The field of pedestrian detection in computer vision is complex and has numerous applications, such as assisting with pedestrian safety, monitoring traffic, and enabling autonomous vehicles. However, accurately detecting pedestrians presents various challenges, including detecting them at multiple scales and in crowded environments with varying lighting conditions and postures. Although numerous pedestrian data sets exist, none encompass all the diverse characteristics found in real-world scenarios. When using pre-trained models for pedestrian detection, performance can suffer when applied to unseen data. We initially created a custom dataset (CamPed: Campus Pedestrians) to address this issue and trained it on different YOLO models. While our custom-trained model performs well in detecting small and occluded pedestrians and pedestrians in different view angles and poses, it struggles to differentiate between pedestrians and objects that resemble them in real-world situations. To resolve this challenge, we enriched our CamPed dataset (Enriched-CamPed Dataset: https://github.com/RahulRaman2/Enriched-CamPed-Dataset ) by curating images from the PnPLO dataset and trained it on the YOLOV7s, YOLOV8m, and YOLOV9c models. The custom-trained model outperforms existing pre-trained models on standard datasets like ETH, PenFudan, and PnPLO. Object detection models DETR and RT-DETR were also trained and tested on the Enriched-CamPed Dataset. The results in the results section showcase that models trained on our proposed dataset perform better by reducing false positives during detection, particularly when identifying pedestrians and pedestrian-like objects.