Updated manifest.xml (because Mac botched the job). Added singular bo…

…unding boxes to Candidate

include/Candidate.hpp

@@ -39,6 +39,7 @@
 #ifndef CANDIDATE_HPP_
 #define CANDIDATE_HPP_
 #include <algorithm>
+#include <iostream>
 #include <limits>
 #include <opencv2/core/core.hpp>
 #include "types.hpp"
@@ -85,6 +86,45 @@ class Candidate {
 	static void sort(vectorCandidate& candidates) {
 		std::sort(candidates.begin(), candidates.end(), descending);
 	}
+
+	/*! @brief create a single bounding box around the detection taken from the part limit
+	 *
+	 * @return a single bounding Rect
+	 */
+	cv::Rect boundingBox(void) const {
+		int minx =  std::numeric_limits<int>::max();
+		int miny =  std::numeric_limits<int>::max();
+		int maxx = -std::numeric_limits<int>::max();
+		int maxy = -std::numeric_limits<int>::max();
+		const int nparts = parts_.size();
+		for (int n = 0; n < nparts; ++n) {
+			const cv::Rect p = parts_[n];
+			if (p.x + p.width/2  < minx) minx = p.x + p.width/2;
+			if (p.x + p.width/2  > maxx) maxx = p.x + p.width/2;
+			if (p.y + p.height/2 < miny) miny = p.y + p.height/2;
+			if (p.y + p.height/2 > maxy) maxy = p.y + p.height/2;
+		}
+		return cv::Rect(minx, miny, maxx-minx, maxy-miny);
+	}
+
+	/*! @brief create a single bounding box around the detection from mean and standard deviation
+	 *
+	 * @return
+	 */
+	cv::Rect boundingBoxNorm(void) const {
+		const int nparts = parts_.size();
+		cv::Mat_<int> xpts(cv::Size(1,nparts));
+		cv::Mat_<int> ypts(cv::Size(1,nparts));
+		for (int n = 0; n < nparts; ++n) {
+			const cv::Point centroid = (parts_[n].tl() + parts_[n].br())*0.5;
+			xpts(n) = centroid.x;
+			ypts(n) = centroid.y;
+		}
+		cv::Scalar xmean, ymean, xstd, ystd;
+		cv::meanStdDev(xpts, xmean, xstd);
+		cv::meanStdDev(ypts, ymean, ystd);
+		return cv::Rect(xmean(0)-1.5*xstd(0), ymean(0)-1.5*ystd(0), 3*xstd(0), 3*ystd(0));
+	}
 };
 
 #endif /* CANDIDATE_HPP_ */

manifest.xml

@@ -1,13 +1,14 @@
-<manifest>
-  <description brief="thing">
-    THING
+<package>
+  <description brief="Parts-based object recognition">
+    This package provides an implementation of parts-based object recognition
+    using mixtures of deformable parts, inspired by the works of Deva Ramanan.
   </description>
-  <author>Not me!</author>
-  <license>Driver's</license>
-  <review status="unreviewed" notes="Yeah, right."/>
+  <author>Hilton Bristow</author>
+  <license>BSD</license>
+  <review status="unreviewed" notes=""/>
 
   <export>
     <cpp cflags="-I${prefix}/include"
          lflags="-L${prefix}/lib -lobject_recognition_by_parts"/>
   </export>
-</manifest>
+</package>

src/Visualize.cpp

@@ -85,6 +85,7 @@ void Visualize::candidates(const Mat& im, const vectorCandidate& candidates, int
 			rectangle(canvas, box, colors[p], LINE_THICKNESS);
 			if (display_confidence) putText(canvas, confidence, Point(box.x, box.y-5), FONT_HERSHEY_SIMPLEX, 0.5f, black, 2);
 		}
+		rectangle(canvas, candidate.boundingBoxNorm(), Scalar(255, 0, 0), LINE_THICKNESS);
 	}
 }
 

src/main.cpp

@@ -92,6 +92,7 @@ int main(int argc, char** argv) {
 	Mat im = imread(argv[2]);
 	if (argc == 4) {
 		depth = imread(argv[3], CV_LOAD_IMAGE_ANYDEPTH);
+		// convert the depth image from mm to m
 		depth = depth / 1000.0f;
 	}
 
@@ -107,7 +108,7 @@ int main(int argc, char** argv) {
 	if (candidates.size() > 0) {
         Mat canvas;
 		Candidate::sort(candidates);
-		visualize.candidates(im, candidates, canvas, true);
+		visualize.candidates(im, candidates, 1, canvas, true);
         visualize.image(canvas);
 		waitKey();
 	}