Creating test for CNN object detection component

21 Jun 2016

Design Specifications:

This components has following tasks: firstly, read an image from command line argument, secondly pass the image to CNNobjectdetection component in raw RGB format, thirdly collect the detection results and finally display them on image.

CSDL interface file for the component:

import "/robocomp/interfaces/IDSLs/objectDetection.idsl";

Component testObjectDetectionComp
{
	Communications
	{
		requires objectDetection;
	};
	language Cpp;
	gui Qt(QWidget);
};

Specific worker compute method:

1. run once: read image file from argument and pass to objectdetectionCNN module.

   Mat src=imread(inputFile,1);

2. convert to ColorSeq format (raw RGB)

   int k=0; ColorSeq rgbMatrix; rgbMatrix.resize(src.rows*src.cols); for(int i=0;i<=src.rows-1;i++) { for(int j=0;j<=src.cols-1;j++) { Vec3b intensity = src.at(i,j); rgbMatrix[k].blue = intensity.val[0]; rgbMatrix[k].green = intensity.val[1]; rgbMatrix[k].red = intensity.val[2]; k++; } }

3. call objectdetectioncnn module and get detections.

objectdetectioncnn_proxy->getLabelsFromImage(rgbMatrix, src.rows, src.cols, result);

1. Display detection Bounding boxes.

        for(unsigned int i=0; i < result.size();i++ )
            {  <<result[i].bb.width<<" "<<result[i].bb.height<<endl;
                Rect r1;
                BoundingBox bb;
                bb=result[i].bb;
                r1.x=bb.x;
    				r1.y=bb.y;
    				r1.width=bb.width;
    				r1.height=bb.height;
    				Scalar rect_color = Scalar( 0, 0, 255 );
                rectangle( src, r1.tl(), r1.br(), rect_color, 2, 8, 0 );
                string name=result[i].name;
                putText(src, name.substr(0, name.find(",")), r1.tl(), CV_FONT_HERSHEY_DUPLEX, 0.5, cvScalar(0,0,250), 1, CV_AA);
            }              
    	imshow("Source image", src ); Results: ============     <img src="images/week5/week5_Real1.png" alt="" width="400"/>  <img src="images/week5/week5_Real2.png" alt="" width="400"/>  <img src="images/week5/week5_Real3.png" alt="" width="400"/>