Vision Middleware

These middleware units are a part of a computer vision system that comprises a benchmark for video anonymization. The system aims to anonymize individuals sensitive objects within videos with an instance-level video transformation approach, as described in INSPIRE: Instance-level Privacy-preservingTransformation for Vehicular Camera Videos.

This repository's specific benchmark offers a comparison of Node.js and Golang in the context of Kafka and computer vision microservices. Futhermore, Node.js offers two libraries, kafka-js and node-rdkafka of which each is said to have its advantanges in write and read operations.

Design

Units

Other components of the system can be found in other repositories within the organization. The following middleware units are featured:

• Latent Generation
  • Generates random latent images for anonymization.
  • Stores latents in a key-value database with expiration.
• Pose-Latent Combination
  • Combines pose estimation tags and segmentation binary masks.
  • Stores intermediate results in caches for efficient processing.
• Segment-New Instance Combination
  • Similar to the Pose-Latent Combination Unit.
  • Handles additional data types for instance replacement.

Technology Stack

• Kafka
• Redis

Installation

Node.js:

./install.sh node

Go:

./install.sh go

Run

Node.js:

./run.sh node

Go:

./run.sh go