I\'m using OpenCV to detect an image. Here is my problem: my function detect_image(mRgba)
needs some time to perform operations and give some results. While functio
To just get a taste at parallelization, the simple approach would be to just use an AsyncTask to process your images:
AsyncTask reference page
A more friendly introduction can be found here:
http://android-developers.blogspot.co.il/2010/07/multithreading-for-performance.html
while this:
http://developer.att.com/developer/forward.jsp?passedItemId=11900176
is a nice all-around introduction to multi-threading on Android.
If you want to just get started, a simple algorithm should work like this:
With this algorithm, if your system can detect 4 images per second while taking a preview at 60fps (for example), you will be able to get a smooth video with a new result about each 20-30 frames on a single processor device, under the realistic assumption that detect_image is CPU intensive while the camera preview/display are I/O intensive.
Capture: x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.... Processing: 1.......1.......1.......1.....1.......1.... time ------------------------------------>
Starting with HoneyComb, a more refined approach would be to account for the number of cores in your CPU (multicore phones/tablets are becoming increasingly common) and start N AsyncTask in parallel (one for each core), feeding a different preview image to each one (maybe using a thread pool...).
If you separate each thread by a fixed delay (about the duration of detectImage/N ), you should get a constant stream of results with a frequency that should be a multiple of the single threaded version.
Capture: x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.x.... Processing: 1.2.3.4.1.2.3.4.1.2.3.4.1.2.3.4.1.2.3.4.... time ------------------------------------>
Hope this helps