The growing need for automated surveillance systems, fueled by heightened public safety concerns, calls for reliable real-time tracking and violence detection features. This study presents a comparative evaluation of three widely used tracking algorithms: Kernelized Correlation Filters (KCF), Channel and Spatial Reliability Tracker (CSRT), and Deep Simple Online and Realtime Tracking (DeepSORT). While KCF offers high speed, its accuracy declines under challenging conditions. CSRT provides enhanced robustness but sacrifices processing speed. DeepSORT, with its deep learning-based appearance descriptors, excels in maintaining object identities across frames, especially in dense environments. Recognizing DeepSORT as the most effective option, this work proposes an integrated framework combining DeepSORT with OpenPose, an advanced pose estimation technique, to improve violence detection in surveillance scenarios. The system incorporates YOLO for object detection, DeepSORT for tracking, and OpenPose for body movement analysis, with an LSTM-based classifier to identify violent behaviors in real time. The proposed solution achieves exceptional performance, offering a reliable and efficient approach for automated violence detection in real-world applications.