Notes

/*
* @file car_detect.cpp
* @author Abhishek Kumar Annamraju
This code provides faster car detection.
Also for the first time multiple cascade files are used to detect objects,with a benefit that no two objects
are detected twice.
Ever car detected in an image goes through a two stage testing.
The number of checkcascades are set to 1.It is desirable not to change this number.
USAGE: ./car_detect IMAGE.EXTENTION checkcas.xml cas1.xml cas2.xml cas3.xml cas4.xml ..........upto n number of main cascade xml files
ckeckcas.xml is the one trained with smallest size parameters and the rest are the main cascades
*/

#include "opencv2/objdetect/objdetect.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/imgproc/imgproc.hpp"
#include <iostream>
#include <iterator>


using namespace std;
using namespace cv;


void help()
{
cout << endl << "USAGE: ./car_detect IMAGE.EXTENTION checkcas.xml cas1.xml cas2.xml cas3.xml cas4.xml ..........upto n number of main cascade xml files" << endl;
cout << endl << "ckeckcas.xml is the one trained with smallest size parameters and the rest are the main cascades" << endl;
}




class cars //main class
{
public: //variables kept public but precaution taken all over the code

Mat image_input; //main input image
Mat image_main_result; //the final result
Mat storage; //introduced to stop detection of same car more than once

CascadeClassifier cascade; //the main cascade classifier
CascadeClassifier checkcascade; //a test classifier,car detected by both main and test is stated as car

int num;

void getimage(Mat src) //getting the input image
{

if(! src.data )
{
cout << "src not filled" << endl ;
}

else
{
image_input = src.clone();
storage = src.clone(); //initialising storage
image_main_result = src.clone(); //initialising result

cout << "got image" <<endl;
}
}


void cascade_load(string cascade_string) //loading the main cascade
{
cascade.load(cascade_string);

if( !cascade.load(cascade_string) )
{
cout << endl << "Could not load classifier cascade" << endl;

}
else
{

cout << "cascade : " << cascade_string << " loaded" << endl;
}

}


void checkcascade_load(string checkcascade_string) //loading the test/check cascade
{
checkcascade.load(checkcascade_string);

if( !checkcascade.load(checkcascade_string) )
{
cout << endl << "Could not load classifier checkcascade" << endl;

}
else
{
cout<< "checkcascade : " << checkcascade_string << " loaded" << endl;
}
}


void display_input() // function to display input
{
namedWindow("display_input");
imshow("display_input",image_input);
waitKey(0);
}

void display_output() //function to display output
{

if(!image_main_result.empty() )
{
namedWindow("display_output");
imshow("display_output",image_main_result);
waitKey(0);
}
}

void setnum()
{
num = 0;
}


void findcars() //main function
{
int i = 0;

Mat img = storage.clone();
Mat temp; //for region of interest.If a car is detected(after testing) by one classifier,then it will not be available for other one

if(img.empty() )
{
cout << endl << "detect not successful" << endl;
}
int cen_x;
int cen_y;
vector<Rect> cars;
const static Scalar colors[] = { CV_RGB(0,0,255),CV_RGB(0,255,0),CV_RGB(255,0,0),CV_RGB(255,255,0),CV_RGB(255,0,255),CV_RGB(0,255,255),CV_RGB(255,255,255),CV_RGB(128,0,0),CV_RGB(0,128,0),CV_RGB(0,0,128),CV_RGB(128,128,128),CV_RGB(0,0,0)};

Mat gray;

cvtColor( img, gray, CV_BGR2GRAY );

Mat resize_image(cvRound (img.rows), cvRound(img.cols), CV_8UC1 );

resize( gray, resize_image, resize_image.size(), 0, 0, INTER_LINEAR );
equalizeHist( resize_image, resize_image );


cascade.detectMultiScale( resize_image, cars,1.1,2,0,Size(10,10)); //detection using main classifier


for( vector<Rect>::const_iterator main = cars.begin(); main != cars.end(); main++, i++ )
{
Mat resize_image_reg_of_interest;
vector<Rect> nestedcars;
Point center;
Scalar color = colors[i%8];


//getting points for bouding a rectangle over the car detected by main
int x0 = cvRound(main->x);
int y0 = cvRound(main->y);
int x1 = cvRound((main->x + main->width-1));
int y1 = cvRound((main->y + main->height-1));



if( checkcascade.empty() )
continue;
resize_image_reg_of_interest = resize_image(*main);
checkcascade.detectMultiScale( resize_image_reg_of_interest, nestedcars,1.1,2,0,Size(30,30));

for( vector<Rect>::const_iterator sub = nestedcars.begin(); sub != nestedcars.end(); sub++ ) //testing the detected car by main using checkcascade
{
center.x = cvRound((main->x + sub->x + sub->width*0.5)); //getting center points for bouding a circle over the car detected by checkcascade
cen_x = center.x;
center.y = cvRound((main->y + sub->y + sub->height*0.5));
cen_y = center.y;
if(cen_x>(x0+15) && cen_x<(x1-15) && cen_y>(y0+15) && cen_y<(y1-15)) //if centre of bounding circle is inside the rectangle boundary over a threshold the the car is certified
{

rectangle( image_main_result, cvPoint(x0,y0),
cvPoint(x1,y1),
color, 3, 8, 0); //detecting boundary rectangle over the final result



//masking the detected car to detect second car if present

Rect region_of_interest = Rect(x0, y0, x1-x0, y1-y0);
temp = storage(region_of_interest);
temp = Scalar(255,255,255);

num = num+1; //num if number of cars detected

}
}

}


if(image_main_result.empty() )
{
cout << endl << "result storage not successful" << endl;
}

}

};




int main( int argc, const char** argv )
{

double t = 0;
t = (double)cvGetTickCount(); //starting timer
Mat image1 = imread(argv[1],1);
Mat image;
resize(image1,image,Size(300,150),0,0,INTER_LINEAR); //resizing image to get best experimental results
cars detectcars; //creating a object


string checkcas = argv[2];

detectcars.getimage(image); //get the image
detectcars.setnum(); //set number of cars detected as 0
detectcars.checkcascade_load(checkcas); //load the test cascade

//Applying various cascades for a finer search.
if(argc > 3)
{
for(int i = 3;i<argc;i++)
{
string cas = argv[i];
detectcars.cascade_load(cas);
detectcars.findcars();
}
}
else
{
help();
cout << endl << "Please provide atleast one main cascade xml file" << endl;
}

t = (double)cvGetTickCount() - t; //stopping the timer

if(detectcars.num!=0)
{
cout << endl << detectcars.num << " cars got detected in = " << t/((double)cvGetTickFrequency()*1000.) << " ms" << endl << endl;
}
else
{
cout << endl << "cars not found" << endl;
}

detectcars.display_output(); //displaying the final result

return 0;
}