Level Set Systems has developed a new method for the compression of 3D point cloud data that can achieve high compression ratios, with a great accuracy. Although the method is lossy, we consider it a “smart lossy” in the sense that it preserves critical and detailed features, with a predefined error bound. One can specify the maximum error for the extracted point cloud after compression. Even though the extracted point cloud is different than the original, the geometry of the objects represented is maintained.
Our compression was originally created for terrain type data, with fine features such as power lines, telephone poles, cars, and trees. However, any application that uses 3D point cloud data can benefit from our compression. One such application would be in the manufacturing industry, where an older part that has not been digitized can be scanned to create a point cloud representation. This data can then be compressed to a much smaller size, while maintaining an error tolerance defined by the user. The same idea can be applied to medical scans that produce point cloud data. Many point cloud data sets can contain several million points, and even if storage space is not an issue, computation time of software applications using that data can be. By compressing these large point cloud data sets, we can drastically reduce the number of data points while keeping geometric accuracy. The software can then run much faster and still produce accurate results.
Our compression is robust to noisy data and is competitive with respect to the computational time of other existing compression methods. However, we are able to get much higher compression ratios than other methods. The quality of our compression has been independently tested by other research companies.
[Examples of Level Set Compression]
Below we provide some examples of our compression, graphically comparing the original data with the data after our compression/uncompression process. The first sets of data were from the Geo* Challenge posed by the Defense Department’s DARPA. They are LIDAR scans of the city of Ottawa, and consist of 1 airborne scan combined 4 ground based scans. The use of multiple scans captures the great detail of the scene including power lines, traffic lights, tree branches, and cars. GeoTurbo preserves these key features of the scene. The original and compressed/uncompressed files are available for download for potential users to compare the detail preservation of our compression. The file format is LAS.
For downloading, click the following file size of samples.
|
Data Name |
Original LAS file size |
Compressed file size |
Compression Ratio |
|
022 |
67:1 |
||
|
149 |
77:1 |
||
|
237 |
47:1 |
_
_
_
| Data Name : 022 | |
 |
 |
| Original Data. 91.5MB file with 4,801,341 points | Data after Compression/Uncompression. Compressed file size: 1.36MB file with 1,676,149 points. |
 |
 |
| Original Data. 91.5MB file with 4,801,341 points | Data after Compression/Uncompression. Compressed file size: 1.36MB file with 1,676,149 points. |
_
_
_
| Data Name : 149 | |
 |
 |
| Original Data. 74.9MB file with 3,927,902 points | Data after Compression/Uncompression. Compressed file size: 0.97MB file with 1,334,959 points |
 |
 |
| Original Data. 74.9MB file with 3,927,902 points | Data after Compression/Uncompression. Compressed file size: 0.97MB file with 1,334,959 points |
_
_
_
| Data Name : 237 | |
 |
 |
| Original Data. 67.0MB file with 3,516,531 points | Data after Compression/Uncompression. Compressed file size: 1.44MB file with 1,747,042 points |
 |
 |
| Original Data. 67.0MB file with 3,516,531 points | Data after Compression/Uncompression. Compressed file size: 1.44MB file with 1,747,042 points |