jyublee / surveillance_yolo

Surveillance YOLO

This Python project utilizes the YOLOv8 object detection model to detect people entering and leaving a camera's field of view. It logs the events and determines whether a person is on side A or side B of the frame.

Prerequisites

Ensure the following are installed on your system:

• Python 3.x
• PyTorch
• NumPy
• OpenCV
• Ultralytics YOLO

Installation

Clone the repository

git clone https://github.com/Jyublee/Surveillance_Yolo.git

Install the Prerequisites

pip install -r requirements.txt

Usage

• The program will start capturing video from the default camera (index 0) and perform person detection and logging during the specified time frame (current set to 9:00 AM to 1:00 PM).

• The program will display the video feed with bounding boxes around detected people and their assigned IDs.

• The program will log events in the logs.txt file whenever a person enters or leaves the frame, along with the timestamp and the side (A or B) they entered or left from.

• The program will start recording video when a person is detected and stop recording after a specified delay (recording_stop_delay) when no person is detected.

• The recorded videos will be saved in the recordings directory with a timestamp in the filename.

• A blinking red circle and "Recording" text will be displayed on the video feed when recording is in progress.

• Press 'esc' to stop the program.

Configuration

• capture_index: The index of the camera to capture video from (default is 0).
• log_delay: The delay in seconds between consecutive log entries for the same person (default is 2.0 seconds).
• presence_threshold: The number of frames a person needs to be present to be considered as entered (default is 10 frames).
• absence_threshold: The number of frames a person needs to be absent to be considered as left (default is 5 frames).
• iou_threshold: The Intersection over Union (IoU) threshold for Non-Maximum Suppression (NMS) (default is 0.5).

TODO

• Modify the Delay to give accurate logs
• Add Function to Save the footage as photos or videos when detection occurs
• Add potential Frontend and UI

License

This project is licensed under the MIT License.