In an era of rapid urbanization and automation, pedestrian safety has become a central concern for surveillance and smart city infrastructure. This paper presents a lightweight yet efficient system for pedestrian detection using YOLOv4-tiny, optimized for real-time video analysis. The system integrates OpenCV with the cv2.dnn module and Python-based inference logic to detect and annotate pedestrian locations in video streams. With the use of confidence filtering and non-maximum suppression, the solution demonstrates high accuracy and frame-wise efficiency even in constrained environments. The results suggest that YOLOv4-tiny provides an effective balance between speed and precision for edge deployment.