The core idea is that when rendering interspersed objects on the terrain, before the lighting calculation stage, they not only have their own lighting calculation information but also can obtain the lighting calculation information of the terrain under their feet. Then, the two are mixed, and the mixed values are passed into the lighting calculation to achieve blending.

Based on the accumulation of relevant technologies in the Unity engine, this engine is used for technical development.

The Render Feature function in the Unity Scriptable Render Pipeline is used, which is an interface provided by Unity for developers to directly access the internal logic of the rendering pipeline [5] .

First, Camera Textures is a new concept proposed in this study.

Its theoretical basis comes from Deferred Rendering, which is a solution mainly used to solve the rendering of a large number of lights. Its essence is to summarize the channel information for lighting calculation from the model and save it in multiple Render Targets, and finally perform lighting calculation uniformly.

Similarly, with the help of the concept of deferred rendering, it is easier to understand what Camera Textures does.

That is, obtaining the channel information for lighting calculation from the perspective of the camera from many models in the scene. For PBR metal flow rendering, the cache set should include Depth cache, Albedo (or Base Color) cache, Normal cache, as well as Metallic, Roughness (or Smoothness), and Ambient Occlusion (AO) cache, as shown in the Fragment section of Figure 3 .

In addition, in the rendering process, deferred rendering is to summarize the information channels at the end to calculate the lighting. The behavior of Camera Textures summarizing the information channels needs to be in a more forward position in the rendering process.

And different from deferred rendering, which saves the results in multiple render targets. The medium used to save Camera Textures can be of different types, such as Texture 2D or Graphic Buffer.

Textures have only four channels, namely red, green, blue, and transparency, as shown in Figure 4 .

First, when choosing a texture as the storage medium, although the depth is a floating-point number information and only occupies one channel, it is often drawn into the red channel of a separate render texture in the rendering pipeline, so it is not additionally processed and is separately saved as a depth texture, as shown in Figure 5 .

Second, the albedo is color information without transparency, with a total of 3 channels for red, green, and blue, and after acquisition, it occupies an albedo texture alone, as shown in Figure 6 .

Again, the normal is a three-dimensional vector, that is, XYZ with a total of 3 channels, and is also stored separately as a normal texture. For the convenience of subsequent reading, the normal stored here is in the world space coordinate system, as shown in Figure 7 .

Next, for roughness, metallic, and ambient occlusion, which are three floating-point numbers, they can be separately converged into a Mask Map. The arrangement of the channels is in accordance with the Unity rule standard, with metallic assigned to the red channel, ambient occlusion to the blue channel, the green channel left empty, and roughness in the alpha channel, as shown in Figure 8 .

Finally, after this step is completed, Camera Textures are set to the global variables of the Shader for use in the subsequent interpolation and blending steps, as shown in Figure 9 .

First, there is no limit to the number of channels in the cache because it is itself a collection of data structures. The data in each structure should include the lighting calculation information mentioned above, as shown in Figure 10 .

Second, because the channel information obtained by the camera is itself a Render Texture (RT), using a cache to save it requires an additional step, that is, the transcription from texture to cache, as shown in Figure 11 .

After rendering and storing Camera Textures, the key difficulty lies in how to read the texture information in Camera Textures when rendering the interspersed objects on the terrain and blend it with the PBR channels of the interspersed objects themselves.

First, the matrix method is used, a technology to restore world coordinates from depth. From the code side, the inverse matrix of the main camera’s view projection matrix (Inversed View and Projection Matrix) is set as the global matrix of the Shader.

Unity provides the useful function, it’s shown in Figure 12 .

The effect of restoring world coordinates from depth using the matrix method is shown in Figure 13 .

Thereafter, the depth cache in Camera Textures can be directly transformed into world coordinates. Then, the distance between this world coordinate and the world coordinate of the fragment pixel of the interspersed object itself is subjected to a Smoothstep algorithm to obtain a blending factor with a value range of 0 to 1.

By using the shader graph, the calculation of blending factor is shown in Figure 14 .

The blending factor can be presented in color, as shown in Figure 15 .

With this blending factor, linear interpolation (Lerp) algorithm can be used to blend the albedo, roughness, metallic, and ambient occlusion.

In particular, for normals, special algorithms such as Spherical Interpolation (Slerp) can be used to maintain the characteristic that their length is always 1 [6] .

The slerp function is shown in Figure 16 .

The effect of spherical interpolation is shown in Figure 17 .

By perturbing the blending factor with procedural noise or texturing, different blending effects can be achieved, as shown in Figure 18 and Figure 19 .

The blending effect mentioned above is always based on the perspective of the main camera. With the help of another medium, that is, using a cache to store Camera Textures, different details in the scene can be baked, that is, Camera Textures of the Additional Camera are stored in hardware.

The Directional Additional Camera is responsible for pre-rendering textures containing depth, albedo, normal, roughness, metallic, and ambient occlusion in the position of the scene requiring high details through scripts in the editor development stage, then transcribing them into the cache, and after storing the required matrix of the additional camera at the same time, packaging and storing them in the asset. The asset structure is shown in Figure 20 .

In the rendering of interspersed objects, after blending Camera Textures of the main camera, Camera Textures of the additional camera need to be blended. The sequence is shown in Figure 21 .

Due to the use of cache as the data medium, there is an upper limit to the data size of a single cache. At the same time, considering performance reasons, the upper limit of the number of additional cameras that can be mixed by a single object is tentatively set to 8.

Transform a large cache at once, can make unity editor crashed. The error it gives is shown in Figure 22 .

A table of the test on different cache size, by bake different resolution of additional camera, is shown in Table 1 . Both them baked 8 camera per-object.

The successful mixing effect is shown in Figure 23 .

In the previous text, the additional camera only bakes with a fixed-direction perspective camera. The Omni Additional Camera bakes the full perspective of the upper, lower, left, right, front, and rear six sides at the same time with a 45-degree Field of View (FOV). Inspired by Unity’s Reflection Probe technology [7] , it does not bake the final rendering result but the various channel information used for lighting calculation. The effect of the omni additional camera is shown in Figure 24 .

The omni additional camera uses the channels of the additional camera mentioned above in the rendering stage of interspersed objects. The omni additional camera is essentially six directional special additional cameras, and a single object occupies the number of six additional cameras that can be mixed at the same time.

The omni additional camera bakes code can be shown in Figure 25 .

First, the concept of Camera Textures is proposed, and a multi-channel lighting information blending framework is constructed. Inspired by deferred rendering, for the first time, the core channels required for lighting calculation of terrain and objects, such as depth, albedo, normal, roughness, metallic, and ambient occlusion, are uniformly stored in the view-space texture set to form a standardized lighting information interaction medium. A cross-object lighting information sharing mechanism is constructed, so that the material properties of the terrain and objects in the junction area, such as roughness differences and normal direction mutations, can be quantitatively calculated and mixed using blending factors, laying a data foundation for “physically real” blending.

Second, the dynamic calculation method of blending factors based on matrix transformation realizes the intelligent connection of geometry and materials. By inferring world coordinates from depth textures and combining the distance metric of the object’s own fragment coordinates, adaptive blending weights are generated to replace traditional manual parameter adjustment or fixed-function mixing. It does not require pre-baked masks or rely on the engine’s built-in layer system, can respond in real time to dynamic interactions (such as object movement and terrain deformation), and automatically generates blending boundaries that conform to the geometric distance attenuation law, significantly reducing manual editing costs.

Again, the “additional camera” and “omni additional camera” systems solve the problem of blending accuracy for extreme perspectives and complex details. A solution to balance cache data size and performance is proposed. A single object can be mixed with up to 8 additional cameras, breaking through the traditional single-viewpoint dependence and providing a universal solution for the omni-directional seamless connection between dynamically rotating objects, such as rotatable mechanical devices, and terrain.

Then, while the research has made the above progress, there are still some areas for improvement. For example, the upper limit of the number of additional cameras that can be mixed by a single object is still 8. It is believed that with the development of hardware devices and further iterations, the problem of cache transmission scale will also be solved.

In summary, this study focuses on the core problem of transitional blending between terrain and objects. Through interdisciplinary technology integration and algorithm innovation, the universality and portability of terrain blending technology are achieved. It has made multiple contributions at the theoretical method, technical implementation, and engineering application levels.

A performance comparison is provided. All results are based on the following configurations.

13th Gen Intel(R) Core(TM) i9-13900HX

NVIDIA GeForce RTX-4070 Laptop GPU

RAM 64.0GB

The test scene is shown in Figure 26 .

First, a table of the new technologies performance, includes platform in Unity Editor and on Windows Standalone, is shown in Table 2 .

Second, the test scene’ Statistics in Unity Editor is shown in Figure 27 .

Third, the test scene’s performance on Windows Standalone is shown in Figure 28 .