This paper provides a review of background subtraction techniques, categorizing them based on speed, memory requirements, and accuracy. Background subtraction is a widely used method for detecting moving objects in videos from static cameras. The core idea is to detect moving objects by comparing the current frame with a reference frame, often called the background image or model. The background image must represent the scene without moving objects and must be regularly updated to adapt to varying lighting and geometry conditions. The paper reviews several methods, including parametric and non-parametric background density estimates and spatial correlation approaches. These methods aim to estimate the background model from the temporal sequence of frames. The methods reviewed range from simple approaches that prioritize speed and low memory usage to more complex ones that focus on high accuracy. All methods aim for real-time performance, implying a lower bound on speed. The paper discusses various approaches, including the running Gaussian average, temporal median filter, mixture of Gaussians, kernel density estimation (KDE), sequential KDE approximation, cooccurrence of image variations, and eigenbackgrounds. Each method has its own trade-offs in terms of speed, memory, and accuracy. For example, the running Gaussian average is fast and memory-efficient but may not be accurate in all situations. The mixture of Gaussians and KDE offer higher accuracy but require more memory and computational resources. The cooccurrence of image variations and eigenbackgrounds address spatial correlation, providing good accuracy with reasonable computational and memory demands. The paper also analyzes the performance of these methods in terms of speed, memory requirements, and accuracy. It highlights that while some methods are fast and memory-efficient, they may sacrifice accuracy. Conversely, methods with higher accuracy often require more computational and memory resources. The paper concludes that the choice of method depends on the specific application and the trade-offs between speed, memory, and accuracy.This paper provides a review of background subtraction techniques, categorizing them based on speed, memory requirements, and accuracy. Background subtraction is a widely used method for detecting moving objects in videos from static cameras. The core idea is to detect moving objects by comparing the current frame with a reference frame, often called the background image or model. The background image must represent the scene without moving objects and must be regularly updated to adapt to varying lighting and geometry conditions. The paper reviews several methods, including parametric and non-parametric background density estimates and spatial correlation approaches. These methods aim to estimate the background model from the temporal sequence of frames. The methods reviewed range from simple approaches that prioritize speed and low memory usage to more complex ones that focus on high accuracy. All methods aim for real-time performance, implying a lower bound on speed. The paper discusses various approaches, including the running Gaussian average, temporal median filter, mixture of Gaussians, kernel density estimation (KDE), sequential KDE approximation, cooccurrence of image variations, and eigenbackgrounds. Each method has its own trade-offs in terms of speed, memory, and accuracy. For example, the running Gaussian average is fast and memory-efficient but may not be accurate in all situations. The mixture of Gaussians and KDE offer higher accuracy but require more memory and computational resources. The cooccurrence of image variations and eigenbackgrounds address spatial correlation, providing good accuracy with reasonable computational and memory demands. The paper also analyzes the performance of these methods in terms of speed, memory requirements, and accuracy. It highlights that while some methods are fast and memory-efficient, they may sacrifice accuracy. Conversely, methods with higher accuracy often require more computational and memory resources. The paper concludes that the choice of method depends on the specific application and the trade-offs between speed, memory, and accuracy.