add motion blur effect

[BudgiePanic]

section in chapter 17 of the book

[BudgiePanic]

working on this looks like it needs the following:

• a new type of shape: the motion shape
  • motion shape wraps around a normal shape and has a velocity vector in units per second
  • motion shape has a bounding box that covers the entire area the shape will move through during the exposure
  • the internal shape is transformed based on the time value of the ray and its velocity vector before intersection test is performed
• a new type of camera: a shutter camera
  • the shutter camera has an exposure time
  • The shutter camera casts multiple rays per pixel, one for each time value during the exposure
  • the color from each ray is averaged before being written to the image canvas
  • if a motion shape is encountered, transform the shape's position based on its velocity and the time value of the ray before performing the intersection test
  • shutter camera should be able to defer ray casts to an internal camera (pinhole or focus camera)? to reduce code duplication

[BudgiePanic]

exposure camera needs to convey more information to motion shapes so they can bake their AABBs

• the duration of exposure
• it will be assumed that time = 0s always when the camera exposure starts
• this is for an optimization step

[BudgiePanic]

following the book's advice on this feature, we need to extend the Ray record to include some more information:

• the time the ray was cast be default this will be 0s but the shutter camera will supply the time value

[BudgiePanic]

tests to create

• motion shape cube
  • cube moves left away from the origin, cast rays at various times, check the ray misses after the cube moved enough
• motion shape plane
  • plane moves backwards, further away from ray origin, check that the distance to intersection is increasing as the object moves further away
• shadows
  • check that the intensity of illumination at a point changes as the motion object moves out of the way
  • the world will need an overload of the isOcculded method to pass in the time that the occlusion check takes place

[BudgiePanic]

created tests for motion shape cube and plane, now working on creating the shutter camera

I believe world will now need method overloads where the time the color is computed / time the occlusion is checked for, must be passed in as a parameter

[BudgiePanic]

To make this work, I think I need to change the perspective camera API, they need to expose the code they use to generate the color for one pixel, so the shutter camera can call it multiple times with different time stamps

[BudgiePanic]

These camera's have a code smell in my opinion, methods across various subclasses sharing the exact same structure The only camera that can't have its takePicture logic extracted to a consumer lambda is depthCamera, but that's because the depth camera is doing a post processing step as part of its takePicture method, which could easily be removed to another method

[BudgiePanic]

you could almost make a single camera class with behaviour delegates for ray generation, pixel imaging, and whole imaging

[BudgiePanic]

tomorrow we need to modify world functionality to use the new time field in rays

[BudgiePanic]

something is up with ray-shape intersection testing when they are nested in a linear motion shape

[BudgiePanic]

motion blur is not appearing in reflective surfaces either

[BudgiePanic]

motion blur is not appearing in reflective surfaces either

fixed now, just had to pass in the time, which is now stored in shading info object

[BudgiePanic]

something is up with ray-shape intersection testing when they are nested in a linear motion shape

I think it is because we are currently transforming the ray by negative velocity in local object space instead of in world space

edit: was actually because not calculating AABB correctly for motion shapes

[BudgiePanic]

alright now that that is working, last thing we need to do is make the shutter camera bake the scene with exposure end time

[BudgiePanic]

future issue to deal with hmm, baking the motion shapes with the camera's exposure end time to optimize their AABBs is going to create an implicit coupling between the camera and the world being imaged, for example, you wouldn't be able to have two shutter cameras imaging the same world at different times

we could modify motion shapes so they have to be provided a motion end time at construction time?

we could use a design pattern to have all motion shapes in a world listen for time end events, and update AABBs when a shutter camera announces their exposure duration or something

[BudgiePanic]

going to change the motion shape so if the end time is unknown, then we don't use the AABB's bounding box this will allow correct behaviour of the program for rays cast at time != 0 for motion shapes even when their end time is not set

going to make shutter camera baking the exposure duration a higher level function so the user is causing that behaviour, rather than the camera automatically baking the times when Camera::imageWorld is called

[BudgiePanic]

looks like everything is working