Architecture for ensuring Privacy in Real Time Video based Computer vision applications
A privacy policy is a legal document that outlines how a software product or service collects, uses, stores, and shares user data. It serves as a transparent communication tool between the software provider and its users regarding the handling of their personal information. Most of the time users agree to privacy policy without reading or properly understanding its content. As a software provider it is important to collect required information with correct format but most of the users tend to skip these steps. How do we solve this problem ?
Sometimes the business logic of a software product does not depend on the raw user data. The architecture of the software can be changed to convert users raw data into other data formats that can be used and stored for future use.
In Real Time Video based Computer vision applications the videos can be converted into numerical data and data can be stored in a database. In the following python code snippet the landmarks detection of a hand movement is explained.
This python code snippet can detect the real time hand movements and it will display the skeleton of the hand as follows.
The print statement in line no 19 will print out the real time numerical data in the console and it is listed below.
This data can be converted into a single json or if it is too large it can be converted into a string and it can be sent to the database by using the backend APIs of the application.
This architecture can be used in almost every field that uses real time computer vision like Autonomous vehicles, Manufacturing and Quality Control, Retail and E-commerce, Agriculture, Security and Surveillance, Healthcare, Robotics and Sports Analytics. With the collected data, it becomes possible to generate skeleton videos and even perform comparisons between two or more user videos, making this process incredibly adaptable and powerful.
By following this architecture, software providers can guarantee that the software will not collect the user videos and it will only generate a numerical object that contains the landmarks of the user video and since this process can be done in mobile or web applications the videos will not be saved in any other places like cloud storage.
Written by Akash Ekanayaka
Keep reading
View all technical blogsMore technical blogs
Next in this series — testing, firmware, wireless, and on-device AI.
Up nextCentralized Beta Testing
How we rolled out centralized beta testing for an ML-driven physiotherapy app across a growing device mix.
Read article
Automated UI Testing for SenzAgro
How automated UI testing keeps SenzAgro’s AIoT application reliable as features and devices expand.
Read articleReal-time Object Detection on the Edge
Running real-time object detection on edge processors with TensorFlow—without sending every frame to the cloud.
Read article
Test Automation for IoT Devices
An open-source approach to automating IoT device tests so quality holds from smart homes to industrial systems.
Read article
Building Animated Skeleton Video
A step-by-step look at building animated skeleton video on mobile while keeping vision data private.
Read article