The constrained multi-UAV platform is composed of multiple unmanned aerial vehicles (UAVs) connected physically and mutually. When multiple UAVs operate collaboratively on an interconnected platform, challenges in external environment perception arise due to limited movement range and sensor disturbance among UAVs. Additionally, real-time perception is crucial for various UAV tasks, posing a significant challenge for conventional single-modal algorithms to balance speed and accuracy effectively. This paper proposes an innovative multi-modal collaborative sensing method for multi-UAV systems, integrating point cloud and image processing algorithms. A feature-level fusion is employed by our framework to accurately determine target positioning in real-time. The multi-UAV decision layer utilizes a Kalman Filter to extract data from multiple individual outputs, yielding stable and accurate positioning and classification information. Simulation experiments validate the robustness of this method, reducing disturbance from structural and mutual obstructions among UAVs, and providing accurate target positioning.