如何在python中对极坐标系数据进行分组或聚类

Question

I have an RPLidar and my task is to detect an obstacle. From the sensor, I am getting a set of truples which contains the quality of signal, angle and distance in mm.

Here are the example:

[(15, 1.375, 2569.0), (15, 3.0, 2491.5), (15, 4.578125, 2528.75), (15, 6.203125, 2419.75), (15, 7.796875, 2299.75), (15, 9.40625, 2200.5), (15, 11.078125, 2104.25), (15, 12.703125, 2047.25), (15, 14.3125, 2091.75), (15, 15.84375, 2132.25), (15, 17.46875, 2183.75), (15, 19.015625, 2232.5), (15, 20.609375, 2288.75), (15, 22.15625, 2341.0), (15, 23.765625, 2411.0), (15, 25.34375, 2481.0), (15, 26.875, 2554.25), (15, 28.453125, 2641.0), (15, 30.046875, 2731.5), (15, 31.609375, 2824.0), (15, 33.1875, 2854.75), (11, 39.8125, 2178.0), (15, 41.390625, 2278.0), (15, 42.984375, 2199.5), (13, 44.546875, 2269.0), (15, 46.171875, 2265.5), (15, 47.71875, 2300.75), (15, 49.3125, 2454.25), (15, 50.890625, 2442.0), (15, 52.484375, 2415.0), (15, 54.109375, 2404.0), (15, 55.703125, 2390.0), (15, 57.28125, 2371.0), (15, 58.796875, 2357.5), (15, 60.421875, 2348.5), (15, 61.984375, 2344.0), (15, 63.5625, 2340.5), (15, 65.171875, 2332.75), (15, 66.75, 2328.25), (15, 68.34375, 2330.75), (15, 69.984375, 2324.5), (15, 71.515625, 2328.25), (15, 73.125, 2332.75), (15, 74.71875, 2335.75), (15, 76.28125, 2341.5)]

I have plot those data in a matplotlib and that is how they look like:

As you can see i am getting the surroundings information as point and each point has quality,angle,distance from the center. Now i want to group them and need to find the length of the each group. how can i achive this? is there any library available in python by which i can do the clustering like as bellow?

Answer 1

Obvious solution #1:

Project your data from polar coordinates to carthesian coordinates. Then you can use any coordinate-based algorithm such as k-means, and anything that works with Euclidean distance (eg, HAC, DBSCAN)

Obvious solution #2:

Compute the pairwise distances in polar coordinates. That will likely be more complicated and thus slower for large data than above approach, but a bit more obvious. When you have the distance matrix, use any distance-based algorithm such as HAC, DBSCAN, affinity propagation, spectral clustering.

Based on your plots, I'd project and use DBSCAN.

Answer 2

This feels like a K-mean clustering problem to me. You can leverage lots of existing solutions to achieve this. Here is blog post about how to achieve this in Panda.

If the volume is potentially large, I did a project to do k-means clustering in Pyspark a while backe. You can find more info here