Code, Algorithms and Probabilities

A stereo matcher produces a depth for almost every pixel. Roughly one in ten of them is wrong, and the matcher does not say which. Everything downstream then has to treat a measurement it can rely on and one it invented as the same kind of thing: a temporal filter... [Read More]

This is the real-time implementation: dense MASDA running on a Jetson TX2 at 31.7 ms per frame at 848×480 — the camera’s full resolution, faster than the camera delivers frames — with the disparity map bit-identical to the CPU matcher on all eight ground-truth scenes. [Read More]

This is the dense stereo matcher: a C++14 implementation of MASDA — max-sum loopy belief propagation for data association — applied to every pixel of an 848×480 infrared pair, on a Jetson TX2. Part 1 derives the message equations and measures the one-to-one constraint on sparse keypoints. This post is... [Read More]

In Faster Data Association with Max-Sum Loopy Belief Propagation (MASDA) I derived MASDA for the tracking problem: associate measurements with objects, allowing for clutter and misdetection. Stereo matching has the same structure, so this post applies MASDA there — to the dense problem, a disparity for every pixel — and... [Read More]

This presents a data association algorithm I call MASDA (Max-Sum Algorithm Data Association) which relies on message passing in a factor graph. [Read More]