A Hybrid Approach Using YOLO and Deep Neural Networks to Enhance Real-Time Perception and Navigation in Self-Driving Cars

Due to the enhancement of self-driving vehicles, there is an increasing interest in research and development of new intelligent and high-efficient perception systems. The basic image processing methods have a hard time meeting the real-time constraints, complex road scenarios and significant environmental variations. To overcome these issues, the paper proposes an improved autonomous vehicles’ image processing framework that employs deep learning-based object detection, lane recognition, and environmental perception. The lane recognition is implemented using a hybrid model that includes Canny Edge Detection, Hough Transform, and a convolutional neural network. The system is based on YOLO (You Only Look Once) real time object detection for accurate AND timely detection of pedestrians, vehicles, and other obstacles within the surrounding environment. Weather dependent image enhancement provides better visibility in adverse condition like fog, rain, and low-light scenarios. The evaluation of the system and the effectiveness in comparison to the existing methods has been performed using publicly available benchmarking datasets like KITTI and Cityscapes along with the real-life working test scenarios. Key performance metrics included: detection accuracy, processing speed (frames per second), false positive rate, Mean Intersection over Union (mIoU), and robustness to environmental variation. The results of the experiment show that the new method is more efficient than conventional techniques and achieves a higher level of accuracy in detection without compromising on the speed of processing. By utilizing deep learning and sophisticated imaging processing methods, this study aids in creating more dependable autonomous vehicle perception systems. Further refinements will address the issues of sensor integration and real-time optimization for broad application in autonomous driving systems.