Fall detection systems are a critical area of application of the Human Activity Recognition (HAR) domain. With the advent of technology, assisted living will emerge as one of the strongest pillars of HAR. This paper investigates a vision-based fall detection system using ORB (Oriented FAST and Rotated BRIEF) features combined with various machine learning algorithms. The approach involves feature extraction using ORB, dimensionality reduction, and evaluation of different classifiers, including Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest, and Decision Tree. Among the models evaluated, Random Forest demonstrated superior performance, giving an accuracy of 0.8473 (± 0.1230 with cross-validation) using handcrafted features only. The study also reviews recent advancements in vision-based fall detection systems, highlighting the effectiveness of different approaches and techniques.