Object Counter

Simple program object for detection and counting. There are two working modes, count objects (in the scene) or gate crossing count.

To detect the objects the program uses a Yolo network. While the demo version comes with a pretrained model that can detect persons, bicycles or cars, object_count.py is suited to detect custom objects. In that case you must provide a trained model.

Counting mode

Gate Crossing mode

Setup

Install or build OpenCV (building is required for GPU support - recomended).
Clone this repo git clone https://github.com/LorBordin/object_counter.
Change directory cd object_counter.
Install requirements pip install -r requirements.txt.

To use the demo

Download the Yolo-COCO weights and move them in models/yolo-coco.
run python demo.py -m counter -v PATH_TO_INPUT_VIDEO.

Optional arguments:

-m MODE: Options: counter or gate_crossing
-v VIDEO_PATH: Path to input video
-y YOUTUBE_URL: YouTube video URL
-o VIDEO_PATH: Path to output video
-s 0_or_1: If 0 doesn't show the live output - deafult 1
-l GATE_COORDS: Gate coords - format: "[Xt,Yt] [Xb,Yb]"
-c CLASS: Object class. Options: person, bicycle, car

Notice: if you use the demo in the gate_crossing mode without the flag -l, you have to choose the gate coordinates by clicking two points on the video (a window will pop up).

To use your custom weights

To use your custo Yolo model you must provide the path to the Yolo .cfg, .weights and .names files. Run

python object_counter.py -m counter -v PATH_TO_INPUT_VIDEO -cfg PATH_TO_CFG \
-weights PATH_TO_WEIGHTS -names PATH_TO_NAMES