Self-Driving Car Using Deep Learning and OpenCV

Engineering notes

Recent advances in techniques of deep learning have significantly enhanced the perception component of such systems.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Self-driving cars are capable of navigating city traffic and have long been studied using architectures that are decomposed into perception, planning, and control components. Recent advances in techniques of deep learning have significantly enhanced the perception component of such systems. These techniques have laid the groundwork for various approaches such as end-to-end learning of steering commands and driving affordances directly from camera images. However, many of the studies are only end-to-end approaches or use traditional image processing techniques in simplified traffic scenarios. In this paper, we propose a modularized architecture and an enhanced design system for miniaturized prototype of a car with extended drive situations which includes traffic sign and light recognition, lane detection, lane change and obstacle detection and avoidance. It integrates state of the art deep learning approaches, such as Haar cascade detectors, and classic computer vision based methods, such as Canny edge detectors, running on a Raspberry Pi. The performance of this hybrid framework achieved the real-world autonomous driving in a limited embedded system.

Links and sources

• Official / arXiv page

Comments