Grayskull: A tiny computer vision library in C for embedded systems, etc. (github.com)

Grayskull is a minimalist, dependency‑free computer vision library written in portable C99 and delivered as a single header under ~1K lines — explicitly aimed at microcontrollers and other resource‑constrained systems. It operates on grayscale images with integer‑centric routines (no C++, no dynamic allocation required) and implements a compact but practical set of algorithms: image ops (copy, crop, bilinear resize, downsample), filters (blur, Sobel), thresholding (global, Otsu, adaptive), morphology (erode/dilate), connected components/contours, perspective warp, FAST/ORB keypoints & descriptors, and LBP cascade detectors. Utilities include PGM I/O and optional gs_alloc/gs_free helpers; an online demo and examples are provided. For the AI/ML and embedded‑vision community this matters because it lowers the barrier to deploy classical computer‑vision pipelines and lightweight feature‑based tracking/detection on tiny devices. The library’s single‑header, no‑dependency design makes integration trivial and predictable for real‑time/real‑world embedded apps, and integer implementations favor deterministic, memory‑tight builds. Features like ORB keypoints (descriptor array uint32_t[8]), FAST interest scoring, and LBP cascades enable on‑device object detection and tracking without heavy ML stacks. Licensed MIT, Grayskull is ready for research and product use where footprint, simplicity, and control matter.