Skip to content

2. Loading Tracked Objects

Learning Goal: Understand how to inspect the output of fmdt-detect.

In this tutorial we are going to take a look into how we can examine the list of objects detected by a call to fmdt-detect.

Detection

We start with a simple detection, continuing to use demo.mp4 as our prototype video. 

import fmdt

vid_in = "demo.mp4"
trk    = "trk.txt"

res = fmdt.detect(vid_in_path=vid_in, trk_path=trk, trk_all=True)

This time, however, we pass in trk_all=True to indicate to fmdt-detect to track all celestial objects (stars, meteors & noise) and we also store the results of a call to fmdt.detect in the res variable. The return of fmdt.detect is a fmdt.res.DetectionResult.

TODO: Insert image here

We will study the DetectionResult more in depth in later tutorials. For now, we are interested in it's trk_list member, which contains a list of all the objects detected by fmdt-detect.

Tracking List

The --trk-path parameters tells fmdt-detect where to store its list of tracked objects. This entity, which we refer to as a Tracking List, is very useful for analyzing the performance of fmdt-detect.

We can examine this tracking list by accessing the trk_list: list[TrackedObject] field of our res.

print(type(res))             # fmdt.res.DetectionResult
print(type(res.trk_list))    # list
print(type(res.trk_list[0])) # fmdt.core.TrackedObject

Remember that if we ever want more information about a class we can just call help(ClassName). We can see how many total objects were detected by inquiring the length of trk_list:

print(len(res.trk_list)) # 83

for i in range(10):
    print(res.trk_list[i])

<Noise (4, 255), pos0: ((643.9, 832.0)), posT: (644.0, 831.9)>
<Noise (4, 255), pos0: ((562.9, 860.5)), posT: (563.8, 861.8)>
<Star (0, 255), pos0: ((909.5, 131.7)), posT: (909.2, 131.7)>
<Star (0, 255), pos0: ((242.6, 138.4)), posT: (242.3, 138.4)>
<Star (0, 255), pos0: ((1639.4, 209.5)), posT: (1639.6, 209.1)>
<Star (0, 255), pos0: ((1099.3, 281.6)), posT: (1099.4, 281.5)>
<Star (0, 255), pos0: ((694.2, 309.4)), posT: (694.1, 309.1)>
<Star (0, 255), pos0: ((1102.4, 361.9)), posT: (1103.0, 361.6)>
<Star (0, 255), pos0: ((1311.2, 451.6)), posT: (1311.5, 451.6)>
<Star (0, 255), pos0: ((240.7, 481.6)), posT: (240.8, 482.0)>

We can retain only the meteors (or stars, noise) by using Python's list comprehension and our TrackedObject class:

meteors = [m for m in res.trk_list if m.is_meteor()]
stars   = [s for s in res.trk_list if s.is_star()]
noise   = [n for n in res.trk_list if n.is_noise()]

because the TrackedObject class comes equipped with the three utility functions .is_meteor(), .is_start(), and .is_noise()

Alternatively, retrieve these lists with the .meteors(), .stars(), and .noise() functions of our res:

meteors = res.meteors()
stars   = res.stars()
noise   = res.noise()

Conclusion

This tutorial taught you how to load, filter, and examine the list of objects that are detected by fmdt-detect (and stored in the trk_path file). We will use this tracking list often when evaluating the performance of fmdt-detect with our database of videos. Keep following along to the next section to learn how to play around with different argument combinations or read up on the DetectionResult class here.