Three dimensional(3D) object recognition was researched under multi-view points. For the shortages of traditional signal feature description for 3D object recognition under multi-view points, a new recognition algorithm fusing multiple features was proposed. Firstly, the object corners were extracted by using the correlation matrixes of anisotropic Gaussian directional derivatives, the particular corners were selected by the skeleton constraint, and the normalized distance between particular corners and the object centroid was taken as the corner descriptor. Then, the geometric moment invariants, affine moment invariants, and the Fourier descriptor of object boundary were extracted, respectively , and the scatter matrixes within and between classes for the four features were calculated. By taking the trace of sample scatter matrix as the weight, the four features were fused. Furthermore, the Independent Component Analysis (ICA) was carried out on the fused vector to obtain independent features. Finally, a Support Vector Machine (SVM) was adopted to complete the whole classification of the experiments. Experimental results show that the recognition accuracy of the proposed approach is higher than that of the signal feature approach by 10% averagely and that in the small training sample (10% of the total samples) condition still achieves more than 80% .It concludes that proposed algorithm meets the demand of theodolites for real-time object recognition.