Use sign-detection.ipynb

The repo utilizes the YOLOv5 architecture with modifications for our use. https://github.com/ultralytics/yolov5

There is a current issue regarding version incompatibility between the version of YOLOv5 we used and Torch. Please refer to ultralytics/yolov5#6948 for fixing this issue.

Examples

Above, frame displaying detection of stop sign

Above, frame displaying detection of speed limit including number on sign

Above, frame displaying false sign detection by YOLOv5, but caught by EasyOCR as urdbl (unreadable)

Usage

Setup

Enters the YOLOv5 directory and double checks the device used for detection

Inference

1. Parameters

The cell where things such as directory and other settings can be adjusted for the rest of inference. Requires setting up a videos folder if not using an existing one already.

2. Functions

Utility functions to be used later on

3. Detection

The call to the YOLOv5 architecture to perform inference on a set up videos as denoted in the paramters cell.

4. Process Labels

Post-processes the raw output from YOLOv5 to fill in gaps, remove outliers, and group together detections.

5. OCR

Runs OCR on groupings of detections to verify results from YOLOv5 and assign speed limit numbers.

6. Generate snippet highlights

Creates video snippets of every grouping of detection for ease of double checking

7. Generate results

Creates a more human readable csv for detections

8. Generate videos

Creates a full video with bounding boxes of detections.