Pedestrian Detection in Robotic Response for the First Responders

The paper presents two object detector networks on USC Pedestrian database in order to detect and localize pedestrians in photos. The COCO metrics was used, because it was considered a very important problem, well addressed. All children in pictures were well localized and the models can be used on autonomous robots and cars. Both Faster R-CNN and SSD missed a single person in the testing C dataset. The research results lead to a precision of 0.733 on Faster R-CNN and 0.583 on SSD on COCO metrics. In future works this results will be used in augmenting data for search and rescue missions conducted by first responders.