Case Study

CUSTOMER

The client who wants to localize and obfuscate (i.e. hides) sensitive information in images/videos in order to preserve the individuals’ anonymity. 

CHALLENGE

In order to use the recorded videos from operation rooms in hospitals for training machine learning models, it is requested to apply full body anonymization on given videos. BMW-Anonymization-Api is the only one open source which provides the free service to accomplish this task. However, it requires a lot of computing resource and slow down the anonymization process. 

SOLUTION

Use several pretrained models mainly for people segmentation in combination with Mask replacement and video processing tools to build an anonymization API which could reach the same goal as well as speed up the anonymization process. Gluoncv provides a huge resource for the implementation of state-of-the-art deep learning algorithms in computer vision. pretrained models can be used directly or further trained base on user expectation. The created API should be executed on linux with gpu support to optimize the computing process and therefore reduce the API running time significantly. Furthermore, video processing tools such as opencv and decord also give a hand in increase the performance of the anonymizer.  

Some pretrained models from Gluoncv and people segmentation: 

+ deeplab_resnet101_voc 

+ deeplab_resnet152_voc 

+ deeplab_resnet101_coco 

+ psp_resnet101_voc 

+ psp_resnet101_coco 

+ Unet 

Video processing tools: 

+ opencv 

+ decord 

RESULT

Built a good api which can anonymize 2 frames of video per second with good performance. The Models can be improved by training the pretrained models with customer created dataset. This provides customer the flexibility to control as well as improve the performance of the trained models. 

TECHNOLOGY

  • Main open source libraries: Gluoncv, Opencv, Decord 
  • Programming language: Python 
  • Operating system / GPU usage: Ubuntu 20.04.3 LTS / GeForce RTX 2070 Super with Max-Q Design 

Article Author: Khoi Vo , Project Owner:  Hieu Nguyen.  Datics Consulting all rights reserved.

Read more: Data Science Service

You’ll also like

Contact us!

Looking to explore the solution?

See the Datics Consulting Privacy Policy and Terms of Use for details on how we collect, use, and share information about you.