Systems registry
researchgithub

Velocity Estimation with Optical Flow

Connects physical-world motion, perception models, calibration assumptions, and embedded capture constraints.

Video Stream
Vehicle ROI
Optical Flow
Motion Conversion
Velocity Output

Repo test video

System Overview

Vehicle speed estimation system using classical optical flow, RAFT, YOLO, Raspberry Pi, and camera-stream constraints.

Research/build system.

Implementation Signals

PythonOpenCVRAFTYOLOKITTIRaspberry Pi

System flow

A compact view of how inputs move through processing, orchestration, validation, and output.

01

Video Stream

02

Vehicle ROI

03

Optical Flow

04

Motion Conversion

05

Velocity Output

Engineering decisions

Decision Record

Compare classical and deep optical flow

Problem: Real-time motion estimation has tradeoffs between speed, accuracy, and hardware constraints.

Approach: Explore Lucas-Kanade, Farneback, RAFT, YOLO ROI handling, and Pi Camera streaming.

Tradeoff: Deep methods add compute cost.

Outcome: Clearer understanding of calibration, frame rate, ROI stability, and embedded limits.

Results / Learnings

Repo

Package notes document RAFT, YOLO, KITTI evaluation, and Pi Camera streaming.

Learning

Speed estimation quality depends on calibration, distance, frame rate, and ROI stability.

Progressive depth

This page keeps the outcome and architecture visible first. Implementation stack, decisions, constraints, and media are available below so technical depth is opt-in rather than forced.

Adjacent systems