Skip to content

Latest commit

 

History

History
46 lines (36 loc) · 2.17 KB

README.md

File metadata and controls

46 lines (36 loc) · 2.17 KB

[UPDATE 2023: This repo is not maintained. The code might not work. No issues will be resolved]

Quick Overview

In this repository you can see 2 main programs: car_counter_yolov3_custom_classes.py and car_counter_yolov3_COCO_6_classes.py

The first one is a lighter version of the second. Basically, I`ve trained YOLOv3 to detect 5 classes:

  • sedan
  • minivan
  • SUV
  • hatchback
  • universal

But, to be honest, .weights file that I got in the end is pretty wack and works not that good on different videos. But it's still here.

How to run it

  • Download yolo-obj_final.weights file for YOLO here
  • Download any test-video with cars driving around and put it to videos/ folder (or use any of those that are already there)
  • Move .weights file to yolo/ folder
  • Go to the project's repository via command line
  • type python car_counter_yolov3_custom_classes.py -y yolo --input videos/THE_NAME_OF_YOUR_TEST_VIDEO --output output --skip-frames 5 and hit Enter

The proccessed video will be saved to the output/ folder

The second one uses pretrained .weights file from this site. So I didn't need to train YOLOv3 myself once again. This program can:

  • detect and track objects of all of 80 COCO classes
  • count objects of each of 6 classes:
    • car
    • truck
    • person
    • motorcycle
    • bicycle
    • bus
  • count the amount of all of those objects on each frame of the video
  • put the results into .json file

How to run it

  • Download YOLOv3-608.weights file for YOLO here

  • Download any test-video with cars driving around and put it to videos/ folder (or use any of those that are already there)

  • Move .weights file to yolo/ folder

  • Go to the project's repository via command line

  • type python car_counter_yolov3_COCO_6_classes.py -y yolo --input videos/THE_NAME_OF_YOUR_TEST_VIDEO --output output --skip-frames 5 and hit Enter

    You can change the skip-frames parameter (the higher it is, the faster the program works). But the accuracy will be lower

    The proccessed video and the .json file will be saved to the output/ folder