Autonomous Space Robotics Lab Speeded Up SURF GpuSurfFeatures.hpp Go to the documentation of this file. /* Copyright (c) 2010, Paul Furgale and Chi Hay Tong All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. * The names of its contributors may not be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef ASRL_GPU_SURF_FEATURES #define ASRL_GPU_SURF_FEATURES //#include <builtin_types.h> #include "CudaSynchronizedMemory.hpp" #include "gpu_globals.h" namespace asrl { class GpuSurfFeatures { public: GpuSurfFeatures(); ~GpuSurfFeatures(); Keypoint * deviceFeatures(){return m_features.d_get();} Keypoint * hostFeatures(){return m_features.h_get();} float * deviceDescriptors(){return m_descriptors.d_get();} float * hostDescriptors(){ return m_descriptors.h_get();} unsigned int * deviceFtCount(){return m_feature_counter.d_get(); } unsigned int * hostFtCount(){return m_feature_counter.h_get(); } Keypoint * downloadFeatures(); unsigned int ftCount(); unsigned int rawFeatureCount(); void clearFeatureCounts(); float * downloadDescriptors(); void setDirty(); CudaSynchronizedMemory<Keypoint> & featuresMem(){return m_features;} CudaSynchronizedMemory<float> & descriptorsMem(){ return m_descriptors;} CudaSynchronizedMemory<unsigned int> & featureCounterMem(){ return m_feature_counter;} CudaSynchronizedMemory<int4> & rawFeatureMem(){ return m_rawFeatures; } template<typename CV_KEYPOINT_VECTOR> void getKeypoints(CV_KEYPOINT_VECTOR & outKeypoints); template<typename CV_KEYPOINT_VECTOR> void getRawKeypoints(CV_KEYPOINT_VECTOR & outKeypoints); template<typename CV_KEYPOINT_VECTOR> void setKeypoints(CV_KEYPOINT_VECTOR const & inKeypoints); private: CudaSynchronizedMemory<Keypoint> m_features; CudaSynchronizedMemory<float> m_descriptors; CudaSynchronizedMemory<unsigned int> m_feature_counter; CudaSynchronizedMemory<int4> m_rawFeatures; bool m_descriptorsDirty; bool m_featuresDirty; bool m_countDirty; }; // This is overly clever. Using a template here avoids including // the OpenCV header which CUDA has terrible problems with. template<typename CV_KEYPOINT_VECTOR> void GpuSurfFeatures::getKeypoints(CV_KEYPOINT_VECTOR & outKeypoints) { // This has an implicit "Thread synchronize" int ftcount = ftCount(); featuresMem().pullFromDevice(); //m_features->descriptorsMem().pullFromDevice(); outKeypoints.resize(ftcount); if(ftcount > 0) { for(int i = 0; i < ftcount; i++) { typename CV_KEYPOINT_VECTOR::value_type & k = outKeypoints[i]; Keypoint * ft = hostFeatures() + i; k.pt.x = ft->x; k.pt.y = ft->y; k.size = ft->size; k.octave= ft->octave; k.angle = ft->angle; k.response = ft->response; } } } template<typename CV_KEYPOINT_VECTOR> void GpuSurfFeatures::getRawKeypoints(CV_KEYPOINT_VECTOR & outKeypoints) { // This has an implicit "Thread synchronize" m_feature_counter.pullFromDevice(); m_rawFeatures.pullFromDevice(); int ftcount = m_feature_counter[1]; outKeypoints.resize(ftcount); for(int i = 0; i < ftcount; i++) { // {hidx_x, hidx_y, threadIdx.z, octave}; typename CV_KEYPOINT_VECTOR::value_type & k = outKeypoints[i]; k.pt.x = (float)m_rawFeatures[i].x; k.pt.y = (float)m_rawFeatures[i].y; k.response = (float)m_rawFeatures[i].z; k.octave= m_rawFeatures[i].w; k.angle = 0; } } template<typename CV_KEYPOINT_VECTOR> void GpuSurfFeatures::setKeypoints(CV_KEYPOINT_VECTOR const & inKeypoints){ ASRL_ASSERT_LT(inKeypoints.size(), ASRL_SURF_MAX_FEATURES, "The device is unable to hold " << inKeypoints.size() << " features. Max: " << ASRL_SURF_MAX_FEATURES); m_feature_counter[0] = inKeypoints.size(); m_feature_counter.pushToDevice(); for(unsigned i = 0; i < inKeypoints.size(); i++) { typename CV_KEYPOINT_VECTOR::value_type const & k = inKeypoints[i]; Keypoint & ft = m_features[i]; ft.x = k.pt.x; ft.y = k.pt.y; ft.size = k.size; ft.angle = k.angle; ft.response = k.response; ft.octave = k.octave; } m_features.pushToDevice(inKeypoints.size()); } } // namespace asrl #endif