ANPR architecture and setup details
One full HD, vandal-proof camera was deployed. The camera was positioned on the weighbridge facing the incoming trucks. The position was adjusted to read the truck number plate after it came to a stop on the weighbridge. The ANPR PoE switch was placed inside the weighbridge control room. It was an eight-channel PoE switch to allow connections between the camera and the Wi-Fi router. The switch also provided power to the camera. The Wi-Fi router provided internet connectivity to the ANPR laptop and connectivity to the FTP server. Put on all images and log files on the FTP server. Connect the router to the local switch at the weighbridge control room. The local network provided connectivity to the FTP server. The FTP server was used to store three types of files: a) One text file for each incoming truck with number plate and arrival timestamp; b) Watermarked image of the truck with number plate and arrival timestamp; and c) log file which captures during of each truck on the weighbridge.
VeraSight Edge software was installed on the ANPR laptop. The laptop was used for both training and live run. VersaSign AiE software was running on the laptop for number plate recognition. Custom code was developed and deployed to generate watermarked images, text files, and log files.
Figure # 1: ANPR architecture
Details of infrastructure
Approximately 50-70 trucks enter and exit the plant daily, with the numbers reaching 90-100 in peak times. Vehicles may be of varying sizes (e.g., 10 MT, 20MT, etc., in capacity). The trucks wait on the weighbridge for around 2-3 minutes on an average, depending on how long the operator takes to enter the weight and print out the material slip.
Key dependencies
a. Availability of sample text file format that SAP can pick up – file naming convention/data format
b. Availability of power, network and some method of transferring files for testing into a folder (FTP VPN, etc.), test folder availability for SAP system
c. Availability of pole for camera, power, network, shared folder availability with required write access to that location
d. Presence of suitable weather conditions for conducting activities on site
e. Availability of 40+ trucks for conducting live trials of ANPR setup.
Summary of ANPR results
1. We have successfully deployed the infrastructure. Day 1 was set-up, Day 2 was data collection, training and testing and Day 3 was the live run. Out of the 45 total trucks present on Day 3, 41 were detected and extracted (the other 4 number plates were not humanly readable).
2. For the detected number plates, we were able to achieve a 100% success rate. The accuracy of the results was dependent on line of sight, lighting, angles and uptime of equipment and devices.
3. For certain cases involving challenges/exceptions (e.g., handwritten number plates, obstructions, halt positions, etc.), their impact has been drawn.
Challenges identified during ANPR test:
(i) Hand-written Number Plates: Some number plates have handwritten characters and follow no standard spacing or character size.
Impact: Optical Character Recognition fails in such cases as the AI systems are tuned for digital printed character sets. Hence, the number plate is not being recognised.
Recommendation: Manual intervention may require to handle such cases.
(ii) Obstruction in front of Number Plate: Sometimes the truck has additional attachments in front of the number plate which are obstructing the number plate. The number plates are not even humanly visible.
Impact: The ANPR system cannot detect the number plate on the truck and hence no recognition occurs either.
Recommendation: Manual intervention may require in such cases.
(iii) Too wide a region for Truck Stop: The trucks did not have a specific area for stopping on the weighbridge. Sometimes the truck would come too close or stand too far out for the camera to pick up.
Impact: The algorithm was designed to pick up the number plate after the truck comes to a halt on the weighbridge. When the truck came too close, the number plate would go out of the camera vision. When the truck stopped too far, the camera did not pick up the number plate as it was waiting for the truck to come closer. The duration computation would also get triggered after the truck stops.
Recommendation: Putting a place marker for the trucks to stop.
(iv) People obstruction: Sometimes the drivers would walk or stand in front of the truck and obstruct the number plate view.
Impact: The system is taking a new reading right after the number plate becomes visible again, thereby producing two separate same truck readings on a single instance. This will result in the inaccurate computation of the duration the truck was waiting at the weighing bridge. It will also result in double counting of the truck.
Recommendation: The issue can be addressed in the following ways: a) The camera can be re-positioned to the other side of the weighbridge as the driver typically walks towards the control room; and b) Add redundancy by putting two cameras on two sides of the weighbridge.
Conclusion
The deployment of ANPR technology to read vehicle registration plates in the plant benefited from reducing human errors related to raw material procurement and helped reduce the waiting time using the ANPR technology.
Sureshbabu Chigurupalli is Plant Head – Balasore Alloys Limited, Odisha. Sureshbabu is leading and managing all plant operations with effective utilisation of all resources and implementing industry best practices such as TPM, Six Sigma, Lean Management and other Business Excellence initiatives that contribute to improve productivity and efficiency. He has exhibited leadership in closely collaborating with numerous Japanese Consultants for implementing TPM to enhance overall plant effectiveness.
A B.Tech in Instrumentation from Andhra University (1994), Sureshbabu is an enterprising leader and planner with a strong record of contributions in streamlining operations, invigorating businesses, heightening productivity, systems and procedures. He has achievement-driven professional experience in spearheading entire unit/plant operations to maintain continuity and match organisational goals through supervising Operations, Quality Control, Production Goals, Automation, Maintenance, Process Improvements, Safety Guidelines, Manpower Development, New Policy/Procedure Guidelines, Resource Allocation and Cost Optimisations.
