7th ISF

COMP027 - Predicting Blood Drop Height and Volume Using Physics Equations, VGG-19, and XGBoost

Blood Pattern Analysis is a technique in forensic science which focuses on leftover bloodstains from the crime to recreate the event. However, the fluctuation in air resistance and drop deformity causes the calculations to deviate from the exact values. Therefore, machine learning models were constructed to overcome this limitation of calculations. A series of experiments was conducted by dropping porcine blood on paper across nine distinct heights: 20, 40, 60, 80, 100, 120, 140, 160 and 180 cm with four different drop volume: 13, 16, 25 and 30 µL resulting in 36 classes. A simple simulation of free-fall spherical object was also created to convert any drop height into impact velocity. Regarding both the empirical data and simulation, the correlation between spreading factor and modified Reynold number along with the number of spines and modified Weber number were expressed as equations which can be used to determine drop height and drop volume. Concurrently, the same dataset as used in physics calculations was used to train machine learning models that implement VGG-19 and XGBoost. For VGG-19, the inputs are images of bloodstains while for XGBoost, the inputs are stain area, stain perimeter, and the number of spines. As a result, the accuracy for physics equations, VGG-19, and XGBoost resulted in 0.26, 0.56 and 0.49, respectively.

Show More