diff --git a/src/openalpr/CMakeLists.txt b/src/openalpr/CMakeLists.txt
index 787e336..d9f3f8c 100644
--- a/src/openalpr/CMakeLists.txt
+++ b/src/openalpr/CMakeLists.txt
@@ -7,6 +7,7 @@ set(lpr_source_files
  config.cpp
  detection/detector.cpp
  detection/detectorcpu.cpp
+ detection/detectorcuda.cpp
  detection/detectorfactory.cpp
  licenseplatecandidate.cpp
  utility.cpp
diff --git a/src/openalpr/config.cpp b/src/openalpr/config.cpp
index 23dabfb..62881a4 100644
--- a/src/openalpr/config.cpp
+++ b/src/openalpr/config.cpp
@@ -130,6 +130,7 @@ namespace alpr
 
     runtimeBaseDir = getString("common", "runtime_dir", "/usr/share/openalpr/runtime_data");
 
+	  gpu_mode = getInt("common", "gpu_mode", GPU_OFF);
     detection_iteration_increase = getFloat("common", "detection_iteration_increase", 1.1);
     detectionStrictness = getInt("common", "detection_strictness", 3);
     maxPlateWidthPercent = getFloat("common", "max_plate_width_percent", 100);
diff --git a/src/openalpr/config.h b/src/openalpr/config.h
index e851f00..b388963 100644
--- a/src/openalpr/config.h
+++ b/src/openalpr/config.h
@@ -47,6 +47,8 @@ namespace alpr
 
       std::string country;
 
+      int gpu_mode;
+
       float detection_iteration_increase;
       int detectionStrictness;
       float maxPlateWidthPercent;
@@ -135,5 +137,13 @@ namespace alpr
       bool getBoolean(std::string section, std::string key, bool defaultValue);
   };
 
+
+  enum GPU_MODE
+  {
+    GPU_OFF=0,
+    GPU_CUDA=1,
+    GPU_OPENCL=2
+  };
+
 }
 #endif // OPENALPR_CONFIG_H
\ No newline at end of file
diff --git a/src/openalpr/detection/detectorcuda.cpp b/src/openalpr/detection/detectorcuda.cpp
new file mode 100644
index 0000000..64c9a7b
--- /dev/null
+++ b/src/openalpr/detection/detectorcuda.cpp
@@ -0,0 +1,140 @@
+/*
+ * Copyright (c) 2013 New Designs Unlimited, LLC
+ * Opensource Automated License Plate Recognition [http://www.openalpr.com]
+ *
+ * This file is part of OpenAlpr.
+ *
+ * OpenAlpr is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License
+ * version 3 as published by the Free Software Foundation
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+*/
+
+
+#include "detectorcuda.h"
+
+
+using namespace cv;
+using namespace std;
+
+namespace alpr
+{
+
+  DetectorCUDA::DetectorCUDA(Config* config) : Detector(config) {
+
+
+
+    if( this->cuda_cascade.load( config->getCascadeRuntimeDir() + config->country + ".xml" ) )
+    {
+      this->loaded = true;
+      printf("--(!)Loaded CUDA classifier\n");
+    }
+    else
+    {
+      this->loaded = false;
+      printf("--(!)Error loading CUDA classifier\n");
+    }
+  }
+
+
+  DetectorCUDA::~DetectorCUDA() {
+  }
+
+  vector<PlateRegion> DetectorCUDA::detect(Mat frame, std::vector<cv::Rect> regionsOfInterest)
+  {
+
+    Mat frame_gray;
+    cvtColor( frame, frame_gray, CV_BGR2GRAY );
+
+    vector<PlateRegion> detectedRegions;
+    for (int i = 0; i < regionsOfInterest.size(); i++)
+    {
+      Mat cropped = frame_gray(regionsOfInterest[i]);
+      vector<PlateRegion> subRegions = doCascade(cropped, regionsOfInterest[i].x, regionsOfInterest[i].y);
+
+      for (int j = 0; j < subRegions.size(); j++)
+        detectedRegions.push_back(subRegions[j]);
+    }
+    return detectedRegions;
+  }
+
+  vector<PlateRegion> DetectorCUDA::doCascade(Mat frame, int offset_x, int offset_y)
+  {
+
+
+    if (frame.cols > config->maxDetectionInputWidth)
+    {
+      // The frame is too wide
+      this->scale_factor = ((float) config->maxDetectionInputWidth) / ((float) frame.cols);
+
+      if (config->debugGeneral)
+        std::cout << "Input detection image is too wide.  Resizing with scale: " << this->scale_factor << endl;
+    }
+    else if (frame.rows > config->maxDetectionInputHeight)
+    {
+      // The frame is too tall
+      this->scale_factor = ((float) config->maxDetectionInputHeight) / ((float) frame.rows);
+
+      if (config->debugGeneral)
+        std::cout << "Input detection image is too tall.  Resizing with scale: " << this->scale_factor << endl;
+    }
+
+    int w = frame.size().width;
+    int h = frame.size().height;
+
+    vector<Rect> plates;
+
+    equalizeHist( frame, frame );
+    resize(frame, frame, Size(w * this->scale_factor, h * this->scale_factor));
+
+    //-- Detect plates
+    timespec startTime;
+    getTime(&startTime);
+
+    float maxWidth = ((float) w) * (config->maxPlateWidthPercent / 100.0f) * this->scale_factor;
+    float maxHeight = ((float) h) * (config->maxPlateHeightPercent / 100.0f) * this->scale_factor;
+    Size minSize(config->minPlateSizeWidthPx * this->scale_factor, config->minPlateSizeHeightPx * this->scale_factor);
+
+    gpu::GpuMat cudaFrame, plateregions_buffer;
+    Mat plateregions_downloaded;
+
+    cudaFrame.upload(frame);
+    int numdetected = cuda_cascade.detectMultiScale(cudaFrame, plateregions_buffer, (double) config->detection_iteration_increase, config->detectionStrictness, minSize);
+    plateregions_buffer.colRange(0, numdetected).download(plateregions_downloaded);
+
+    for (int i = 0; i < numdetected; ++i)
+    {
+      plates.push_back(plateregions_downloaded.ptr<cv::Rect>()[i]);
+    }
+
+
+
+    if (config->debugTiming)
+    {
+      timespec endTime;
+      getTime(&endTime);
+      cout << "LBP Time: " << diffclock(startTime, endTime) << "ms." << endl;
+    }
+
+    for( uint i = 0; i < plates.size(); i++ )
+    {
+      plates[i].x = (plates[i].x / scale_factor) + offset_x;
+      plates[i].y = (plates[i].y / scale_factor) + offset_y;
+      plates[i].width = plates[i].width / scale_factor;
+      plates[i].height = plates[i].height / scale_factor;
+    }
+
+    vector<PlateRegion> orderedRegions = aggregateRegions(plates);
+
+    return orderedRegions;
+
+  }
+
+}
\ No newline at end of file
diff --git a/src/openalpr/detection/detectorcuda.h b/src/openalpr/detection/detectorcuda.h
new file mode 100644
index 0000000..927e046
--- /dev/null
+++ b/src/openalpr/detection/detectorcuda.h
@@ -0,0 +1,55 @@
+/*
+ * Copyright (c) 2013 New Designs Unlimited, LLC
+ * Opensource Automated License Plate Recognition [http://www.openalpr.com]
+ *
+ * This file is part of OpenAlpr.
+ *
+ * OpenAlpr is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License
+ * version 3 as published by the Free Software Foundation
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef OPENALPR_DETECTORCUDA_H
+#define	OPENALPR_DETECTORCUDA_H
+
+#include <stdio.h>
+#include <iostream>
+#include <vector>
+
+#include "opencv2/objdetect/objdetect.hpp"
+#include "opencv2/imgproc/imgproc.hpp"
+#include "opencv2/core/core.hpp"
+#include "opencv2/ml/ml.hpp"
+#include "opencv2/gpu/gpu.hpp"
+
+#include "detector.h"
+
+namespace alpr
+{
+
+  class DetectorCUDA : public Detector {
+  public:
+      DetectorCUDA(Config* config);
+      virtual ~DetectorCUDA();
+
+      std::vector<PlateRegion> detect(cv::Mat frame, std::vector<cv::Rect> regionsOfInterest);
+
+  private:
+
+      cv::gpu::CascadeClassifier_GPU cuda_cascade;
+
+      std::vector<PlateRegion> doCascade(cv::Mat frame, int offset_x, int offset_y);
+  };
+
+}
+  
+#endif	/* OPENALPR_DETECTORCUDA_H */
+
diff --git a/src/openalpr/detection/detectorfactory.cpp b/src/openalpr/detection/detectorfactory.cpp
index cdc9617..07d7e94 100644
--- a/src/openalpr/detection/detectorfactory.cpp
+++ b/src/openalpr/detection/detectorfactory.cpp
@@ -2,10 +2,18 @@
 
 namespace alpr
 {
-
   Detector* createDetector(Config* config)
   {
+    if (config->gpu_mode == 0)
+    {
+      // CPU mode
       return new DetectorCPU(config);
+    }
+    else if (config->gpu_mode == 1)
+    {
+      return new DetectorCUDA(config);
+    }
   }
 
-}
\ No newline at end of file
+}
+
diff --git a/src/openalpr/detection/detectorfactory.h b/src/openalpr/detection/detectorfactory.h
index 8e3d926..8b50800 100644
--- a/src/openalpr/detection/detectorfactory.h
+++ b/src/openalpr/detection/detectorfactory.h
@@ -21,6 +21,7 @@
 #define	OPENALPR_DETECTORFACTORY_H
 
 #include "detectorcpu.h"
+#include "detectorcuda.h"
 #include "config.h"
 
 namespace alpr