The invention discloses a power patrolling task-oriented method for binocular vision obstacle sensing of an unmanned aerial vehicle and belongs to the technical field of machine vision. The method comprises the steps of designing a dual template ratio detection operator to extract power line pixels from a binocular aviation image and performing line detection based on line fitting hypothesis by combination with Hough transform; constructing two pairs of homonymy points on a power line based on an epipolar geometry constraint principle, obtaining a fitting linear equation, and calculating the shortest distance between the unmanned aerial vehicle and the power line; according to other typical obstacles, performing parallel completion of overall three-dimensional reconstruction, constructing a flight safety channel of the unmanned aerial vehicle, and adopting a specific policy to extract obstacles; and finally outputting orientation and physical dimension information of the obstacles relative to the unmanned aerial vehicle. The method includes all static threat information of a corridor environment covered by the unmanned aerial vehicle power inspection, the calculation is small, a sensing result is more accurate, and the real-time requirements of obstacle sensing are met.