Back to projects
Real-Time Object Recognition
Abstract.An assistive app that recognizes everyday objects from a live camera and announces them by audio.
- Role
- AI Developer Intern
- Period
- Jun to Sep 2022
- Context
- Digipodium, Inc. · Lucknow, India
- Team
- Solo
| ~1,000 | Self-collected training images |
| Real-time | Detection with audio feedback |
01From problem to approach
The problem
Visually impaired users need to know what is in front of them quickly, without asking for help.
The approach
Train a detector on a self-collected dataset and speak each result aloud.
02How it works
- Input Camera feed Capture live video frames from the camera.
- Step 02 Detect (TensorFlow) A trained model identifies objects in each frame.
- Step 03 Annotate (OpenCV) Draw bounding boxes and labels with confidence scores.
- Output Announce by audio Speak the detected objects aloud.
03What I built
- Collected and labelled a dataset of about 1,000 images and trained a TensorFlow detection model.
- Built the OpenCV real-time video pipeline and a Streamlit web UI with audio feedback.
- Designed the database and wrote the basic design documents (ER diagrams, functional and screen design).
04Technologies
- Python
- TensorFlow
- OpenCV
- Pandas
- Streamlit
- SQL Server