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
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
|
/* Copyright (c) 2021 Michael Welter <michael@welter-4d.de>
*
* 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.
*/
#pragma once
#include "ui_neuralnet-trackercontrols.h"
#include "model_adapters.h"
#include "deadzone_filter.h"
#include "preview.h"
#include "options/options.hpp"
#include "api/plugin-api.hpp"
#include "cv/video-widget.hpp"
#include "cv/translation-calibrator.hpp"
#include "cv/numeric.hpp"
#include "compat/timer.hpp"
#include "video/camera.hpp"
#include "cv/affine.hpp"
#include <QObject>
#include <QThread>
#include <QMutex>
#include <QHBoxLayout>
#include <QDialog>
#include <QTimer>
#include <memory>
#include <cinttypes>
#include <array>
#include <opencv2/core.hpp>
#include <opencv2/imgproc.hpp>
namespace neuralnet_tracker_ns
{
using namespace options;
enum fps_choices
{
fps_default = 0,
fps_30 = 1,
fps_60 = 2,
fps_MAX = 3
};
struct resolution_tuple
{
int width;
int height;
};
static const std::array<resolution_tuple, 7> resolution_choices =
{{
{ 320, 240 },
{ 640, 480 },
{ 800, 600 },
{ 1024, 768 },
{ 1280, 720 },
{ 1920, 1080},
{ 0, 0 }
}};
struct Settings : opts {
value<int> offset_fwd { b, "offset-fwd", 200 }, // Millimeters
offset_up { b, "offset-up", 0 },
offset_right { b, "offset-right", 0 };
value<QString> camera_name { b, "camera-name", ""};
value<int> fov { b, "field-of-view", 56 };
value<fps_choices> force_fps { b, "force-fps", fps_default };
value<bool> show_network_input { b, "show-network-input", false };
value<double> roi_filter_alpha{ b, "roi-filter-alpha", 1. };
value<double> roi_zoom{ b, "roi-zoom", 1. };
value<bool> use_mjpeg { b, "use-mjpeg", false };
value<int> num_threads { b, "num-threads", 1 };
value<int> resolution { b, "force-resolution", 0 };
value<double> deadzone_size { b, "deadzone-size", 1. };
value<double> deadzone_hardness { b, "deadzone-hardness", 1.5 };
value<QString> posenet_file { b, "posenet-file", "head-pose.onnx" };
Settings();
};
struct CamIntrinsics
{
float focal_length_w;
float focal_length_h;
float fov_w;
float fov_h;
};
class NeuralNetTracker : protected virtual QThread, public ITracker
{
Q_OBJECT
public:
NeuralNetTracker();
~NeuralNetTracker() override;
module_status start_tracker(QFrame* frame) override;
void data(double *data) override;
void run() override;
Affine pose();
std::tuple<cv::Size, double, double> stats() const;
QMutex camera_mtx_;
std::unique_ptr<video::impl::camera> camera_;
private:
bool detect();
bool open_camera();
void set_intrinsics();
cv::Mat prepare_input_image(const video::frame& frame);
bool load_and_initialize_model();
void draw_gizmos(
const std::optional<PoseEstimator::Face> &face,
const Affine& pose);
void update_fps(double dt);
// Secretly applies filtering while computing the pose in 3d space.
QuatPose compute_filtered_pose(const PoseEstimator::Face &face);
// Compute the pose in 3d space taking the network outputs
QuatPose transform_to_world_pose(const cv::Quatf &face_rotation, const cv::Point2f& face_xy, const float face_size) const;
QString get_posenet_filename() const;
Settings settings_;
std::optional<Localizer> localizer_;
std::optional<PoseEstimator> poseestimator_;
Ort::Env env_{nullptr};
Ort::MemoryInfo allocator_info_{nullptr};
CamIntrinsics intrinsics_{};
cv::Mat grayscale_;
std::array<cv::Mat,2> downsized_original_images_ = {}; // Image pyramid
std::optional<cv::Rect2f> last_localizer_roi_;
std::optional<cv::Rect2f> last_roi_;
static constexpr float HEAD_SIZE_MM = 200.f; // In the vertical. Approximately.
mutable QMutex stats_mtx_;
double fps_ = 0;
double inference_time_ = 0;
cv::Size resolution_ = {};
static constexpr double RC = .25;
int num_threads_ = 1;
bool is_visible_ = true;
QMutex mtx_ = {}; // Protects the pose
std::optional<QuatPose> last_pose_ = {};
Affine last_pose_affine_ = {};
Preview preview_;
std::unique_ptr<cv_video_widget> video_widget_;
std::unique_ptr<QHBoxLayout> layout_;
};
class NeuralNetDialog : public ITrackerDialog
{
Q_OBJECT
public:
NeuralNetDialog();
void register_tracker(ITracker * x) override;
void unregister_tracker() override;
bool embeddable() noexcept override;
void set_buttons_visible(bool x) override;
private:
void make_fps_combobox();
void make_resolution_combobox();
Ui::Form ui_;
Settings settings_;
// Calibration code mostly taken from point tracker
QTimer calib_timer_;
TranslationCalibrator trans_calib_;
QMutex calibrator_mutex_;
QTimer tracker_status_poll_timer_;
NeuralNetTracker* tracker_ = nullptr;
private Q_SLOTS:
void save() override;
void reload() override;
void doOK();
void doCancel();
void camera_settings();
void update_camera_settings_state(const QString& name);
void startstop_trans_calib(bool start);
void trans_calib_step();
void status_poll();
void onSelectPoseNetFile();
};
class NeuralNetMetadata : public Metadata
{
Q_OBJECT
QString name() override { return QString("neuralnet tracker"); }
QIcon icon() override { return QIcon(":/images/neuralnet.png"); }
};
} // neuralnet_tracker_ns
namespace neuralnet_tracker_tests
{
void run();
}
using neuralnet_tracker_ns::NeuralNetTracker;
using neuralnet_tracker_ns::NeuralNetDialog;
using neuralnet_tracker_ns::NeuralNetMetadata;
|
