The issue of safety risks existing when large construction machinery operates in proximity to live-line working is addressed in this paper. A combined early warning method based on image feature recognition and laser point cloud is proposed herein. Firstly, by combining the methods of deep separable convolution and triple attention mechanism, the visual features are extracted, and the calculation formula for the safety distance is deduced. Subsequently, based on clock synchronization and data preprocessing, three-dimensional space reproduction is achieved, and on-site real-time ranging is carried out. Finally, a comparative analysis study of different distances and different methods is conducted to verify the feasibility of the combined early warning method of image feature recognition and laser point cloud. The method presented in this paper provides a novel and feasible solution for the safety distance early warning of construction machinery.