nihalsoans91 / phantom_obstacles

Removes phantom obstacles from maps created from robot point clouds

Remove Phantom obsticles from point cloud maps ( Python )

Nihal Solomon Soans

[email protected]

We solve the problem of removing obsticle noise using a filter of size 41 x 41. As shown in Figure 2 and 3 overthe whole image. This filter slides along the image similar tohow a convolution Neural Network works. The filter checksif there is any cluster of objects near it if there is it discard thethat iteration of the filter and moves to the next one. Figure2 will help understand this problem. All cells in Figure 1 and 2 colored in green are obstacles and white are empty and onescolored in grey are unexplored. In our experiment we use a 41 by 41 filter size but for simplicity we use a filter of size 6by 6.

Figure 1: Filter values that are got from the image which does not satisfy the two conditions

The filter has an area called the target area this is shown in the red border in Figure 2 and Figure 3 this area is the wherethe detection occurs. If there is no object in this area that is ifthe value is either 0 or 127 then the current area of interest is discarded and the next one is taken. This is done to preventunwanted resource utilization. Doing so helps us move tonext one without wasting resources. The target area can be changed as per the users choice. Increasing this value will get better accuracy but will take a hit on resources. Upon trying multiple sizes of target area we recommend the size of 2 by 2.

Once a pixel of number 255 or an object is discovered inthis target area. we then check all the values in the surround-ing filter if there is a object or value 255 current filter is dis-carded. Figure 2 shows a good example where the filter val-ues are discarded and the next one is taken. There are 255 inthe the target area and then there are multiple 255 around itthis hows. It figures out if there is a object close to it

Figure 2: Filter values that are got from the image which satisfies the two conditions

Execution or Usage

This can be addded directly to your code

First initialise the code

import rem_noise from rem_noise
obj=rem_noise()

To execute

postproc=obj.remove_phan_obj(map_as_an_numpy_array)

postproc will have the map which do not have any phantom obsticles. Which is a numpy array of size of the input image.

obj.save(location.jpg)

Will save the image in jpg format at location