"FOCUS" is a project that allows for the analysis of human emotion in images via a bot. It utilizes AI - a software that learns from patterns and features in the data using the combination of algorithms and data - to analyze the facial expressions and surroundings in said video/image. Its purpose is to detect potentionally dangerous situations.
The bot is tagged in these images on Twitter and classifies the emotion picked up. This bot controls a Twitter account using its Twitter API keys and tokens from the Twitter developer site. In order to allow the bot to detect emotion, a pre-trained emotion recognition neural network was used. The network selects the most likely emotion per face and replies to the user. This neural network was run on a virtual machine (VM), that was created in Google Cloud to run a Docker image and code.
The bot is running continuously and constantly listens for mentions of its handle.
The keys and tokens were placed in an .env file to secure them and pulled into the Python code. Data sets (ie. security footage and screen frame by our images to detect danger) and sources were put into a bucket. The dataset was used to test the bot.
The bot was coded in a Jupyter notebook using AI platform. The bot code downloads an image from twitter using its URL and replies even if the tweet contains no image.
The end goal is to use this project to help move towards a more equal and just society. Events such as the George Floyd incident and other inhumane instances of police brutality sparked the need for this project as the projects aims to detect emotion and possibly intentions of these recorded instances to help analyze each unfortunate situation and ultimately create a change in society.
- Create an env file for twitter credentials with the format:
CONS_KEY=your_key
CONS_SECRET=your_secret
ACCESS_TOKEN=your_access_token
ACCESS_SECRET=your_access_secret
-
Download the DemoDir files (model files) https://drive.google.com/file/d/0BydFau0VP3XSYk9ZVnVNd0ZvVk0/view
-
Modify the docker-compose mount to mount the demodir
-
Start the server from within the docker directory and run with:
docker-compose up
- Tag the bot in a tweet
- Upload an image to the tweet
The bot will respond to the tweet classifying an emotion per face.
- Find image above tagged tweet (easy)
- Analyze a video (intermediate)
- Swap out API keys (easy)
- Permanently house on a server (easy)
Gil Levi and Tal Hassner. Emotion Recognition in the Wild via Convolutional Neural Networks and Mapped Binary Patterns. Proc. ACM International Conference on Multimodal Interaction (ICMI), Seattle, 2015.
https://talhassner.github.io/home/publication/2015_ICMI
Adarsh Menon
https://towardsdatascience.com/face-detection-in-2-minutes-using-opencv-python-90f89d7c0f81
https://github.com/adarsh1021/facedetection
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent