Abstract: In fully mechanized underground coal mining faces, the presence of dense dust, suspended fog, and fluctuating illumination severely impairs image perception quality, posing challenges to the reliable operation of intelligent monitoring systems. Traditional 2D electronic fence methods suffer from limited spatial adaptability, the absence of depth information, and high false detection and omission rates, making accurate identification and early warning of personnel intrusion difficult. To enhance image quality and spatial perception capabilities, this study proposes an image preprocessing approach that integrates the dark channel prior with an improved contrast-limited adaptive histogram equalization (CLAHE) algorithm, enabling precise ambient light estimation and effective enhancement of dust-obscured images. Addressing the limitations of 2D fence recognition, an underground object keypoint detection algorithm—MSS-YOLO (MSEE SCDH SPSC-YOLO)—is developed. Combined with the Feature-Enhanced Semi-Global Matching (FE-SGBM) stereo vision model, the algorithm enables detection of keypoints on the trailing cable trough and acquisition of corresponding spatial depth information, thereby supporting the construction of a dynamic 3D electronic fence. Furthermore, a virtual collision detection method based on Unity3D is introduced to simulate interactions between underground personnel and the dynamic 3D fence, forming an intrusion detection and early warning system tailored for fully mechanized mining faces. Experimental results demonstrate that the proposed system achieves a personnel false detection rate of only 4.56% under conditions of dense dust and varying illumination. The virtual platform attains a collision warning frame matching rate of 94.4% for intrusion scenarios and 98.5% for alarms within hazardous regions, indicating excellent detection accuracy and real-time performance. This study provides a feasible technical framework for constructing intelligent 3D safety warning systems in fully mechanized coal mining environments.