Comparison of Autonomy and Study of Deep Learning Tools for Object Detection in Autonomous Self Driving Vehicles

Abstract

Computer vision is a field of computer systems that can comprehend images and scenes and also identify them. Computer vision technology is very popular in recent times and it has numerous applications namely smart monitoring and surveillance, drones, medicine, sports, entertainment, industrial robotics and self driving cars. Image localization, classification and detection are the basis of the above technology applications. Current improvements in Convolutional Neural Networks i.e. CNNs also lead to exceptional feat in tasks related to visual identification. CNNs are largely applicable for image classification tasks, particularly, looking at an image and classifying various features in it. In this paper study is carried out on various computer vision algorithms and deep learning techniques and corresponding tools for object detection in autonomous self driving cars. Deep learning has revolutionized computer vision techniques namely for object detection. CNN for object detection include three main deep learning techniques; SSD (Single shot multibox detector), R-CNN i.e. Region based CNN, R-FCN i.e. Region based fully convolutional networks. In this paper, an effort is made to show how above deep learning techniques are effective in ameliorating pre-trained networks' performance using existing datasets under different weather and terrestrial conditions. A brief comparison is also done for human centered autonomy and full autonomy for self driving cars.