I want to create an Android app which uses BOW + SVM in native (using C++) for predicting. Unfortunately I have problem with building the native part. Since the non-free module is not included in the OpenCV SDK for Android, I need to build the module myself, using this tutorial. It seems I built the .so file successfully. Here is the output:
[armeabi-v7a] Prebuilt : libopencv_java.so <= /home/crash-id/Development/SDK/OpenCV-2.4.9-android-sdk/sdk/native/jni/../libs/armeabi-v7a/
[armeabi-v7a] SharedLibrary : libnonfree.so
[armeabi-v7a] Install : libnonfree.so => libs/armeabi-v7a/libnonfree.so
[armeabi-v7a] Install : libopencv_java.so => libs/armeabi-v7a/libopencv_java.so
So the problem comes here, when I have to add this .so file to my project. I added libnonfree.so to my jni folder. Then I edited the Android.mk. Here I provide my .mk files.
Android.mk
LOCAL_PATH := $(call my-dir)
include $(CLEAR_VARS)
LOCAL_MODULE := nonfree_prebuilt
LOCAL_SRC_FILES := libnonfree.so
include $(PREBUILT_SHARED_LIBRARY)
include $(CLEAR_VARS)
OPENCV_CAMERA_MODULES := on
OPENCV_INSTALL_MODULES := on
include /home/crash-id/Development/SDK/OpenCV-2.4.9-android-sdk/sdk/native/jni/OpenCV.mk
#LOCAL_SHARED_LIBRARIES := nonfree_prebuilt #if I add this, it says undefined reference for everything in the cv namespace.
LOCAL_SRC_FILES := SVMDetector.cpp
LOCAL_MODULE := svm_detector
LOCAL_C_INCLUDES += /home/crash-id/Development/SDK/OpenCV-2.4.9-android-sdk/sdk/native/jni/include
LOCAL_CFLAGS := -Werror -O3 -ffast-math
LOCAL_LDLIBS += -llog -ldl
include $(BUILD_SHARED_LIBRARY)
Application.mk
APP_STL := gnustl_static
APP_CPPFLAGS := -frtti -fexceptions
APP_ABI := armeabi-v7a
APP_PLATFORM := android-15
But this doesn't work. When I try to build the app, I get the following error:
./obj/local/armeabi-v7a/objs/svm_detector/SVMDetector.o: in function Java_org_elsys_thesisdiploma_cammect_FrameProcess_SVMDetect:jni/SVMDetector.cpp:23: error: undefined reference to 'cv::initModule_nonfree()'
When I click right button on initModule_nonfree();
, Eclipse opens the nonfree.hpp file and here are it's contents:
#ifndef __OPENCV_NONFREE_HPP__
#define __OPENCV_NONFREE_HPP__
#include "opencv2/nonfree/features2d.hpp"
namespace cv
{
CV_EXPORTS_W bool initModule_nonfree();
}
#endif
But I'm not sure the linker knows where is the implementation of this method. Since it gives an error, it doesn't.
EDIT
If I add LOCAL_ALLOW_UNDEFINED_SYMBOLS := true
, the project compiles successfully but it causes run time error:
02-17 00:15:58.197: E/AndroidRuntime(8793): FATAL EXCEPTION: main
02-17 00:15:58.197: E/AndroidRuntime(8793): Process: com.example.cammect, PID: 8793
02-17 00:15:58.197: E/AndroidRuntime(8793): java.lang.UnsatisfiedLinkError: dlopen failed: cannot locate symbol "_ZN2cv18initModule_nonfreeEv" referenced by "libsvm_detector.so"...
Do you know what am I doing wrong? Thanks in advance!
D:\adt-bundle-windows-x86_64-20140702\android-ndk-r10d\
)D:\CODE\OpenCV-2.4.10-android-sdk\
), Download link
D:\CODE\OpenCV-2.4.10\
), Download link
We actually only need to copy a few files from OpenCV-2.4.10
source code to OpenCV-2.4.10-android-sdk
, namely:
Copy the nonfree folder from OpenCV-2.4.10\sources\modules\nonfree\include\opencv2\
to OpenCV-2.4.10-android-sdk\sdk\native\jni\include\opencv2
.
Create a folder to hold our new project for libnonfree.so
. Here, I call it libnonfree
. Create a jni
folder under libnonfree
. Copy the following files from OpenCV-2.4.10\sources\modules\nonfree\src
to libnonfree\jni\
folder:
Building libnonfree.so
:
Create Android.mk
and Application.mk
scripts. This Android.mk
is used to build libnonfree.so
.
OPENCV_PATH
where your OpenCV-2.4.10-android-sdk
is)cd
into the project folder libnonfree
and type ndk-build
to build the libnonfree.so
.
So far, you have got libnonfree.so
along with libopencv_java.so
and libgnustl_shared.so
in libnonfree\libs\armeabi-v7a
folder.
You can easily build any SIFT or SURF applications using those libraries. If you want to use SIFT and SURF in JAVA code in your Android application, you only need to write JNI interfaces for the functions you want to use.
Create a project folder call libnonfree_demo
. Create a jni
folder inside the project folder. Then copy libnonfree.so
along with libopencv_java.so
and libgnustl_shared.so
into jni
.
Create a nonfree_jni.cpp in jni
. It is simple SIFT test program. It basically reads an image and detects the keypoints, then extracts feature descriptors, finally draws the keypoints to an output image.
Create Android.mk
and Application.mk
inside jni
:
OPENCV_PATH
where your OpenCV-2.4.10-android-sdk
is)cd
into the project folder libnonfree_demo
and type ndk-build
to build the libnonfree_demo.so
.
At this point you can easily extend the sample app with your SVMDetector
. Just copy the source and include files int to the folder libnonfree_demo\jni
and add cpp files to LOCAL_SRC_FILES
in Android.mk
.
The whole source can be downloaded from: https://github.com/bkornel/opencv_android_nonfree.
Original source from: http://web.guohuiwang.com/technical-notes/sift_surf_opencv_android
May I add that, in order to use the new libraries in the running application, there are the following steps to do:
1) in your folder libnonfree/libs/[TARGET PLATFORM]/, there are now 3 files: - libgnustl_shared.so - libnonfree.so - libopencv_java.so
in your own project (my IDE is the Android Studio), you have a folder src/main/, with the subfolders: - java - res
create a new folder (if not already there): "jniLibs" [this folder is auto-parsed by Gradle]
COPY the 3 above-mentioned folders under "libnonfree/libs/" into the "jniLibs" folder. you end up with a structure like that: screenshot of the jniLibs folder
/app/src/main/jniLibs/[armeabi, armeabi-v7a, ...]/[libgnustl_shared.so, libopencv_java.so, libnonfree.so]
2) Somewhere in your code, you have a line like this:
OpenCVLoader.initAsync(OpenCVLoader.OPENCV_VERSION_2_4_10, this, mLoaderCallback);
this line tells your app to dynamically load the pre-compiled library from the locally-installed OpenCV Manager. In order to use the self-compiled non-free version, we replace the above line with the following:
if(!OpenCVLoader.initDebug())
{
}
else
{
System.loadLibrary("nonfree");
}
now, we made sure to use the nonfree-included libraries we provide with the app.
3) well, run a SURF descriptor:
Bitmap mPhotograph = BitmapFactory.decodeFile(_image_path);
Mat real_image = new Mat();
Utils.bitmapToMat(mPhotograph, real_image);
MatOfKeyPoint keypoints_real = new MatOfKeyPoint();
FeatureDetector detector = FeatureDetector.create(FeatureDetector.SURF);
detector.detect(real_image, keypoints_real);
while before, the app would return with a bad signal, this time it does its job and you can evaluate the resulting keypoints.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With