Ewenwan/MVision

Combines multiple autonomous driving perception pipelines (object detection, semantic segmentation, tracking, depth estimation) with visual-inertial SLAM backends and supports multi-sensor fusion through IMU integration for robust localization. Provides comprehensive implementations across the full autonomous stack—from low-level feature extraction and optical flow estimation to high-level motion planning and collision avoidance—with references to production frameworks like Apollo and practical datasets (KITTI). Heavily emphasizes sensor calibration toolchains and offers curated learning resources spanning classical computer vision (HOG+SVM, CRF) to modern deep learning approaches (FCNs, DPMs, end-to-end steering models).

8,589 stars. No commits in the last 6 months.