CMSC 436 Research Paper Extra Credit

Using Haar Feature-Based Cascade Classifiers for Face Detection on a Raspberry Pi for IoT Applications

• Link to Draft on Google Drive
• Link to Positive Image Set
• Link to Negative Image Set
• Link to OpenCV3.4
• Link to OpenCV4

Project Setup

Prequisites: python-opencv installed via pip, opencv installed via building from source (link to macos).

1. Clone the repo.
2. Download the positive image set. Then rename labels to yolo_labels, label2 to labels, images to pos_images.
3. Download the negative image set. Rename the image directory to neg_images and remove all samples containing human faces with rm Human*. Put the neg_images directory in the archive directory. This is your complete dataset.

Usage

Run the face detection model with the following command:

> python face_detection.py [-r] [-t]

The -r flag denotes that the script is being run on a Raspberry Pi, and the script will use the Picamera2 module accordingly. The -t flag denotes that we are logging the time it takes to compute a frame.

Run the model evaluation script with the following command:

> python test_model.py --config <path_to_config>

Reference the config files in config/ to understand how to write a config file for this script.

Run the profiling script with the following command:

python test_utilization.py face_detection.py