<p>Even though numerous deep-learning-based methods have done a good job of detecting walkers who are easy to see, their total performance is still not good enough, particularly if heavily occluded pedestrians are added to the scene. When there are a lot of people, objects, or pedestrians in the way, only a small part of the body can be used to find someone. The parts of the walker that can be seen are small, but the sizes of the parts that can’t be seen and the parts that can be seen in the identical image are distinct. It is proposed that enhanced proportional feature fusion and CBAM on FPN can combine four standard datasets of occluded people, KITTI, WiderPerson, CrowdHuman, and INRIA, are used to improve key characteristics with new feature data of different sizes. The findings indicate that the suggested method works well with deep learning models like Faster, Cascade, and Mask RCNN to obtain findings like the average precision and miss rate index in restricted human recognition tasks.</p>