Ouster recently announced the new REV8 OS product family, billed as the ‘World’s First Color Lidar’. This might not sound like a significant advancement, as “colorized” lidar data has been available for over twenty years, fusing aerial camera and lidar data collected from the same flight. If you dig deeper, though, true-color lidar stands out as a significant development. Let’s look further.
Intensity (monochromatic return signal amplitude data) was available starting in the late 1990s on airborne lidar systems. That allowed users to distinguish between two surfaces at the same distance or elevation, because the different surfaces had different reflectivity at the laser wavelength. An airport runway was therefore distinguishable from the grass around it and the stripes on it, by virtue of their different reflectivity, even though they would all be at a similar elevation.
Later on, sensor manufacturers developed software that could associate the color of a particular pixel in a camera image with a particular lidar data point. The result: colorized lidar data. This became useful in creating more photorealistic lidar data and, as lidar point densities increased, resulted in impressive data sets.
The challenge with colorization techniques lies in a couple of design limitations. First, the imaging sensor collects large images at relatively slow intervals (seconds). Lidar collects a series of individual data points (or a number of parallel lines) at high rates (think MHz). This results in an inherent time mismatch between the collection of lidar points and the associated camera pixels. Think about a stopped vehicle in the camera data, but not in the associated lidar data. Hence, road-textured cars and car-colored roads. Second, clouds, shadows, or nightfall limit the chromatic accuracy of lidar data colorized by camera images.
The first challenge above can be addressed using what might be called a “single pixel imager”, where a single color pixel is collected each time the laser fires. However, this still does not solve the second challenge, dealing with lighting variations.
“Color lidar” solves both issues. Although multi-spectral lidar systems have been available for about a decade, these systems tend to use lasers at readily-available wavelengths, such as 532nm (green), 1064nm (infrared) and 1540nm (infrared). Not entirely within human perception, colorization based on current multi-spectral lidar is not generally photorealistic. That said, a lidar system using three wavelengths, all within the visible spectrum, has the potential to deliver accurate color rendering. As an active sensor, the color data derived from the three wavelengths will be consistent, regardless of illumination. This means equally accurate color data, day or night, greatly simplifying the task of classifying data based on color.
Ouster’s new multicolor design therefore allows distinction between different surfaces at the same distance and having the same overall reflectivity, so long as there is some difference in at least one color band. A green street sign and a red stop sign would look distinctly different in a true-color lidar, even at the same range and having the same overall reflectivity. In addition, those same road signs will be consistent whether the data was collected in daylight or at night, in direct sunlight or in shade.
So, a true multicolor lidar solves lots of issues, particularly in the areas of alignment and illumination. It also offers a potential boost in maximum-range performance. Maximum range is typically limited by signal-to-noise ratio (SNR) and improves as the square root of the laser output. Combining the return signal of three lidar channels results in a 73 percent increase in maximum range, close to Ouster’s claims of doubled maximum range.
Whether used for mapping, vehicle guidance or machine-vision applications, this new technique offers lots of benefits. Perception sensing has made another step forward.