Year: 2024
Type: conference-paper
Abstract: This research introduces a machine-learning human body classification model for detecting defensive and aggressive human stances in real time using the Media Pipe holistic model. This method extracts relevant landmarks from video streams and detects aggressive and defensive poses. The process involves a handcrafted dataset that captures landmarks and the corresponding coordinates into a .csv file and then trains and evaluates four different classification models performing a test-train split and cross-validation. The best-performing model is then selected and used for testing. The methodology was replicated on a Raspberry Pi 4 and on a Desktop computer, with results indicating computational efficiency across both platforms, highlighting its adaptability to low-computational power environments. This research introduces an algorithm for detecting defensive and aggressive human behaviors, pinpointing the importance of multi-modal gesture model application in real-world scenarios.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4410825482