cv_object_detection_webinarthincai's Introduction

Object Detection in images and videos using the YOLO Object Detector

Overview

Applies the YOLO Object Detector to detect and classift ~80 object types in a given image or video. The algorithm was first described by Redmon et al.,in their paper You Only Look Once: Unified, Real-time Object Detection. In this, we used YOLO3, which has some improvements over their original model (YOLOv3: An Incremental Improvement). All their code can be found at https://pjreddie.com/yolo/.

The file size of the model weights is greater than github's size limit, so it's downloadable at https://pjreddie.com/yolo/ or instead just run download_model_weights.sh.

Running the YOLO Object Detector on one image

$ python --image images/baggage_claim.jpg
- Required arguments:
  - --image (path to input image)
- Option arguments:
  - --threshold (threshold when applying non-maxima suppression, type=float)
  - --confidence (minimum probability to filter weak detections, type=float)

Running the YOLO Object Detector on a video

$ python yolo_video.py --input videos/airport.mp4 --output output/airport_output.avi
- Required arguments:
  - --input (path to input video)
  - --output(path to output video)
- Optional arguments:
  - --threshold (threshold when applying non-maxima suppression, type=float)
  - --confidence (minimum probability to filter weak detections, type=float)

Running the YOLO Object Detector on a live webcam video

$ python yolo_real_time.py
- Option argument:
  - --output (path to output video if you want to save)
  - --threshold (threshold when applying non-maxima suppression, type=float)
  - --confidence (minimum probability to filter weak detections, type=float)

Credit: