CAMEO-SIM has been developed as a physics-based, broadband, scene simulation toolset to enable the quantitative evaluation of both current and future camouflage in visible and infrared wavebands. This is necessary because field trials are expensive, subject to the vagaries of the weather and cannot be used to design new systems. The software development has been undertaken by Hunting Engineering Ltd under contract to DERA for the UK MoD.

The goal of the CAMEO-SIM system is to produce synthetic, high resolution, physically accurate radiance images of target vehicles in operational scenarios, at any wavelength between 0.4 and 14 microns. The software was developed with a scaleable rendering kernel in which imagery can be produced at different fidelities and frame rates depending on the image application and wavelength of operation.

CAMEO-SIM has two key elements that distinguish it from conventional ray-tracing packages. Firstly, there is a complete audit trail between the material properties used by the renderer and the final rendered image. This means that it is possible to conduct carefully controlled parametric manipulations of scene properties, which can be repeated. It allows a scientific investigation of important variables. Secondly because CAMEO-SIM is physics based it is predictive. That is, once a particular scenario has been created it can then be used to predict effects under different conditions such as diurnal variations and weather conditions. The same scene geometry can be used and different material properties assigned for different times of the year. Traditional ray-tracers rely on arbitrary parameter manipulation in order to simulate changes in the environment (e.g. change of atmosphere). CAMEO-SIM on the other hand incorporates these directly as a result of solving the underlying physical equations.

All geometric objects forming the synthetic environment are modelled using textured faceted structures. Texel values in these textures are mapped to real materials, which have measured physical properties, associated with them e.g. bi-directional reflectance, solar absorptivity, conductivity, and density. Each texel is then considered as a mixture of up to three different materials. This means that the physical properties of both soil and grass are modelled when using a ‘grass’ texture. Complex objects are modelled as a number of polygons. This means that the three dimensional effects of trees, including shadow effects can be simulated.

CAMEO-SIM computes the radiance in specified sub-bands for each pixel in the image. These sub band radiance images can then summed to produce an ‘in-band’ radiance image.

The software can display each sub-band as a grey scale image or any three sub-bands as a false colour image. CAMEO-SIM can also display visible band true-colour imagery. This is done by first evaluating the spectral radiance image cube for a user-defined number of sub-bands between 0.38 and 0.78 microns. The spectral image cube is then converted into device independent colour space represented by the tristimulus values X, Y, Z, using the CIE 1931 Colorimetric Standard Observer. The monitor that is used for the image display is calibrated both in terms of luminance and phosphor radiance allowing the X, Y, Z values to be converted to R, G, B values. Various luminance transforms can be employed to make best use of the limited CRT dynamic range.

The scaleable rendering kernel means that images can be rendered at different levels of fidelity. In the lowest fidelity mode all surfaces are diffuse and no shadow effects are modelled. The high fidelity renderer can model directional and bidirectional reflectance effects, geometric occlusion of point and extended sources, and ensures spectral integration of the optical properties with the atmospheric illumination terms to ensure correct radiative transport. As may be expected the higher fidelity images will take longer to render so the level of fidelity required can be selected before rendering the image, so that appropriate imagery can be generated for different applications.

Observer trials

CAMEO-SIM produces a cube of data with the spectral and spatial information. The images produced by CAMEO-SIM can also be displayed on a monitor for observer trials These observer trials are conducted under controlled conditions in the laboratory. The CAMEO-SIM imagery is played back onto a computer display that has been carefully calibrated both colorimetrically and for luminance output in order to provide true-colour imagery to observers (see input from Dr I R Moorhead).

Fidelity issues/validation

CAMEO-SIM has been successfully subjected to range of verification tests. Validation work has included tests where the imagery is compared to measurements made on a simple scene. The scene was created within CAMEO-SIM. However validating a model against the real world is notoriously difficult. All the input parameters must be carefully measured, so that the model has the best chance of producing a solution that is close to that which exists in reality.

One of the difficult problems is assessing what is an appropriate level of fidelity for different tasks, such as target detection. A research programme is investigating different methods of quantifying image fidelity. A range of techniques is being evaluated including models based on human vision performance (see input by Dr T Troscianko and Dr D Tolhurst) and statistical analysis of images using higher order statistics.

In the first year a large data set of imagery of a site on Salisbury Plain in the UK was obtained. Similar imagery was created using CAMEO-SIM, in different wavebands, and rendered at different levels of fidelity eg. using different fidelity trees, with and without textures, and with different shadow options.

Statistical analysis of the images has looked at local effects by comparing a synthetic image with a similar real image, and global effects by comparing image statistics of real and synthetic scenes. The statistics can use different ‘filters’ such as blob and horizontal or vertical bar filters.

Analysis using the different metrics has shown that similar trends are observed using all metrics. However the impact of the ‘global’ fidelity on the ability to detect a target (a ‘local’ effect), has yet to be assessed.

Further work/areas of interest

The main issue is how to evaluate how ‘real’ synthetic imagery is. Methods to quantify the effect of different rendering techniques on image realism must be identified. The impact of human perception on the required fidelity must be assessed so that appropriate imagery can be rendered for different applications. For example; different effects might have to be captured when using imagery in a flight simulator compared with when the imagery is used to assess how easily different targets can be detected in a cluttered environment. Another issue is: how important is movement? Is the overall image fidelity different when the whole image is moving, compared with when an object is moving within the scene?