summaryrefslogtreecommitdiffhomepage
path: root/tracker-easy/point-extractor.cpp
blob: 037d06c05fefd76b548c837134602698d4db0eb2 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
/* 
 * Copyright (c) 2019 Stephane Lenclud
 *
 * Permission to use, copy, modify, and/or distribute this software for any
 * purpose with or without fee is hereby granted, provided that the above
 * copyright notice and this permission notice appear in all copies.
 */

#include "point-extractor.h"
#include "preview.h"
#include "tracker-easy.h"

#include "cv/numeric.hpp"
#include "compat/math.hpp"
#include <opencv2/imgproc/types_c.h>

#include <cmath>
#include <algorithm>
#include <cinttypes>
#include <memory>

#include <QDebug>

using namespace numeric_types;

namespace EasyTracker
{

    PointExtractor::PointExtractor() : s(KModuleName)
    {
    
    }


    void PointExtractor::extract_points(const cv::Mat& aFrame, cv::Mat* aPreview, std::vector<vec2>& aPoints)
    {
        //TODO: Assert if channel size is neither one nor two
        // Make sure our frame channel is 8 bit
        size_t channelSize = aFrame.elemSize1();
        if (channelSize == 2)
        {
            // We have a 16 bits single channel. Typically coming from Kinect V2 IR sensor
            // Resample to 8-bits
            double min = std::numeric_limits<uint16_t>::min();
            double max = std::numeric_limits<uint16_t>::max();
            //cv::minMaxLoc(raw, &min, &max); // Should we use 16bit min and max instead?
            // For scalling to have more precission in the range we are interrested in
            min = max - 255;
            // See: https://stackoverflow.com/questions/14539498/change-type-of-mat-object-from-cv-32f-to-cv-8u/14539652
            aFrame.convertTo(iFrameChannelSizeOne, CV_8U, 255.0 / (max - min), -255.0*min / (max - min));
        }
        else
        {
            iFrameChannelSizeOne = aFrame;
        }


        // Make sure our frame has a single channel
        // Make an extra copy if needed
        const int channelCount = iFrameChannelSizeOne.channels();
        if (channelCount == 3)
        {
            // Convert to grayscale
            // TODO: What's our input format, BRG or RGB?
            // That won't make our point extraction work but at least it won't crash
            cv::cvtColor(iFrameChannelSizeOne, iFrameGray, cv::COLOR_BGR2GRAY);
            // TODO: Instead convert to HSV and use a key color together with cv::inRange to sport the color we want.
            // Key color should be defined in settings.
        }
        else if (channelCount == 1)
        {
            // No further convertion needed
            iFrameGray = iFrameChannelSizeOne;
        }
        else
        {
            eval_once(qDebug() << "tracker/easy: camera frame depth not supported" << aFrame.channels());
            return;
        }

        // Contours detection
        std::vector<std::vector<cv::Point> > contours;
        cv::findContours(iFrameGray, contours, CV_RETR_EXTERNAL, CV_CHAIN_APPROX_SIMPLE);
    
        // Workout which countours are valid points
        for (size_t i = 0; i < contours.size(); i++)
        {
            if (aPreview)
            {
                cv::drawContours(*aPreview, contours, (int)i, CV_RGB(255, 0, 0), 2);
            }
        

            cv::Rect bBox;
            bBox = cv::boundingRect(contours[i]);
       
            // Make sure bounding box matches our criteria
            if (bBox.width >= s.min_point_size
                && bBox.height >= s.min_point_size
                && bBox.width <= s.max_point_size
                && bBox.height <= s.max_point_size)
            {
                vec2 center;
                center[0] = bBox.x + bBox.width / 2;
                center[1] = bBox.y + bBox.height / 2;
                aPoints.push_back(vec2(center));

                if (aPreview)
                {
                    cv::rectangle(*aPreview, bBox, CV_RGB(0, 255, 0), 2);
                }
            }
        }

        // Keep only the three points which are highest, i.e. with lowest Y coordinates
        // That's most usefull to discard noise from features below your cap/head.
        // Typically noise comming from zippers and metal parts on your clothing.
        // With a cap tracker it also successfully discards noise from glasses.
        // However it may not work as good with a clip user wearing glasses.
        while (aPoints.size() > 3) // Until we have no more than three points
        {
            int maxY = 0;
            size_t index = -1;

            // Search for the point with highest Y coordinate
            for (size_t i = 0; i < aPoints.size(); i++)
            {
                if (aPoints[i][1] > maxY)
                {
                    maxY = aPoints[i][1];
                    index = i;
                }
            }

            // Discard it
            aPoints.erase(aPoints.begin() + index);
        }
    }

}