Extended-depth-of-field object detection with wavefront coding imaging system

As an important issue in the field of computer vision, object detection has broad application prospects. Recent researches using convolutional neural networks (CNN) have shown the state-of-the-art results in the challenge competition. Most of them focused on improving the precision under ideal imaging conditions. However, it is hard to ensure that the optical imaging system works in the focused state in practice. In this study, we examine the impact of defocus on detection accuracy. The results show that even the state-of-the-art network is sensitive to a different defocus situation. Thus we put forward wavefront coding (WFC) technique for improving the performance over a large range of depth of field (DOF). Simulation results indicate the improvement on average precision of detection results by applying WFC. In addition, we propose a novel WFC method for overcoming the defects of the traditional one. Then the optical imaging system is designed under the guidance of the proposed theory. Experiments are conducted to suggest that the detection accuracy rate can be enhanced considerably with WFC.