Pedestrian Detection and Tracking in Challenging Conditions

Driver safety is essential in this time of the world as the roads are filled with vehicles and pedestrians and accidents frequently occur especially during tough weather conditions. We used the YOLOv4 model which is pretrained with generic images with the standard COCO weights to detect pedestrians. We chose a custom dataset consisting various other datasets which we used for training our new improved YOLOv4. We got the performance metrics of the YOLOv4 detection model which is compared with the Improved YOLOv4 detection model and resulted with an improvement of mean Average Precision (mAP) by 7%.