Region-based Fully ConvNet (R-FCN) designed for general object detection is difficult to be directly applied for pedestrian detection, due to being with large human pose and scale changes, and even with partial occlusion in surveillance scenarios. This paper presents a rapid pedestrian detection method with partial occlusion handling, which builds on the framework of R-FCN. We introduce a deep Omega-shape feature learning and multi-paths detection to make our detector be robust to human pose and scale changes. A novel predicted boxes fusion strategy is proposed to reduce the number of false negatives caused by partial occlusion in crowded environment. Our end-to-end approach achieved 95.35% mAP on the Caltech dataset, 96.22% mAP on DukeMTMC dataset and 97.43% mAP on Bronze dataset at a test-time speed of approximate 86 ms per image.