The invention provides a collision detection method for objects in a large-scale scene. The method includes the first step of establishing an AABB bounding box, the second step of conducting dimension reduction processing, the third step of obtaining a dynamic list and the fourth step of carrying out collision detection calculation. The Sweep and Prune algorithm is comprehensively used, objects which do not collide in the future are removed by scanning states of moving objects in the three-dimensional scene, and therefore the calculation quantity of collision detection is greatly reduced, and real-time rendering efficiency of a physical engine is improved. The latest research achievements of current computer vision and physical engine technical fields are introduced into a power transmission line inspection simulating training system, and a virtual power transmission line inspection training environment which has high simulation, is interactive and is deeply immersed is showed to trainees on the basis of an efficient large-scale scene collision detection algorithm. The collision detection method lays key theory and application foundations for the collision detection technology and physical engine technology in electric system simulation training and has very good application prospects.