CIO Insider

CIOInsider India Magazine


Visionbot: AI and Deep Learning Enabled Intelligent Video at Work

Amit Chakraborty,Co-Founder

Visionbot is in a continuous mode of developing new solutions around ML and AI. Headquartered in Bengaluru, the Visionbot platform leverages the scalability and high availability of cloud services to offer computer vision solutions. The company offers the power of computer Vision and AI to extract objective analyzable data from subjective visuals. Further, the company's founder will tell us how they ensure to provide an excellent user experience on this platform.

In conversation with Amit Chakraborty, Co-Founder, Visionbot

One of the tops AI trends to watch in 2019 is the growth of edge AI across applications. As an AI technology domain player, how has Visionbot grown and adapted to the changes?
Visionbot was started with an idea of getting into a large market sector primarily from the point of view of computer vision. Visionbot offers businesses to get the most out of their visual content thereby helping them to derive powerful insights and driving decision making. Designed as cloud-based Software as Services (SaaS) model, Visionbot lets users start using the system with minimal investment. A setup fee will be charged for customizations in the premium tier only. For the rest, it is only paid as per their use. Therefore, Visionbot is an adaptive

platform and learns on its own training. However, users can expedite training by providing preferences that make it more efficient.

Visionbot offers businesses to get the most out of their visual content thereby helping them to derive powerful insights and driving decision making

AI has already proved its huge potential and is here to stay. Elaborate on your distinguished AI services and which among them do you consider as your flagship offering that distinguishes Visionbot from its competitors?
One of the specialties of Visionbot is the ability to integrate camera to the cloud system on the go. That is one of the major usability characteristics that we have built into the system. We were providing similar services to the customer in the US, South-East Asia, and India. Likewise, we helped one of the clients in the UK with a parking area next to Gatwick airport. We created a system by which they do not need to send their engineers to the client's site to understand the problem they are facing. We used our media streaming server to connect them. We did all the processes from our Indian office and the entire training of license related recognition systems too. Second is the algorithm that we have developed from the deep neural network which we have customized for detecting objects and detecting moments in videos. We give importance to transient behavior, not merely the object or face detection. That is the use of RNN (recurring neural network) into it and it is still in infancy and we have done some significant work on that. The other thing is defining vision events; we are giving a definition of the particular event in the hands of the customer.

The security of data remains a challenge across the entire AI and machine learning spectrum. How do you deal with these challenges?
The data we have is coming from the camera or the video feed. We have two levels of security. First is the AAA mechanism which is the authentication, authorization and accounting system by which a user is authenticated and only authorized people are allowed to access. The second is secure transient data.

Please quote 1 or more implementation story that has or have left a mark in the company’s success.
In logistics, we are talking about identification and movement tracking of large containers at port terminal. The use-cases for a multinational logistics company which moves containers from ships on to trucks using cranes – hundreds are moved on a daily basis. Every container has a maximum tilt angle which if breached would attract demurrage charges. Currently it is the port operator who reports with substantive data that are prone to errors. Downstream insurance companies are also interested in data from such incidents. Visionbot application is used to accurately report such incidents by analyzing video feed from camera on a 24X7 basis. This eliminates human error due to fatigue and reporting failures.

What further innovation do you plan to include in your AI technology offerings?
The biggest problem in AI and computer vision is custom object detection because these require programming. The first part of our road map is to do custom adaptive training in which the system itself will adapt to the type of the object that a customer wants to detect. Also, automating the further movement detection system more intelligently and to increase the accuracy of the movement detection will be part of our future R&D.

Current Issue
Startups Leading the AI Race