To address the issues of insufficient positioning accuracy, high cost, and the requirement for dedicated onboard positioning modules in existing indoor UAV positioning schemes, this paper proposes a high-precision three-dimensional (3D) UAV positioning system based on WiFi channel state information (WiFi-CSI) sniffing. Specifically, the system deploys a number of low-cost ESP32-S3 sensors to construct a UAV sensing network, which directly sniffs the WiFi signals transmitted by the UAV and parses the CSI signals from them. Combined with the altitude information acquired by time of flight (ToF) sensors, this system achieves a lightweight and high-precision positioning solution that eliminates the need for dedicated positioning receiving modules on the UAV. Considering CSI characteristics at different altitude layers, we innovatively design a hierarchical training strategy and a sparse neural regression network (SIRD). Fusing the SE attention mechanism, Inception multi-scale convolution, and residual connections, the network establishes the mapping between CSI and 2D coordinates. 3D positioning is finally realized by fusing predicted 2D coordinates with ToF altitude. Experimental results verify the effectiveness of the proposed system are the mean localization error (MLE) is 0.107 9 m, the localization mean squared error (LMSE) is 0.100 3 m, and the coefficient of determination R2 is 0.956 2. Compared with existing indoor UAV positioning methods, the proposed method achieves a significant improvement in positioning accuracy, providing a low-cost and high-precision solution for indoor UAV applications.