search

mvondracek / VUT-FIT-POVa-2018-Pedestrian-Tracking

Computer vision system for tracking pedestrians in a scene observed by multiple cameras.
10 stars 4 forks source link

Task definition #3

Closed mvondracek closed 5 years ago

mvondracek commented 5 years ago

Suggestion for possible visual representation of final result -The Marauder's Map? What's this rubbish?

EDIT: Harry Potter's Marauder's Map: Localizing and Tracking Multiple Persons-of-Interest

xstast24 commented 5 years ago

TASK 1) Vezmeme 1 ci vice kamer, umistime je do mistnosti. Zmerime jejich pozice. Natocime se pri chozeni v mistnosti a pri vstupu/opusteni zaberu. Chceme, at se obrazy 2 kamer castecne prekryvaji, ale pote muzeme udelat i stopaz, kde se neprekryvaji (pro pripadne rozsireni). 2) Spojime obrazy z kamer pomoci OpenCV - sachovnice? 3) Pomocí Open pose zdetekujeme postavy - ty barevné "kostry", co dělá open pose. Pokud jdou (nevim), muzeme zkusit i bounding boxy, pro snazsi praci pozdeji pri trackovani. 4) Pomoci triangulace zjistíme pozice osob. 3) Trackujeme jednotlivé osoby (napříč snímky pospojujeme příslušné osoby). Pokud se predavaji pouze "kostry" z open pose, bude potreba vytahnout jejich okoli z puvodniho obrazu, abychom mohli napr. pocitat histogram atp. Metodu trackovani vhodne vygooglime. :D 5) Nyni mame stopu osoby. Tu vyznacime do 2D pudorysu. Cili výsledek může být třeba zobrazení barevných trajektorií podle pohybu osob v půdorysu.

mozna rozsireni 1) Detekce pozice/natoceni lidi (by @Hradis) - treba do pudorysu ukazat smer, kterym koukaji? 2) Re-identifikace lidi po odchodu a opetovnem navraceni (by @Hradis)

Co muze trvat (by @Hradis) 1) Spojit obraz z vice kamer (openCV, sachovnice) muze trvat dlouho. Z jedne kalibrovane kamery by mohlo jit snadno trojclenkou! 2) Rozchodit open pose je OK, ale muze se na necem zaseknout (1h az 2dny).

mvondracek commented 5 years ago

@xstast24 Please finalize and translate task definition, then place it to the introduction of the report.

mvondracek commented 5 years ago

The aim of this project was to create a computer vision system capable of tracking pedestrians in video sequences captured by multiple stationary cameras. The input should be video streams from the cameras place in a room. The output should be a visualisation of tracks of the people who appeared in front of the cameras. A flow of the system can be described by the following steps:

1) Let's have a room with 1 or multiple cameras with known positions. 2) Record videos of people walking in the room. 3) Detect figures in the videos. 4) Find out matching figures (the same person) across the videos from multiple cameras. 5) Locate the person in 3D space. 6) Track the targets across the video sequence (chronologically). 7) Mark down 2D track for each person and visualise it in a floor plan of the room.