This paper introduces a privacy-preserving system for detecting presence using Wi-Fi Received Signal Strength Indicator (RSSI) variations, capable of distinguishing humans from various objects. Multiple low-cost Wi-Fi nodes (arranged in a grid) and 2.4 GHz RF source capture signal disturbances, which are interpolated into heatmaps and combined with raw and filtered RSSI features for analysis via a Random Forest machine learning model. To address the challenge of labeling wireless data, cross-modal supervision integrates synchronized camera frames during training to guide the learning process, after which the system operates solely on RSSI inputs. The approach achieves 94% classification accuracy in evaluated scenarios, with potential to differentiate humans and objects through expanded node deployment. Unlike camera-dependent methods, it detects presence through obstacles without privacy concerns. Built with $50 hardware, it offers an affordable alternative for smart environment monitoring.