Post 4

Reimagined Reality

Temporary Name)

Version 01-

In this version, I want to gather surveillance camera footage, isolate individuals from their original backgrounds, and blend them into the same settings, forming a partially altered reality. This explores the intersection of privacy and the increasingly ambiguous boundary between reality and technology in contemporary society.

Methodology

First, access to public surveillance camera footage. Second, use machine learning model is employed to identify and remove the background from the acquired footage.

1. Some footage that I can access

2. Technical Test

ml5 with camera

Tensorflow with the image I host on my website.

Visual Reference:

I recorded this video before. I like the unclear sense of space created by the blurred characters and background.

Some images that I feel will be great as references with different senses of space and movement.

Some problems & progress :

1. The browser's cross-domain resource restrictions on different websites prevent me from directly importing the footage into the website for processing.

2. I found that I can load the material into my webpage by adding some tags, but machine learning part will determine it as a contaminated canvas (it cannot be edited.)

3. I'm thinking about the option of not using machine learning. Therefore, I tried to compare the pixel changes to draw the changed pixels in the picture and got this result. The reason for choosing this method is because the monitor is a fixed-angle camera, so it may be a good choice to leave moving objects.

Version 02-

In this version I wanted to try out different AI generation tools. Use AI that describes the picture to describe the content of what the ai saw in the surveillance camera, and then hand it over to the drawing AI to regenerate the picture from the content. And connect them together to form an AI surveillance camera based on the “realistic” description.