Pedestrian Detection

In this project, we compare three famous object detection algorithms - Faster RCNN, YOLOv8, and DETR(DEtection TRansformers) for pedestrian detection. For this purpose, we captured a video with pedestrians and vehicles at a busy intersection.

The video below shows the output for Faster RCNN.

The video below shows the output for YOLOv8.

The video below shows the output for DETR.

From the comparison, we observe that DETR excels in accurately detecting pedestrians, including those located far away. While Faster R-CNN performs well overall, it can sometimes misclassify objects such as bikes and parking meters as pedestrians. YOLOv8, on the other hand, shows fewer misclassifications compared to Faster R-CNN but fails to detect a pedestrian crossing the street. Despite these accuracy differences, YOLOv8 stands out for its real-time performance capabilities, whereas Faster R-CNN and DETR require more time for inference.