With the rapid development of unmanned aerial vehicle (UAV) technology and the increasingly urgent demand for low-altitude security control, unimodal detection methods exhibit significant limitations in complex environments. Fusing multimodal information, including radar, electro-optical, and radio frequency (RF) data, has become a key approach to enhancing low-altitude target detection performance. To address the challenges faced by existing methods in temporal alignment, feature complementarity, and computational efficiency, this paper proposes a real-time detection framework named MRLT-DF (Multimodal Real-time Lightweight Detection Framework) for low-altitude targets based on radar-electro-optical-RF multimodal fusion. The framework employs an asynchronous parallel processing mechanism, combined with an attention-guided soft-gated fusion network and a multi-level adaptive routing strategy, to effectively address the challenges of spatiotemporal alignment, feature complementarity, and real-time inference among heterogeneous multimodal data. By constructing a lightweight encoding network, introducing a cross-modal attention mechanism, and adopting a phased progressive training strategy, the proposed method significantly improves detection accuracy and inference speed in complex environments. Experimental results demonstrate that the proposed method achieves excellent recognition performance across various trajectory lengths, with an accuracy exceeding 93% in long-trajectory tasks, while meeting the stringent real-time response requirements of low-altitude target detection systems. This approach can provide technical support for low-altitude security and UAV management.