YOLO-Object-Counting-API

Real time Object Counting api. Implemented with the YOLO algorithm and with the SORT algorithm

The implementation is using model in same format as darkflow and darknet. Weight files, as well as cfg files can be found here. Darklow supports only YOLOv1 and YOLOv2. Support for YOLOv3 has not yet been implemented.

In order to achieve the best performance, you should have Cuda and tensorflow-gpu installed on Your device.

Demo

Count objects of a specified class crossing a virtual line

Graduate pedestrian

Counting pedestrains

Highway traffic counting

Count objects on a video

Count objects on a single frame

Set up

Dependencies

-tensorflow 1.0
-numpy
-opencv 3

Getting started

You can choose one of the following three ways to get started with darkflow.

1. Just build the Cython extensions in place. NOTE: If installing this way you will have to use ./flow in the cloned darkflow directory instead of flow as darkflow is not installed globally.

   python3 setup.py build_ext --inplace
   

2. Let pip install darkflow globally in dev mode (still globally accessible, but changes to the code immediately take effect)

   pip3 install -e .
   

3. Install with pip globally

   pip3 install .
   

Required files

The YOLO algoritym impementation used in this project requires 3 files. Configuration of network (.cfg), trained weights (.weights) and labels.txt.

YOLO implementation used in this project enables usage of YOLOv1 and YOLOv2, and its tiny versions. Support for YOLOv3 has not yet been implemented.

.cfg files

Configuration file determines a network architecture. Configurations can be found here. In example scripts we assume that the configuration is placed in cfg/ folder. Location of used .cfg file is specyfied in the options object used in the code.

The .cfg file can be downloaded using the following command:

wget https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov2.cfg -O cfg/yolov2.cfg

.weights files

The .weights files contain trained parameters of a network. In example scripts we assume the weights are placed in bin/ folder. Location of used .weights file is specyfied in the options object used in the code.

The .weights file can be downloaded using the following command:

wget https://pjreddie.com/media/files/yolov2.weights -O bin/yolov2.weights

labels.txt files

This file is list of classes detected by a YOLO netowork. It shoud contain as many classes as it is specyfied in a .cfg file.

Run counting

Once You have all dependencies instaled and all required files You can start counting objects. Object counting is carried out by an ObjectCuntingAPI object.

Examples of counting below

Count cars on crosing a virtual line

python3 count_cars_crosing_virtual_line.py

Count objects on video from Video Camera

python3 count_objects_from_camera.py

Count people on image

python3 count_people_on_image.py

Credits

The following open source projects were used in the implementation

Darkflow

The YOLO algorithm impementation - Darkflow

Python Traffic Counter

Object counting with YOLO and SORT. Similar project, but instead of using the darklow YOLO implementation, it uses the opencv YOLO implementation, so there is no GPU acceleration.

Deep Sort

Object tracking and counting - SORT

Modifications for Graduate Student Diploma Counting

This fork has been modified to count graduating students collecting their diplomas during commencement ceremonies. The key enhancement ensures that each student is counted only once when crossing the virtual line, preventing double-counting if they step backward.

Modified Files

object_counting_api.py - Main implementation file for one-way counting logic

Summary of Changes

1. One-Way Counting Implementation (Lines 142, 191-196)

Line 142 - Added object tracking list:

detect = []

• Stores IDs of objects that have already crossed the line
• Prevents duplicate counting when objects move backward

Lines 191-196 - Modified line-crossing detection logic:

if intersect(p0, p1, line[0], line[1]):
    if indexIDs[i] not in detect:
        counter += 1
        detect.append(indexIDs[i])
        print(indexIDs[i])
        print(detect)

• Line 191: Detects when a person's path crosses the virtual line
• Line 192: Checks if person's ID is NOT in the detect list (new check)
• Line 193: Increments counter only for first-time crossings
• Line 194: Records the person's ID to prevent future counting
• Lines 195-196: Debug output to track counted IDs

2. Additional Improvements

Lines 136-138, 207 - Disabled video output writing for improved performance:

# fps, height, width = get_output_fps_height_and_width(cap)
# fourcc = cv2.VideoWriter_fourcc(*'XVID')
# output_movie = cv2.VideoWriter(output_path, fourcc, fps, (width, height))

Lines 212-213 - Added pause functionality:

if cv2.waitKey(1) & 0xFF == ord(' '):
    cv2.waitKey(0)

• Press spacebar to pause video during processing
• Press 'q' to quit

Line 230 - Changed video source to camera:

cap = cv2.VideoCapture(0)  # Live camera feed instead of file

How It Works

The one-way counting algorithm ensures accurate diploma collection tracking:

1. YOLO detects and tracks each student with a unique ID
2. When a student crosses the red line, the system checks if their ID exists in the detect list
3. If it's their first crossing, increment the counter and add their ID to detect
4. If they step backward over the line, their ID is already in detect, so no increment occurs
5. Result: Each student is counted exactly once

Original Repository

Forked from: tugot17/YOLO-Object-Counting-API

Images and Videos sources

Highway surveillance video

Pedestrian surveillance video

Authors

• tugot17

License

This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details

That's all.