https://github.com/vermavinay982/tex-tractor

Searching Words in Live Video using OCR & Computer Vision
https://github.com/vermavinay982/tex-tractor

computer-vision hacktoberfest hacktoberfest2022 ocr-recognition python video

Last synced: 9 days ago
JSON representation

Searching Words in Live Video using OCR & Computer Vision

• Host: GitHub
• URL: https://github.com/vermavinay982/tex-tractor
• Owner: vermavinay982
• Created: 2022-09-17T20:35:27.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2022-10-15T22:32:20.000Z (about 2 years ago)
• Last Synced: 2024-10-14T14:11:37.429Z (24 days ago)
• Topics: computer-vision, hacktoberfest, hacktoberfest2022, ocr-recognition, python, video
• Language: Python
• Homepage:
• Size: 12.5 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Changelog: CHANGELOG.md

Awesome Lists containing this project

README

        
# TexTractor - Video text searching utility

Goal: Trying to find the specific text in a video

Motivation: While watching tutorial, need to find a specific word on screen, but had to watch the whole video. So thought to make it for the world.

Benefits: faster inference on videos, better analytics, threading for performance, multi processing

![Image](assets/gui_look.gif)

![Image](assets/Demo.gif)

it will give the frame, or timestamp where the text was present

for all words - timestamps will be generated 

if the user searched few words - all those timestamps will be returned to user

- Find all the text in complete video, mark red the low confidence text, and black with confident text

- Allows you to find certain words, and get that slide created for you

- If you need the last page where the word was written, you can get that too by going through all pages

- Can summarize the whole video, and provide you the slides, super easy to read in emergency

- Works on your laptop, so no issue of privacy

## More Points

- search all at once

- search more 

- getting range of video where text can be present - for faster search

- location of detected text also can be mentioned 

- tracking of text can be done - for each frame - to reduce load on processing

- for each time stamp the text can be found out

- option to give frame skipping option to user is beneficial

- showing range where the word was present, some words are constantly present

- can highlight for each word by user where it was present

- close to it, could also be added - to make suggestion to user

- streamlit for deployment and params 

Mini Goals

- find best text detection model

- infer it on single frames

- generate frames using cv2

- save data and present it in csv form 

- use trees and other algorithms to merge the presense of word in all images