Numerous modern technologies fundamentally depend on precise indoor object localization. Conventional approaches for the signal strength calculation and the network architecture design must rely, in most cases, on the 3D models or floor maps of the sites, which can be computationally expensive and time-intensive to generate. This paper presents a novel machine learning-based approach for indoor object localization that simplifies these processes while reducing resource requirements. The approach makes use of visual data and efficient algorithms that allow for the real-time recognition and localization of indoor objects and generate lightweight models that can be used for various applications. One of the applications of this strategy is the prediction of signal strength that can increase the accuracy of the placement of the transmitter and the receiver through real-time identification and localization of indoor objects. The preliminary results show the potential of this method to streamline network design, optimize resource allocation, and reduce environmental impact, aligning with sustainability goals. Since both the 3D models and the floor plans are not extensively used, the technique also promotes sustainable practices in the planning and deployment of the networks. This work provides a basis for the development of wireless systems that have a high capacity and are energy efficient, while paving the way for the increased use of machine learning for indoor localization and network optimization purposes.