Computer Vision
Super high speed camera acquisition with raspberry pi : 660 FPS article The tutorial for 660 FPS (simple)
1000 FPS article The tutorial (complex)
Cheap and precise dimension reading with linear array
Detection of actions using neuronal networks
Photogrametry : level over 9000 - modelling a whole hotel in 3D just with photos and a motion sensor...
Start photogramety with a smartphone camera
OPENCV Ai kit:
Open/convert tensorflow network with opencv
DepthAi:
Docs downloadable here - and API
OAK-1 (or MegaAI) : here
OAK-D (or BW1098OBC) : here
Available trained and optimised models for Aikit : Modelplace.ai
Explanation of One stage (single stage) vs Two stage detectors in DNN machine vision - An explanation of single stage specifically.
Devnote
Can deeplabcut be used to train easily a network and export (how is it saved ? tensorflow style ?) it to be loaded onto OAK-1 ?
If so, the goal would be to track mouse and it's orientation (nose, ears, back, tail in the best conditions) live, onto the camera, and process this information to for example move the hanging point of the rotary joint (less constraint on the animals : better results to behavioral tasks)
Image registration
Summing multiple exposures of the same scene improve signal to noise ratio
Image obtained with the great software Twibright registrator http://ronja.twibright.com/registrator.php
- diffeomorphic mapping (diffeomorphometry) :
-
Image Edge detection ref
-
Filtre de Canny ref -
.PNG)
.PNG)
Le filtre de Canny (ou détecteur de Canny) est utilisé en traitement d'images pour la détection des contours.
.PNG?uselang=fr)
.PNG?uselang=fr)