thank you um that of this uh presentation is a state driven particle filter for

multi person tracking um these are so work with uh for that it let us

to and they're gonna look this

this is the other end of the of the presentation first uh define the introduction

to the problem of tracking

then i will uh

formulate a explain how the particle filtering general is formulated

uh you have touched and all that yeah

uh

afterwards

it's similar compute its own a computer

this is not this computer this is the laptop

oh

so it's uh

okay thank you

and

afterwards i will uh explain our proposal which is based on particle filtering and use

the state driven particle filtering and finally we present experimental results on the completion

well us most of you may know a multi person tracking have different applications for

instance or radiance driver assistance uh robot interaction with the media

um in sometimes uh the some pickups that some typical problems such as occlusions or

more than fifteen or lost targets detection is not uh addressed by the literature uh

people some things that present day um we propose something general or ready

basis in which these uh occlusions and problems doesn't happen uh so in this case

we have an example of target hijacking we have a target in the foreground weeks

recruits one in the background and high get the model

oh here we would have the problem of uh lost target detection uh and occluded

person by the bigram soul or by the scene

and um in general approaches and tracking dealing with tracking and not they do not

take these situations into account uh but typically um they use heuristics which are based

on a book um solutions for example detecting the targets overlapping or um stopping the

updating with the model updating when this happens getting the targets when detections are not

associated for a given set of frames et cetera uh some specific and papers which

uh really thinking talk on this problem by Q for example using um with the

common sff um so who proposed a post processing you know that to um you

know it to uh relate the relation the interactions between objects but this is made

as a post processing for the tracking for the for the acquisition and one hundred

depending on who proposed uh occlusion up that uh structures

a orange or object if is to formalize these problems uh in a in a

whole general framework you know what i what chuck hijacking model drifting decrease the false

positives the false negatives and also probably the identity switch yeah in the trucks

how do we make that well our contribution is and based on a graph uh

continue containing states this estates are formalising their in the nodes and uh the transitions

and between the nodes

are represented by the arcs

uh this assigned the scheme of our proposal i will go into it later but

uh just to right

say that uh one of the main uh

the important things of the of our proposal is that it's as and generic enough

so or the papers other proposals the specific techniques for occlusions to drop can be

already included here

now i will uh briefly overview of our particle filtering works uh just a set

of particle filtering use of a would method which is able to um follow the

objects in the scene um and we uh make use of five tracking-bydetection by reading

this means that we have and detections with our which are painted in general we

have the tracks which are painted in

blue in this case and uh with the corresponding id and a set of particles

which are the ones that are distributed along the frames in order to search for

the new appearance of the of the truck

particle filter has two ways that's the first one is the prediction also known as

a importance function this is the equation of the importance function um

and here you see the main components of these equations this is this will be

used later in our formulation uh it's a normally based on human detector are for

example whole class svm the historic oriented gradients and a classifier uh the dynamics pdf

uh for example in our case we use a random walk but we could use

of our higher-order approaches an upright which is open a predefined way to search for

the um new occurrences of the truck

the second is that if the is the correction which is the weighting of the

particles uh in this case we use a color model very simple is the colour

histogram of the of this torso region of the pedestrian

and that you see here uh we have uh the particles are and painted according

to the weighting of the of the of the colour matching so the brighter the

particle the highway

uh this is a proposal of we propose a

state based tracking we have potential tracked occluded lost states and also the additional that

when we keep the track

and now we will explain each one of the uh specific a steps not all

the details but uh the main important components that uh will highlight why this proposal

is useful

first um we have a detection in general and the first we make use to

create a new track which is painted in dashed blue and uh we initialize these

track with the state potential

after a set of frames if and their track has corresponding detections we obtain that

the state from potential to track

otherwise if the track does not have enough corresponding detections in the next frames we

give a soul goes from potential to that

then in the case that the person is correctly tracked so out of the results

of are several uh frames there's a corresponding detection we can apply frame-by-frame corresponding actions

specifically them at the track state this is an example there in a set of

actions for example the weighting of the particles gives an initial model used in the

data association between detections and the track and i will just highlight uh for model

of time their importance function which is how we draw we distribute the particles in

the frames in the case in this case uh we only use the dynamics without

thinking and talk on the detections in order to deceive it is practical so as

you see here we from the original equation we only use one part

then um you know to make the transitions

uh from track to do that and two lost we propose to conditions are quite

simple the first one is that the isolation of the tracks so if two tracks

overlap each other

uh in this case we use that pascal a criterion that pascal overlapping criterion is

also known the jaccard overlapping criterion is quite simple and the other um condition is

the classifier um and studied in this case we use a online classifier would uh

more than is the colour of the course of the person and

we track it a long time and in the case that we see in the

yellow uh line we see that tracked person in which the confidence on the of

the online model uh is maintained a long time

in a real we see the case of an occluded target in which there and

confidence of the of the color model and jumps down so we detect this uh

first we filter the signal which someone the kalman filter and then we test these

and jumps with the generalized likelihood ratio test

so in the case that and these conditions are fulfilled so the um pedestrian the

person is not isolated and there's a jump in their classifier we go to the

occluded uh

state

and so one here we have a additional uh conditions from occluded to track come

from the same occluded to this to the occluded the state and

i highlight this again the importance function in this case is based on detections so

when the truck uh when the person disappears which because it so you the by

another person we go to the occluded and the state and the particles are not

drawn according to i don't know what is it wasn't importance function of the track

state but are drawn according to the detections around

and so one so um we have the same conditions with the same tests in

order to make the transitions between the states in this case uh we see that

the lost target is occluded by their scene

and so on

and we finally if a target has been lost for different a different set of

frames week

to sum up um these are the experimental results um we have a me to

use of three datasets the tud-crossing to decompose and the one on pets two thousand

and nine and compared to the original um formulation of the particle filtering and our

proposed state based

one and give several statistics

the first ones are the with the object tracking precision and accuracy in the case

of the precision it evaluates the uh overlapping between the ground truth and the track

itself and we see in its around plus or minus one percent but this is

not significant if since uh we have um already detected the person but there's a

slight displacement with respect to the position of the tracks with respect to the ground

truth

uh the one which is a really significantly fees the accuracy in this case with

our proposal we gain about seven percent and also the false negative rate on the

false positives pretty much which are also decreased in our proposal and finally the identity

switches this is when a truck has to be really se reinitialised and given a

different id from the one but it had in our case uh the number channel

and then i show you some examples

and this case we have attract a pedestrian which is isolated and it works quite

well both with the original proposal and on the with the original approach and the

state based approach we see uh the potential track which is initialized then it uh

divorced what tracked track then uh you can see she has several detection so now

false positives will which will disappear but the track is currently uh track i don't

time so there's no problem this case

in the case in which we really see that the differences between the traditional approach

in our proposal is in this case for example have an occluded uh case we

have to target which overlap each other and in the first case the um person

on the front high yets the person on the back so the model uh drifts

and it gets a stark on the background and finally there's a new uh track

for the new detection and then new id for these same person so he would

have a switching in the idea

in our case uh the same happens the two targets okay with each other but

the system is able to reinitialize the track with the same id by detecting that

in the third frame in this case and there's an occlusion

and finally the case in which a track has been lost in this case the

track is computed by the background but the scene um uh by the person but

the our system is able to detect that it hasn't lost and it finally uh

detect that in matches the track that have we lost with a new detection in

the system that colour model matches and it we initialize the tracker and with the

same i

so to sum up the conclusions are that we have presented a and the state

based uh tracking approach which is able to deal with it got tracking problems that

are their creations that hijacking of the target one and drifting at sea trial it

uh gives a performance improvements with respect to the traditional on the state based approach

it's applicable to different existing approaches such as uh or occlusion classifiers specifically some classifiers

are used we saw that kind of filter in the test and one with a

different state of the art

uh_huh lost target detections the could be actually included in our proposal and especially in

the future work um since this is a on uh mutual to be framework this

is we have make use of very simple ingredients so for example the um the

colour model for the for the person is quite simple socially going to more advanced

that um

components for instance appearance models for the for the for the tracks or an increasing

the number of particles depending on the state of the person uh we think that

it would uh provide a better results

thank you

yeah

uh experimentally there are uh

as many as the ones you would use for a non state based approach but

multiplied by the number of states

uh_huh

and then they are related with the transitions for the for example for the isolation

uh we use the syllable five which is typically used uh

for the overlapping and for the classifier itself it's also i just the bike and

by looking experimentally

no not at the moment we uh have one frame per second

yeah

for

the case is my

ten people more or less yeah can people yeah

to become an attempt works

yeah in fact it would because uh the particle filtering uh allows you to process

independently the different uh tracks and seems uh each one of the states allows you

to um to make use of different parameters depending on this date you could spend

less particle so less processing on the trucks that for instance are isolated so by

accelerating these things ending in these things we can we think that it would be

to reach we have then

yeah

it is based on the chance writing

or tracking on particle filtering

yeah

yeah i don't know which one but it's based on that

okay fine

a speaker