OpenCV KAKUUCHI

EdgeBoxes

Gallery

src 元動画

halfway エッジ抽出

dst 検出

what process?

エッジを使って物体を検出し、バウンディングボックスを生成する。

比較的検出が速いのでこの処理手法を使ってバウンディングボックスで物体があるかを特定した後に、その物体が何かをカスケードを使って分類の計算すれば、検出率は良さそうです。

* model.yml は下記のリンクからダウンロードして実行時に読み込んでください。
https://github.com/opencv/opencv_extra/blob/master/testdata/cv/ximgproc/model.yml.gz

cook code C++

#include <iostream>
#include <opencv2/core.hpp>
#include <opencv2/core/utility.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/videoio.hpp>
#include <opencv2/ximgproc.hpp>
#include <opencv2/ximgproc/edgeboxes.hpp>

int main(int argc, const char* argv[])
{
    // ./out -src=[image path]
    cv::String keys = "{src||}";
    cv::CommandLineParser parser(argc, argv, keys);
    cv::String src_path = parser.get<cv::String>("src");

cv::VideoCapture capture(src_path);
    if (!capture.isOpened()){
        //error in opening the video input
        std::cerr << "Unable to open file!" << std::endl;
        return 0;
    }

/* 初期設定 */
    int cap_width = int(capture.get(cv::CAP_PROP_FRAME_WIDTH));
    int cap_height = int(capture.get(cv::CAP_PROP_FRAME_HEIGHT));
    int fps = int(capture.get(cv::CAP_PROP_FPS));
    int fourcc = cv::VideoWriter::fourcc('m','p','4','v');
    cv::Size video_size = cv::Size(cap_width,cap_height);

cv::VideoWriter writer("output.mp4", fourcc, fps, video_size);

/* read model */
    std::string modelFileName = "model.yml";
    cv::Ptr<cv::ximgproc::StructuredEdgeDetection> obj 
        = cv::ximgproc::createStructuredEdgeDetection(modelFileName);

while(true){
        cv::Mat frame, mat_32f;
        cv::Mat edges, orimap, edges_nms, scores, _scores;
        std::vector<cv::Rect> boxes;

capture >> frame;
        if (frame.empty())
            break;

frame.convertTo(mat_32f, CV_32F, 1/255.0);
        obj->detectEdges(mat_32f, edges);

obj->computeOrientation(edges, orimap);
        obj->edgesNms(edges, orimap, edges_nms);

cv::Ptr<cv::ximgproc::EdgeBoxes> boxobj =  cv::ximgproc::createEdgeBoxes();
        boxobj->setMaxBoxes(30);

boxobj->getBoundingBoxes(edges_nms, orimap, boxes, scores);

if (0 < boxes.size()){
            for(int i = 0; i < boxes.size();i++){
                cv::rectangle(frame, boxes.at(i), cv::Scalar(0,255,0), 1, cv::LINE_AA);
            }
        }
        writer << frame;
    }
    
    capture.release();
    writer.release();

return 0;
}

cook code Python

import numpy as np
import cv2
import argparse

def main():

parser = argparse.ArgumentParser()
    parser.add_argument('--src')
    args = parser.parse_args()

cap = cv2.VideoCapture(args.src)

cap_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
    cap_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    fps = cap.get(cv2.CAP_PROP_FPS)

fmt = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')
    writer = cv2.VideoWriter('output.mp4', fmt, fps, (cap_width,cap_height))

#read model
    model='model.yml'
    obj = cv2.ximgproc.createStructuredEdgeDetection(model)
    boxobj = cv2.ximgproc.createEdgeBoxes()
    boxobj.setMaxBoxes(30)

while cap.isOpened():
        _ret, img = cap.read()
        
        if img is None:
            break

imgrgb=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)/255
        imgrgb=imgrgb.astype(np.float32)
        edges=obj.detectEdges(imgrgb)

orimap = obj.computeOrientation(edges)
        edges_nms = obj.edgesNms(edges, orimap)

boxes, scores = boxobj.getBoundingBoxes(edges_nms, orimap)

if 0 < boxes.shape[0]:

for rect in boxes:
                cv2.rectangle(img,rect,(0,255,0),1,cv2.LINE_AA)

writer.write(img)

cap.release()
    writer.release()

if __name__ == '__main__':
    main()

Reference

thesis : C. Lawrence Zitnick and Piotr Dollár. Edge boxes: Locating object proposals from edges. In ECCV, 2014.
[Pixabay:Photographer]Coverr-Free-Footage
opencv-contribのedge-boxesを試す
OpenCVdoc - edgeboxes_demo.py

戻る(Back Page)