Individuals with visual impairments encounter considerable difficulties when navigating outdoor environments, where limited accessible information can lead to accidents or discomfort. To address this issue, an intelligent system is proposed, utilizing deep learning and computer vision to assist users with visual impairments in safely navigating park environments. The system adopts a structured machine learning approach, comprising phases of data preprocessing, model training, and evaluation. The dataset used in this study, though small and self-collected, includes images of common park elements such as benches, trees, and vehicles. This limited dataset presents an opportunity to explore transfer learning and fine-tuning techniques to improve model performance. Several convolutional neural network (CNN) models are evaluated using classification metrics like Hamming Loss, which is especially important for multi-label classification tasks. Results show promising accuracy and effective performance in real-world applications. Future work will aim to expand the dataset, include additional object categories, and enhance system adaptability by integrating advanced sensory inputs and incorporating user feedback.