In the industrial production quality control process, the detection of defects in automotive parts has received significant attention in the last few decades. The traditional visual inspection process may be superior in some aspects but very prone to failure. The application of deep learning algorithms has led to improvements in the efficiency and reduction of human error in automotive defect identification systems. However, the real-time performance of such systems is challenging in three aspects namely memory, processing time, and hardware limitations. This comprehensive survey presents brief insights into deep neural network-based defect detection and classification methods. Firstly, a brief background of defects within the automobile assemblies like bearing grooves, gear surface, wheelbase, casting, and engine cylinder, is discussed with their characteristics. Secondly, recent deep learning-based video processing techniques are discussed. Third, recent deep neural network architectures such as CNN, LSTM, RNN, and RCNN-based methods are reviewed with their characteristics, strengths, and performance. Through investigation, we found that CNN-based architectures like YOLO outperform other DNN methods in real time with high precision and fast detection. Finally, we outline available datasets and current research challenges in this domain.