Video surveillance is widely used for automated security monitoring in public and private environments. Object detection enables intelligent analysis of surveillance footage by identifying entities such as pedestrians and vehicles. This paper presents a structured survey of object detection techniques used in video surveillance, covering both traditional motion-based methods and contemporary deep learning–based approaches. Conventional techniques, including background subtraction and optical flow, are discussed alongside deep learning frameworks such as Convolutional Neural Networks (CNNs), Region-based CNNs (R-CNN family), Single Shot Multibox Detector (SSD), and You Only Look Once (YOLO) [6, 7, 9, 12]. The study analyzes their detection accuracy, computational efficiency, and robustness under practical surveillance challenges such as occlusion, illumination variation, and real-time processing constraints. In addition, recent enhancements involving hybrid detection models, multi-object tracking, data augmentation, and lightweight architectures are examined [1, 5, 13, 14, 16]. The study also highlights open challenges and future research directions for improving robustness and real-time deployment.