The invention discloses a method and device for recognizing an object in an image. The method comprises the following steps that corresponding feature vectors are acquired on different extraction positions of an image to be detected; a training feature vector on a node most similar to each feature vector is found in a retrieval tree which is established in advance; the object category which each searched training feature vector belongs to is acquired, and a relative position relation between a position corresponding to each training feature vector and the center of the object is acquired; the number of times when each point in each voting image serves as the center of the object is calculated, and when the number of times when one point in one voting image serves as an estimation position of the center of the object is greater than a preset threshold value, the fact that an object which falls into the object category corresponding to the voting image exists in the image to be detected can be judged. The method and device for recognizing the object in the image solves the technical problem of long detection time caused by the fact that local feature vectors of current positions are contrasted with local feature vectors in a database one by one in the prior art.