Object Detection and Tracking for Autonomous Vehicles using Deep Learning Technique- YOLO

For self-driving cars, Optical Perception is the most important factor. By accurately detecting road signs, foottravelers and vehicles, it could help and guide auto driven vehicles to drive as safely as human beings. In recent times, several CNN (Convolutional Neural Network) centered classification- after-localization methods have increased detection results under various situations. It is not possible for the two-stage methods to be used in real-time situations as it has very slow recognition speed. In recent times, (YOLO) You Only Look Once, was anticipated, can straight forwardly degenerate to object class positions and scores from input image. This model has single network structure and has greater detection precision than present-day real-time methods. This model handles images at 45 frames per second on PASCAL VOC 2007 dataset. In this detailed study and research, YOLO model is implemented to verify its general applicability, on three unlike datasets. This helped to completely scrutinized its behavior from several features on KITTI dataset which is specifically dedicated for self-directed driving. KITTI is my key dataset. In addition, for pre-training, ImageNet dataset is used and rest three datasets are implemented for other class domains are FDDB (Face Detection Dataset & Benchmarks), Road Signs, Pascal VOC-2007/2012