Reimplement new laser point detection method

[mzur]

• Implement new script for automatic detection
• Update manual detection so it does not need to download the images any more
• Update the manual based on Color threshold lp detection #95

Based on: #95
References: #44

[mzur]

@dlangenk Is this really in meters?

[dlangenk]

This is the comparison between delphi and new DoG based algorithm (v14).

Metrics:

• perfect/perfect (%): Number/Percentage of images where all laserpoints were found within a radius of 15px
• gt: Number of ground truth images
• graceful/graceful (%): Number/Percentage where the failsafe image area too large was triggered
• partial/partial(%): Either not all laserpoints have been found or some were located incorrectly
• num_partial_lps_found: Number of laserpoints that have been found in these partial images

dataset	version	f1	perfect	perfect (%)	gt	graceful	graceful (%)	partial	partial (%)	num_partial_lps_found
1	v14	0.8961	518	74.1	699	6	0.9	173	24.7	519
2	delphi	0.5036	107	30.7	349	N/A	N/A	16	4.6	33
2	v14	0.9952	344	98.6	349	1	0.3	4	1.1	12
3	delphi	0.0198	4	1.0	400	N/A	N/A	0	0.0	0
3	v14	0.9350	356	89.0	400	0	0.0	36	9.0	72
4	delphi	0.0080	4	0.4	991	N/A	N/A	0	0.0	0
4	v14	0.8532	780	78.7	991	0	0.0	131	13.2	262
5	delphi	0.1470	30	6.3	479	N/A	N/A	12	2.5	24
5	v14	0.9885	468	97.7	479	1	0.2	10	2.1	30
6	v14	0.9447	178	89.0	200	2	1.0	20	10.0	40
7	delphi	0.9950	198	99.0	200	N/A	N/A	0	0.0	0
7	v14	0.8107	136	68.0	200	9	4.5	45	22.5	90
8	v14	0.9644	829	89.3	928	0	0.0	99	10.7	297
9	v14	0.9994	549	99.8	550	0	0.0	1	0.2	3
10	v14	0.8604	746	85.4	874	0	0.0	8	0.9	32
11	v14	0.7392	370	50.0	740	0	0.0	354	47.8	708
12	v14	0.3301	36	14.1	256	0	0.0	97	37.9	194

[dlangenk]

I didn't touch the TODOs in your code (manual lps for tiled images etc.)