The aim of this project is to use modified Single Shot Detector (SSD) approach in pedestrian detection and then improve it by two new proposed approaches. In the first proposed approach, we extract initial Region of Interests (RoIs) using SSD approach, while we employ a set of parallel Deep Convolutional Neural Networks (DCNNs) to estimate some different parts of body. In the second proposed approach, we extract initial Region of Interests (RoIs) using SSD approach, while we employ a set of parallel DCNNs to detect pedestrians in some different sets of camera viewing angles. The proposed approaches take the advantage of a secure margin in each initial RoI both to create an Extended Region of Candidate Pedestrian (ERCP) and to extract multi-RoIs. Comprehensive experimental results demonstrate that the proposed methods, Deep Model based on the Best Candidate Pedestrians and Different Parts Extraction (DM-BCP+DPE) and Deep Model based on the Best Candidate Pedestrians and Changes in Camera Viewing Angle (DM-BCP+CCVA) are highly effective methods that achieve very competitive performance on Caltech dataset.