Occluded pedestrian identification is a critical and challenging issue in the field of pedestrian detection. Currently, using pedestrian components or human approaches is the most popular way to get over the problem of occlusion in pedestrian recognition. Only a small portion of the body may be used for detection when there is severe occlusion from crowds, or human or pedestrian components. The viewable portions of the occluded pedestrian have a small scale, but the scales of the unobscured and occluded pedestrians in the same image are different. Improved Parallel Feature Fusion with CBAM (IPFF-CBAM) on Feature Pyramid Network is put forward that can integrate new feature data of various sizes which are applied to four benchmark datasets KITTI, WiderPerson, CrowdHuman, and INRIA individuals of occluded pedestrians, in order to enhance key attributes. According to the findings the proposed method performs satisfactorily on deep learning approaches i.e., Faster RCNN, Cascade RCNN and Mask RCNN to obtain results of parameters like Average Precision (AP) and Miss rate index (MR) in obstructed pedestrian detection tasks.