The objective of research into navigation assistance is to facilitate autonomous living for individuals who are visually impaired. Although numerous navigation aids employ contemporary technologies and methodologies, they do have certain drawbacks, including portability, object detection, convenience, and the need for extensive training. When developing a navigation aide for individuals with visual impairments, portability and user-friendliness without the need for extensive training are additional crucial factors. Certain navigation systems might fail to furnish precise details regarding the nature of hurdles that can be identified, a critical component for making well-informed decisions while in real-time travel. This research proposes the EfficientNet with entirely contorted Convolutional Neural Networks, for the development of a navigation assistance system in order to circumvent all of these challenges. The system's EfficientNet is a portable object detection model that permits the execution of multiple operations, including uniform expansion, resolution, depth, and breadth. The completely convolutional CNN, a pre-trained model, finds extensive application in object detection and computer vision tasks. The evaluation of the proposed system is conducted using the metrics of Distance, Mean Average Precision (mAP), and Accuracy.