YoloDetection 🚀

Welcome to YoloDetection! This is a real-time object and digit detection and verification application using YOLO (You Only Look Once). The application can take input from a webcam or a video file and performs object and digit detection, verification, and optional visualization and transmission of the results. The repository also contains an algorithm for validating the sequential digits both individually and as a sequence versus a database of valid combinations.

Introduction 📚

This repository contains code for a live object and digit detection and verification application using YOLO (You Only Look Once). The application is designed to be flexible and easy to use, with several options for customization. You can enable or disable verification and visualization, adjust the confidence and IoU thresholds, and choose the class of objects to track. The application also includes options for logging and transmitting the results.

The application is built on top of the yolov8 repository by Ultralytics. YOLOv8 is a state-of-the-art object detection model that is fast and accurate, making it suitable for real-time applications. This repository extends the functionality of YOLOv8 by adding digit sequence detection and verification, and providing a user-friendly interface for running the application on a webcam or video file.

Installation 💻

To install and set up the repository, follow these steps:

1. Clone the repository:

git clone https://github.com/Tottowich/YoloTracking.git

2. Install the required dependencies:

pip install -r requirements.txt

3. (Optional) Download the pretrained models and place them in the TrainedModels folder.

Usage 🚀

To run the application, use the sequence_verification.py script. Here is an example command:

python sequence_verification.py --webcam "0" --track --track_digits --visualize --time 60 --disp_pred --object_frames 10 --class_to_track 1 --verbose

For a detailed explanation of the input arguments, see the Input arguments section below.

Input Arguments 🛠️

Object detection

Argument	Type	Default	Description	Example
--weights	str	ROOT/'./TrainedModels/object/object.onnx'	Path to the object detection model's weights.	--weights ./path/to/weights.onnx
--data	str	ROOT/"./TrainedModels/Object/data.yaml"	Path to the dataset configuration file.	--data ./path/to/data.yaml
--max_det	int	1000	Maximum number of detections per image.	--max_det 500
--conf_thres	float	0.6	Confidence threshold for object detection.	--conf_thres 0.5
--iou_thres	float	0.1	Intersection over Union (IoU) threshold for NMS.	--iou_thres 0.2
--track	action: BooleanOptionalAction		Enable tracking.	--track
--imgsz/--img/--img-size	int/list[int]	448	Inference size (height and width) for the input image.	--imgsz 512/--img-size 640 480

Digit detection

Argument	Type	Default	Description	Example
--track_digits	action: store_true		Enable digit tracking.	--track_digits
--digit_frames	int	3	Number of frames to track for digit certainty.	--digit_frames 5
--weights_digits	str	"./TrainedModels/digit/digit.onnx"	Path to the model for digit detection.	--weights_digits ./path/to/digit_model.onnx
--conf_digits	float	0.3	Confidence threshold for digit detection.	--conf_digits 0.5
--iou_digits	float	0.1	IoU threshold for digit detections.	--iou_digits 0.2
--ind_thresh	float	0.1	Individual threshold for digit sequences.	--ind_thresh 0.2
--seq_thresh	float	0.2	Sequence mean threshold for digit sequences.	--seq_thresh 0.3
--out_thresh	float	0.35	Output threshold for sequence mean history.	--out_thresh 0.4
--data_digit	str	"./TrainedModels/digit/data.yaml"	Path to the dataset configuration file for digit detection.	--data_digit ./path/to/digit_data.yaml
--imgsz_digit	int/list[int]	448	Inference size (height and width) for digit detection.	--imgsz_digit 512/--imgsz_digit 640 480
--combination_file	str	"./TrainedModels/data/combinations.txt"	Path to the combination file.

Inference settings

Argument	Type	Default	Description	Example
--object_frames	int	3	Number of frames to track for object certainty.	--object_frames 5
--tracker_thresh	float	0.6	Tracker threshold for object tracking.	--tracker_thresh 0.5
--class_to_track	int	1	Class index to track.	--class_to_track 2
--augment	action: store_true		Augmented inference.	--augment
--agnostic-nms	action: store_true		Class-agnostic NMS.	--agnostic-nms
--half	action: store_true		Use FP16 (half-precision) inference.	--half
--device	str	'cuda:0'	Which device to run inference on, e.g. mps, cpu, cuda.	--device cuda:0
--ckpt	str	None	Path to the pretrained model checkpoint.	--ckpt ./path/to/checkpoint.pth
--auto	action: store_true		Auto-size using the model.	--auto

Visualization

Argument	Type	Default	Description	Example
--visualize	action: BooleanOptionalAction		Enable visualization.	--visualize
--wait	action: BooleanOptionalAction	Help: Wait for keypress after each visualization	--wait
--prog_bar	action: BooleanOptionalAction		Enable progress bar.	--prog_bar
--hide_labels	action: store_true	False	Hide object labels in visualizations.	--hide_labels
--hide_conf	action: store_true	False	Hide object confidences in visualizations.	--hide_conf
--line_thickness	int	3	Thickness of bounding box lines for visualizations.	--line_thickness 2

Logging

Argument	Type	Default	Description	Example
--verbose	action: store_true		Print information during execution.	--verbose
--save_time_log	action: BooleanOptionalAction		Save time log.	--save_time_log
--save_csv	action: BooleanOptionalAction		Save results as CSV.	--save_csv
--log_time	action: BooleanOptionalAction		Log time during execution.	--log_time
--disp_pred	action: BooleanOptionalAction		Display predictions.	--disp_pred
--disp_time	action: BooleanOptionalAction		Display execution time.	--disp_time
--log_all	action: BooleanOptionalAction		Log all information.	--log_all

General

Argument	Type	Default	Description	Example
--ip	str	None	IP address to transmit to.	--ip 192.168.0.1
--port	int	None	Port number to transmit to.	--port 8080
--name_run	str	randomly generated names	Name of the run to save the results.	--name_run my_run
--transmit	action: BooleanOptionalAction		Transmit data.	--transmit
--webcam	str		Use webcam as input. Which webcam to use.	--webcam "1"
--classes	int/list[int]		Filter detections by class index.	--classes 0/--classes 0 2 3
--source	str	None	Path to the input source.	--source ./path/to/input.mp4

Example usage:

python sequence_verification.py --webcam "0" --track --track_digits --visualize --time 60 --disp_pred --object_frames 10 --class_to_track 1 --verbose

Contributing 🤝

Contributions are welcome! Please follow the coding style of the existing code and submit a pull request with your changes. If you are adding a new feature or fixing a bug, please include tests for your changes.