Existing video surveillance systems in public parks face several challenges in ensuring child safety. Traditional systems often rely on basic motion detection algorithms, which can result in false alarms or fail to detect subtle threats. These systems may also lack the ability to scale effectively, leading to limitations in monitoring large park areas in real time. Furthermore, privacy concerns and the need for data security pose additional challenges in implementing effective surveillance systems. The proposed Real-Time Video Surveillance-based System for Enhancing Child Safety in Public Parks using Jetson TX2 and the SafetyNet-102 Model addresses these challenges by leveraging advanced technologies. The use of the Jetson TX2 platform allows for high-performance computing on the edge, enabling real-time processing of video data. The SafetyNet-102 Model, with its deep learning capabilities with 102 layers, enhances the system's ability to detect and respond to potential threats. The 102 layers in the SafetyNet-102 Model are formed using ADAM optimizer and SiLU activation functions, enabling efficient learning and inference. The model performs both object and event detection such as fainting incidents, garbage accumulation, restricted area breaches, weapon presence, and overcrowding, and classify them accordingly, providing a comprehensive understanding of the objects present in the input video. The system also includes audio-visual alerts to notify authorities of potential threats in real-time. The proposed SafetyNet-102 Model is evaluated against a customized COCO dataset, comparing it with various algorithms such as YOLOv5, YOLOv7, YOLOv8, ObjectViViT, and Mask-RCNN. The results show that the proposed model outperforms all these state-of-the-art techniques, achieving an accuracy of 98.8% and a loss of almost 0.2%. This demonstrates the effectiveness of the proposed system in enhancing child safety in public parks.