In the procedure of surface defects detection of large-aperture aspheric optical elements, it is necessary to scan the surface of the element to achieve full coverage inspection. Since the curvature of the aspherical element is constantly changing from the center to the edge, it is of great difficulty to carry out efficient path planning. In addition, the machine vision system is a microscopic system with limited depth of field, and the sub-aperture imaging of aspherical elements has a visual depth along the object side. When the object depth is greater than the depth of field, out-of-focus blur will generate, so the object depth needs to be as small as possible. In response to these problems, this paper proposes a fast path planning algorithm based on the minimum object depth of a sub-aperture. To ensure minimum object depth, the machine vision system collects images along the normal direction of the sub-aperture plane. To address the problem of the surface curvatures of aspheric elements being different and the overlap coefficient difficult to determine, this paper proposes an image processing based overlap coefficient self-optimization algorithm. When scanning with full coverage of elements, there is only one connected domain in the horizontal projection image of all sub-apertures. According to this premise, the overlap coefficient is optimized through an image processing method to obtain a local optimal path planning strategy. According to the obtained path planning strategy, combining the component parameters and mechanical structure, the mapping matrix of the path planning algorithm transplanted to the detection system is calculated. Through computer programming, automatic sub-aperture acquisition is realized, and the self-edited sub-aperture stitching program is applied to reconstruct the collected sub-apertures. Our algorithm can complete path planning within 5 s, and the experimental results show that the maximum stitching misalignment error of the collected sub-apertures is no more than four pixels, and the average is one pixel. The reconstruction accuracy satisfies the needs of subsequent image processing and digital quantization.