The application of autonomous vehicles has the potential to significantly lessen the variety of current harmful externalities, such as accidents, traffic congestion, security, and degradation of the environment. Therefore, autonomous vehicles become an emerging topic of research, and different papers have been published by a large number of global researchers in recent years. And in this paper, a literature review of autonomous vehicle development was conducted and many different Deep Learning models are built using Image and LiDAR point cloud data for Object Recognition and Classification for the navigation system. Different filtering techniques are being proposed in the study to optimize the object recognition and classification in those models, based on three critical assessment factors: performance, user experience, and sustainability. One of the notable findings in the Literature Review section is that autonomous vehicles will become an indispensable and greener solution in the long-term vision, even though they are not yet an environmental-friendly solution in the current transportation system. 5 different deep learning models, YOLOv5s, EfficientNet-B7, Xception, MobilenetV3, and InceptionV4, are built and analyzed for the 2D object recognition task, with the finding that YOLOv5s and EfficientNet-B7 are the best among them by the 3 assessment factors. Moreover, the study proposed Hessian, Laplacian, and Hessian-based Ridge Detection filters for object detection and classification in 2D image data. The results showcase that they can increase the mean Average Precision up to 11.81%, reduce the detection time up to 43.98%, and significantly lessen the energy consumption by up to 50.69% when applied to YOLOv5s and EfficientNet-B7 models. The research also proposed Random Sample Consensus (RANSAC) and Least Median of Squares (LMedS) ground filtering techniques for LiDAR point cloud data in object recognition. The outcomes indicate that the models can be more 10.14% efficient, take 35.75% less detection time, and use 30.63% less energy when using ground filtering. Overall, all the experiment results are very intriguing and can be used in other research studies. The outcomes can act as a reference to optimizing not only object recognition, but also other cognitive tasks in autonomous vehicles. These results may lead to more studies in the future into various filtering algorithms for object detection and classification in autonomous cars, allowing researchers and businesses to involve in further development. And different recommendations and future work have been clearly defined in the study.