Detection of different objects for autonomous driving applications

Summary

The aim of this dissertation is to use real-time video to locate and classify distinct motion objects. Two ways were employed and compared to achieve this. The Berkeley DeepDrive dataset was used to train the two YOLO and Faster RCNN models so that they could compare their performance and create a similar mAP table as well as matching diagrams of normalized total loss and average accuracy (mAP). Then, with a focus on autonomous driving and attempting to compare the models' performance, brief FPS and mAP measurement movies were generated.