Aiming at the influences on the operational positioning of distribution network live working robots caused by outdoor lighting, vibration of aerial devices, varied operation scenes, and other factors, this paper proposed an intelligent perception method. This method is based on scene semantics which integrated multidirectional data perception, automatic device object recognition, automatic assistance of device semantics and other technics as a whole to solve the problems of difficult target object recognition and low positioning accuracy in the operation of distribution network live connection of leads. This paper presented the spatial 3D point cloud perception and acquisition algorithm, which adopted the method of carrying laser radar on tail end of the operational mechanical arm to perform full situational awareness and scanning to realize multi-angle and multi-position point cloud data acquisition, avoided interference of strong light on the sensor. This paper discussed the point cloud connected domain search strategy algorithm based on device semantics, which conquered the challenge of the lead target recognition under the conditions of uneven data acquisition, edge noisy point and acquisition disconnection in strong light. With regard to the standard operational requirements on live connection of drainage leads, this paper realized automatic selection of the positions and postures of working row lines and lead operating points, by means of combining semantic location information of the leads, working arm and other devices.