Add RELATIVE_ORIENTATION sensor implementation on macOS to //device/generic_sensor

device/generic_sensor/platform_sensor_relative_orientation_using_accelerometer.cc

Index: device/generic_sensor/platform_sensor_relative_orientation_using_accelerometer.cc
diff --git a/device/generic_sensor/platform_sensor_relative_orientation_using_accelerometer.cc b/device/generic_sensor/platform_sensor_relative_orientation_using_accelerometer.cc
new file mode 100644
index 0000000000000000000000000000000000000000..186eeee58238841b333a0409e864cc24ce527596
--- /dev/null
+++ b/device/generic_sensor/platform_sensor_relative_orientation_using_accelerometer.cc
@@ -0,0 +1,164 @@
+// Copyright 2017 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "device/generic_sensor/platform_sensor_relative_orientation_using_accelerometer.h"
+
+#include <stdint.h>
+
+#include <cmath>
+
+#include "base/bind.h"
+#include "base/memory/ptr_util.h"
+#include "device/generic_sensor/generic_sensor_consts.h"
+#include "device/generic_sensor/platform_sensor_provider.h"
+#include "device/generic_sensor/public/cpp/sensor_shared_buffer_reader.h"
+
+namespace {
+
+constexpr double kQuaternionThreshold = 0.01;
+
+bool IsSignificantlyDifferent(const device::SensorReading& reading1,
+                              const device::SensorReading& reading2) {
+  return (std::fabs(reading1.values[0] - reading2.values[0]) >=
+          kQuaternionThreshold) ||
+         (std::fabs(reading1.values[1] - reading2.values[1]) >=
+          kQuaternionThreshold) ||
+         (std::fabs(reading1.values[2] - reading2.values[2]) >=
+          kQuaternionThreshold);
+}
+
+void ComputeQuaternionFromEulerAngles(double alpha,
+                                      double beta,
+                                      double gamma,
+                                      double* x,
+                                      double* y,
+                                      double* z,
+                                      double* w) {
+  double cx = std::cos(beta / 2);
+  double cy = std::cos(gamma / 2);
+  double cz = std::cos(alpha / 2);
+  double sx = std::sin(beta / 2);
+  double sy = std::sin(gamma / 2);
+  double sz = std::sin(alpha / 2);
+
+  *x = sx * cy * cz - cx * sy * sz;
+  *y = cx * sy * cz + sx * cy * sz;
+  *z = cx * cy * sz + sx * sy * cz;
+  *w = cx * cy * cz - sx * sy * sz;
+}
+
+}  // namespace
+
+namespace device {
+
+PlatformSensorRelativeOrientationUsingAccelerometer::
+    PlatformSensorRelativeOrientationUsingAccelerometer(
+        mojo::ScopedSharedBufferMapping mapping,
+        PlatformSensorProvider* provider)
+    : PlatformSensor(mojom::SensorType::RELATIVE_ORIENTATION,

[shalamov, 2017/06/14 07:22:45]

Would it be better to create 'fusion' (PlatformSensorFusionRelativeOrientation)
sensor that will.

1. Get required sensors from platform sensor provider.
2. Based on available senors, create appropriate fusion algorithm instance.
3. Observe sensors for data changes using PlatformSensor::Client
4. Feed data to algorithm and provide fused data to fusion sensor clients.

Relative orientation data can be created (fused) from Accel or Accel + Gyro
combination.

[juncai, 2017/06/19 23:23:25]

Thanks. For this CL, the relative orientation fusion sensor is implemented by
using Accel.

Accel + Gyro relative orientation fusion sensor will be implemented in future
follow-up CLs.

Done.

+                     std::move(mapping),
+                     provider) {
+  provider->CreateSensor(
+      mojom::SensorType::ACCELEROMETER,
+      base::Bind(&PlatformSensorRelativeOrientationUsingAccelerometer::
+                     CreateAccelerometerCallback,
+                 base::Unretained(this)));
+}
+
+PlatformSensorRelativeOrientationUsingAccelerometer::
+    ~PlatformSensorRelativeOrientationUsingAccelerometer() = default;
+
+mojom::ReportingMode
+PlatformSensorRelativeOrientationUsingAccelerometer::GetReportingMode() {
+  return mojom::ReportingMode::ON_CHANGE;

[Reilly Grant (use Gerrit), 2017/06/14 00:59:56]

This should return accelerometer_->GetReportingMode() which means that you need
to delay calling the CreateSensorCallback passed to CreateSensor until
CreateAccelerometerCallback is called.

[juncai, 2017/06/19 23:23:25]

Done.

+}
+
+PlatformSensorConfiguration
+PlatformSensorRelativeOrientationUsingAccelerometer::GetDefaultConfiguration() {
+  return PlatformSensorConfiguration(kDefaultAccelerometerFrequencyHz);
+}
+
+bool PlatformSensorRelativeOrientationUsingAccelerometer::StartSensor(
+    const PlatformSensorConfiguration& configuration) {
+  if (!accelerometer_ ||
+      !accelerometer_->StartSensor(
+          PlatformSensorConfiguration(kDefaultAccelerometerFrequencyHz))) {
+    return false;
+  }
+
+  accelerometer_reader_ = base::MakeUnique<SensorSharedBufferReader>(
+      static_cast<const SensorReadingSharedBuffer*>(
+          accelerometer_->shared_buffer_mapping().get()));
+
+  timer_.Start(

[Reilly Grant (use Gerrit), 2017/06/14 00:59:56]

|accelerometer_| already has a timer and can notify this object through its
client interface when the value changes so we don't need another timer here.

[juncai, 2017/06/19 23:23:25]

Done.

+      FROM_HERE,
+      base::TimeDelta::FromMicroseconds(base::Time::kMicrosecondsPerSecond /
+                                        configuration.frequency()),
+      this, &PlatformSensorRelativeOrientationUsingAccelerometer::PollForData);
+
+  return true;
+}
+
+void PlatformSensorRelativeOrientationUsingAccelerometer::StopSensor() {
+  timer_.Stop();
+  if (accelerometer_)
+    accelerometer_->StopSensor();
+}
+
+bool PlatformSensorRelativeOrientationUsingAccelerometer::
+    CheckSensorConfiguration(const PlatformSensorConfiguration& configuration) {
+  return configuration.frequency() > 0 &&
+         configuration.frequency() <=
+             mojom::SensorConfiguration::kMaxAllowedFrequency;
+}
+
+void PlatformSensorRelativeOrientationUsingAccelerometer::
+    CreateAccelerometerCallback(scoped_refptr<PlatformSensor> accelerometer) {
+  accelerometer_ = accelerometer;
+}
+
+void PlatformSensorRelativeOrientationUsingAccelerometer::PollForData() {

[shalamov, 2017/06/14 07:22:45]

Wouldn't it be better to move all algorithms outside of actual sensor? Then we
will have set of different fusion / filtering algorithms that can be used
whenever needed.

We can use 'Magwick' algorithms for relative orientation (gyro + accel) and 
absolute (magnetometer + accel + gyro) orientation.
http://x-io.co.uk/res/doc/madgwick_internal_report.pdf

[juncai, 2017/06/19 23:23:25]

Done.

+  SensorReading reading;
+  reading.timestamp = (base::TimeTicks::Now() - base::TimeTicks()).InSecondsF();
+
+  if (!accelerometer_reader_->GetSensorReading(&reading))
+    return;
+
+  // Transform the accelerometer values to W3C draft angles.
+  //
+  // Accelerometer values are just dot products of the sensor axes
+  // by the gravity vector 'g' with the result for the z axis inverted.
+  //
+  // To understand this transformation calculate the 3rd row of the z-x-y
+  // Euler angles rotation matrix (because of the 'g' vector, only 3rd row
+  // affects to the result). Note that z-x-y matrix means R = Ry * Rx * Rz.
+  // Then, assume alpha = 0 and you get this:
+  //
+  // x_acc = sin(gamma)
+  // y_acc = - cos(gamma) * sin(beta)
+  // z_acc = cos(beta) * cos(gamma)
+  //
+  // After that the rest is just a bit of trigonometry.
+  //
+  // Also note that alpha can't be provided but it's assumed to be always zero.
+  // This is necessary in order to provide enough information to solve
+  // the equations.
+  //
+  double alpha = 0.0;
+  double beta = std::atan2(-reading.values[1], reading.values[2]);

[shalamov, 2017/06/14 07:22:44]

Do we need to filter (low-pass) accelerometer readings for smoother roll / pitch
estimation? Should this algorithm check when Y is aligned with gravity vector?

[juncai, 2017/06/19 23:23:25]

The main purpose of this CL is to add relative orientation fusion sensor (using
Accel) for Mac, which makes //device/generic_sensor have the same
platform-specific implementation details as //device/sensors for device
orientation event. So for this CL, the implementation is the same as:
https://cs.chromium.org/chromium/src/device/sensors/data_fetcher_shared_memor...

In future follow-up CLs, other fusion algorithm such as Accel + Gyro relative
orientation will be implemented.

+  double gamma = std::asin(reading.values[0] / kMeanGravity);
+
+  ComputeQuaternionFromEulerAngles(
+      alpha, beta, gamma, &reading.values[0].value(),
+      &reading.values[1].value(), &reading.values[2].value(),
+      &reading.values[3].value());
+
+  if (!IsSignificantlyDifferent(reading_, reading))
+    return;
+
+  reading_ = reading;
+  UpdateSensorReading(reading_, true);
+}
+
+}  // namespace device