A compound eye has the advantages of a large field of view, high sensitivity, and compact structure, showing that it can be applicable for 3D object detection. In this work, an artificial compound eye system is developed for 3D object detection, consisting of a layer of lenslets and a prism-like beam-steering lens. A calibration method is developed for this system, with which the correspondences between incident light rays and the relevant image points can be obtained precisely using an active calibration pattern at multiple positions. Theoretically, calibration patterns at two positions are sufficient for system calibration, although more positions will increase the accuracy of the result. 3D positions of point objects are calculated to evaluate the system, which are obtained by the intersection of multiple incident light rays in the least-squares sense. Experimental results show that the system can detect an object with angular accuracy of better than 1 mrad, demonstrating the feasibility of the proposed compound eye system. With a 2D scanning device, the system can be extended for general object detection in 3D space.