Objective to use Cloud Concepts and Machine Learning in their Offerings

The objective of this project was to develop an API which returns whether a person in video is matched with ID proof he or she has provided with some additional information such as age, gender, emotions, eyewear and other facial landmarks in a few minutes. This API was the first step of customer verification which clients can use to integrate with their own application.

Identifying & validating customers in real time: The customers frequently upload their identity/proof documents through web and mobile service platforms. They needed a solution to verify using available recognition softwares with minimal customization to make it developer friendly for future requests as well.

Performance & Scalability: Their previous solution was deployed on a dedicated server which took 2-3 minutes to run 500 simulations. Increased focus on platform/ Infrastructure during certain business quarters and additional workload for in-house teams to manage the SLA for their stack was a continuous challenge.

The main objective of this project is to apply knowledge of Cloud Concepts and Machine Learning in development of real life applications which make the life of users a lot easier and faster.

Searce’s Solution using Amazon‘s compute instance Elastic Compute Cloud (EC2)

The work dealt with face detection, head pose estimation, smile detection, blink detection, face comparison and hosting API on Amazon‘s compute instance Elastic Compute Cloud (EC2). Various head pose and blink detection is used for checking liveness of users. Hence input video contains recording of users with 7 different head poses within the specified input time range given by the system.

Primary goal of this project was to develop an API which will be available for Kristal customers on their web & mobile platforms used to identify whether the person in the video is the same as the person in Valid Identity Proof document with very good accuracy and faster.

The Business Impact of using AWS S3 and AWS Rekognition in Kristal.AI’s Operations

• Continuous learning: It is built upon deep learning based algorithms hence provides high accuracy. Moreover, service is providing continuous trained models to address any shortfalls / mitigate the prior accuracy/ image recognition
• Simple integration: API reduces lots of complexity in building actual learning algorithms. Instead one can build image and video analysis into any web or mobile application easily and quickly.
• Scalable: The current application which uses millions of images and tons of videos. And the platform is configured with autoscaling groups and policies to address any overload activities
• Easy Integration: Rekognition can be integrated with other AWS services such as s3 or lambda. Hence, clients can take advantage of other services such as S3 and lambda are both scalable in terms of application usage demand. Also to integrate with other applications/ 3rd party services with an API to address additional business offerings in line with the current business logics
• Low cost: The platform provides pay as you use policy. Hence, it comes with the lowest price. Client will have to pay for number of images that are analyzed, minutes of video to process and total metadata stored