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

use derive_more::{AsMut, AsRef, Constructor, Deref, DerefMut, From, Into};
use nalgebra::{Matrix3, Point2, Point3, Vector2};

/// A point on an image frame. This type should only be used when
/// the point location is on the image frame in pixel coordinates.
/// This means the keypoint is neither undistorted nor normalized.
#[derive(
    Debug,
    Clone,
    Copy,
    PartialEq,
    PartialOrd,
    AsMut,
    AsRef,
    Constructor,
    Deref,
    DerefMut,
    From,
    Into,
)]
pub struct ImageKeyPoint(pub Point2<f32>);

/// A point in normalized image coordinates. This keypoint has been corrected
/// for distortion and normalized based on the camrea intrinsic matrix.
/// Please note that the intrinsic matrix accounts for the natural focal length
/// and any magnification to the image. Ultimately, the key points must be
/// represented by their position on the camera sensor and normalized to the
/// focal length of the camera.
#[derive(
    Debug,
    Clone,
    Copy,
    PartialEq,
    PartialOrd,
    AsMut,
    AsRef,
    Constructor,
    Deref,
    DerefMut,
    From,
    Into,
)]
pub struct NormalizedKeyPoint(pub Point2<f32>);

impl NormalizedKeyPoint {
    pub fn with_depth(self, depth: f32) -> CameraPoint {
        CameraPoint((depth * self.coords).push(depth).into())
    }
}

impl From<CameraPoint> for NormalizedKeyPoint {
    fn from(camera: CameraPoint) -> Self {
        NormalizedKeyPoint(camera.xy() / camera.z)
    }
}

/// A 3d point in camera coordinates (relative to camera).
/// If the point is divided by the `z` component (projected onto a plane at `z` = 1),
/// then the `x` and `y` components form a `NormalizedKeypoint`. This is because a point
/// at that location would appear on the camera at that location.
#[derive(
    Debug,
    Clone,
    Copy,
    PartialEq,
    PartialOrd,
    AsMut,
    AsRef,
    Constructor,
    Deref,
    DerefMut,
    From,
    Into,
)]
pub struct CameraPoint(pub Point3<f32>);

/// This contains intrinsic camera parameters as per
/// [this Wikipedia page](https://en.wikipedia.org/wiki/Camera_resectioning#Intrinsic_parameters).
///
/// For a high quality camera, this may be sufficient to normalize image coordinates.
/// Undistortion may also be necessary to normalize image coordinates.
#[derive(Debug, Clone, Copy, PartialEq, PartialOrd, Constructor)]
pub struct CameraIntrinsics {
    pub focals: Vector2<f32>,
    pub principal_point: Point2<f32>,
    pub skew: f32,
}

impl CameraIntrinsics {
    /// Creates camera intrinsics that would create an identity intrinsic matrix.
    /// This would imply that the pixel positions have an origin at `0,0`,
    /// the pixel distance unit is the focal length, pixels are square,
    /// and there is no skew.
    pub fn identity() -> Self {
        Self {
            focals: Vector2::new(1.0, 1.0),
            skew: 0.0,
            principal_point: Point2::new(0.0, 0.0),
        }
    }

    pub fn focals(self, focals: Vector2<f32>) -> Self {
        Self { focals, ..self }
    }

    pub fn focal(self, focal: f32) -> Self {
        Self {
            focals: Vector2::new(focal, focal),
            ..self
        }
    }

    pub fn principal_point(self, principal_point: Point2<f32>) -> Self {
        Self {
            principal_point,
            ..self
        }
    }

    pub fn skew(self, skew: f32) -> Self {
        Self { skew, ..self }
    }

    #[rustfmt::skip]
    pub fn matrix(&self) -> Matrix3<f32> {
        Matrix3::new(
            self.focals.x,  self.skew,      self.principal_point.x,
            0.0,            self.focals.y,  self.principal_point.y,
            0.0,            0.0,            1.0,
        )
    }
}

/// This contains basic camera specifications that one could find on a
/// manufacturer's website. This only contains parameters that cannot
/// be changed about a camera. The focal length is not included since
/// that can typically be changed and images can also be magnified.
///
/// All distance units should be in meters to avoid conversion issues.
#[derive(Debug, Clone, Copy, PartialEq, PartialOrd, Constructor)]
pub struct CameraSpecification {
    pub pixels: Vector2<usize>,
    pub pixel_dimensions: Vector2<f32>,
}

impl CameraSpecification {
    /// Creates a [`CameraSpecification`] using the sensor dimensions.
    pub fn from_sensor(pixels: Vector2<usize>, sensor_dimensions: Vector2<f32>) -> Self {
        Self {
            pixels,
            pixel_dimensions: Vector2::new(
                sensor_dimensions.x / pixels.x as f32,
                sensor_dimensions.y / pixels.y as f32,
            ),
        }
    }

    /// Creates a [`CameraSpecification`] using the sensor width assuming a square pixel.
    pub fn from_sensor_square(pixels: Vector2<usize>, sensor_width: f32) -> Self {
        let pixel_width = sensor_width / pixels.x as f32;
        Self {
            pixels,
            pixel_dimensions: Vector2::new(pixel_width, pixel_width),
        }
    }

    /// Combines the [`CameraSpecification`] with a focal length to create a [`CameraIntrinsics`].
    ///
    /// This assumes square pixels and a perfectly centered principal point.
    pub fn intrinsics_centered(&self, focal: f32) -> CameraIntrinsics {
        CameraIntrinsics::identity()
            .focal(focal)
            .principal_point(self.pixel_dimensions.map(|p| p as f32 / 2.0 - 0.5).into())
    }
}