Taking eggplant fruits as examples, one method was introduced for recognizing position and attitude of object based on improved generalized Hough transforms (IGHT). Scaling and rotation operations were added to the parameter index list establishment, which enlarged the representative ranges of sample. And the method of “shape similarity degree” was adopted to primarily select scaling index and rotation index of images awaiting recognization, which could effectively reduce the searching scope. Properly increasing step size of gradient index could avoid excessive dispersion of array in accumulator, and make it easy to search the final reference point coordinates. The experiment demonstrated that the depth recovering results for large target and small target were decreased by 4.0 and 7.9 percentage points, respectively, by using improved generalized Hough transforms, and this method is feasible and effective to recognize different pose and partially occluded eggplant fruits.