Ask Your Question

ANDROID: Using 20+ CascadeClassifiers to detect different objects in camera frames. Is realtime still possible?

asked 2013-09-01 03:30:44 -0500

Max Muster gravatar image

updated 2013-09-01 03:32:57 -0500

hey there.

as you can see from my past questions, im currently working on an app that detects different buttons in a car. im using CameraBridgeViewBase / JavaCameraView to grab frames from the camera. well, in the onCameraFrame()-method i let my detectors try to detect the objects that they were trained for. currently i only got 2 trained classifiers and i get about 15 fps on older devices and about 30 fps on newer devices.

my question: will i still be able to get those FPS when im using 20 or more classifiers (each trained for one button) in this method to detect all the different buttons the could appear in a picture or will that many classifiers slow down the program? i am using a very small area inside each frame that will be checked for buttons only to avoid unnecessary time loss.

edit retag flag offensive close merge delete



thanks for your commet, i will see what i can do about that :) i will let you know once more questions about this come up.

Max Muster gravatar imageMax Muster ( 2013-09-03 00:49:41 -0500 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2013-09-02 03:57:35 -0500

I am afraid NO :) It is already quite challenging to make a robust model and let it work realtime, let alone stand out 20 models. HOWEVER, you could do some processing of the frame, by using simple filters like a naive bayes or a basic color thresholding, ... It can make you destinguish buttons into categories, resulting in only needing to apply some cascades. 2 or 3 cascades might be possible to do realtime with enough processing power.

edit flag offensive delete link more

Question Tools


Asked: 2013-09-01 03:30:44 -0500

Seen: 749 times

Last updated: Sep 01 '13