0:00:15thank you um that of this uh presentation is a state driven particle filter for
0:00:22multi person tracking um these are so work with uh for that it let us
0:00:29to and they're gonna look this
0:00:32this is the other end of the of the presentation first uh define the introduction
0:00:36to the problem of tracking
0:00:38then i will uh
0:00:41formulate a explain how the particle filtering general is formulated
0:00:46uh you have touched and all that yeah
0:00:57it's similar compute its own a computer
0:01:00this is not this computer this is the laptop
0:01:05so it's uh
0:01:08okay thank you
0:01:13afterwards i will uh explain our proposal which is based on particle filtering and use
0:01:19the state driven particle filtering and finally we present experimental results on the completion
0:01:25well us most of you may know a multi person tracking have different applications for
0:01:30instance or radiance driver assistance uh robot interaction with the media
0:01:35um in sometimes uh the some pickups that some typical problems such as occlusions or
0:01:43more than fifteen or lost targets detection is not uh addressed by the literature uh
0:01:50people some things that present day um we propose something general or ready
0:01:57basis in which these uh occlusions and problems doesn't happen uh so in this case
0:02:03we have an example of target hijacking we have a target in the foreground weeks
0:02:09recruits one in the background and high get the model
0:02:14oh here we would have the problem of uh lost target detection uh and occluded
0:02:19person by the bigram soul or by the scene
0:02:23and um in general approaches and tracking dealing with tracking and not they do not
0:02:29take these situations into account uh but typically um they use heuristics which are based
0:02:36on a book um solutions for example detecting the targets overlapping or um stopping the
0:02:43updating with the model updating when this happens getting the targets when detections are not
0:02:49associated for a given set of frames et cetera uh some specific and papers which
0:02:56uh really thinking talk on this problem by Q for example using um with the
0:03:00common sff um so who proposed a post processing you know that to um you
0:03:08know it to uh relate the relation the interactions between objects but this is made
0:03:13as a post processing for the tracking for the for the acquisition and one hundred
0:03:19depending on who proposed uh occlusion up that uh structures
0:03:26a orange or object if is to formalize these problems uh in a in a
0:03:31whole general framework you know what i what chuck hijacking model drifting decrease the false
0:03:38positives the false negatives and also probably the identity switch yeah in the trucks
0:03:44how do we make that well our contribution is and based on a graph uh
0:03:50continue containing states this estates are formalising their in the nodes and uh the transitions
0:03:57and between the nodes
0:04:00are represented by the arcs
0:04:02uh this assigned the scheme of our proposal i will go into it later but
0:04:08uh just to right
0:04:10say that uh one of the main uh
0:04:13the important things of the of our proposal is that it's as and generic enough
0:04:17so or the papers other proposals the specific techniques for occlusions to drop can be
0:04:23already included here
0:04:25now i will uh briefly overview of our particle filtering works uh just a set
0:04:31of particle filtering use of a would method which is able to um follow the
0:04:37objects in the scene um and we uh make use of five tracking-bydetection by reading
0:04:45this means that we have and detections with our which are painted in general we
0:04:49have the tracks which are painted in
0:04:52blue in this case and uh with the corresponding id and a set of particles
0:04:58which are the ones that are distributed along the frames in order to search for
0:05:03the new appearance of the of the truck
0:05:07particle filter has two ways that's the first one is the prediction also known as
0:05:12a importance function this is the equation of the importance function um
0:05:18and here you see the main components of these equations this is this will be
0:05:22used later in our formulation uh it's a normally based on human detector are for
0:05:30example whole class svm the historic oriented gradients and a classifier uh the dynamics pdf
0:05:37uh for example in our case we use a random walk but we could use
0:05:41of our higher-order approaches an upright which is open a predefined way to search for
0:05:47the um new occurrences of the truck
0:05:50the second is that if the is the correction which is the weighting of the
0:05:53particles uh in this case we use a color model very simple is the colour
0:05:58histogram of the of this torso region of the pedestrian
0:06:03and that you see here uh we have uh the particles are and painted according
0:06:08to the weighting of the of the of the colour matching so the brighter the
0:06:12particle the highway
0:06:16uh this is a proposal of we propose a
0:06:20state based tracking we have potential tracked occluded lost states and also the additional that
0:06:28when we keep the track
0:06:30and now we will explain each one of the uh specific a steps not all
0:06:34the details but uh the main important components that uh will highlight why this proposal
0:06:40is useful
0:06:42first um we have a detection in general and the first we make use to
0:06:47create a new track which is painted in dashed blue and uh we initialize these
0:06:52track with the state potential
0:06:56after a set of frames if and their track has corresponding detections we obtain that
0:07:02the state from potential to track
0:07:05otherwise if the track does not have enough corresponding detections in the next frames we
0:07:11give a soul goes from potential to that
0:07:16then in the case that the person is correctly tracked so out of the results
0:07:21of are several uh frames there's a corresponding detection we can apply frame-by-frame corresponding actions
0:07:29specifically them at the track state this is an example there in a set of
0:07:35actions for example the weighting of the particles gives an initial model used in the
0:07:39data association between detections and the track and i will just highlight uh for model
0:07:45of time their importance function which is how we draw we distribute the particles in
0:07:51the frames in the case in this case uh we only use the dynamics without
0:07:57thinking and talk on the detections in order to deceive it is practical so as
0:08:00you see here we from the original equation we only use one part
0:08:07then um you know to make the transitions
0:08:11uh from track to do that and two lost we propose to conditions are quite
0:08:16simple the first one is that the isolation of the tracks so if two tracks
0:08:22overlap each other
0:08:24uh in this case we use that pascal a criterion that pascal overlapping criterion is
0:08:29also known the jaccard overlapping criterion is quite simple and the other um condition is
0:08:36the classifier um and studied in this case we use a online classifier would uh
0:08:44more than is the colour of the course of the person and
0:08:49we track it a long time and in the case that we see in the
0:08:54yellow uh line we see that tracked person in which the confidence on the of
0:08:59the online model uh is maintained a long time
0:09:03in a real we see the case of an occluded target in which there and
0:09:09confidence of the of the color model and jumps down so we detect this uh
0:09:14first we filter the signal which someone the kalman filter and then we test these
0:09:21and jumps with the generalized likelihood ratio test
0:09:25so in the case that and these conditions are fulfilled so the um pedestrian the
0:09:32person is not isolated and there's a jump in their classifier we go to the
0:09:38occluded uh
0:09:40and so one here we have a additional uh conditions from occluded to track come
0:09:48from the same occluded to this to the occluded the state and
0:09:53i highlight this again the importance function in this case is based on detections so
0:09:58when the truck uh when the person disappears which because it so you the by
0:10:03another person we go to the occluded and the state and the particles are not
0:10:08drawn according to i don't know what is it wasn't importance function of the track
0:10:13state but are drawn according to the detections around
0:10:20and so one so um we have the same conditions with the same tests in
0:10:26order to make the transitions between the states in this case uh we see that
0:10:30the lost target is occluded by their scene
0:10:36and so on
0:10:38and we finally if a target has been lost for different a different set of
0:10:42frames week
0:10:45to sum up um these are the experimental results um we have a me to
0:10:49use of three datasets the tud-crossing to decompose and the one on pets two thousand
0:10:55and nine and compared to the original um formulation of the particle filtering and our
0:11:02proposed state based
0:11:05one and give several statistics
0:11:09the first ones are the with the object tracking precision and accuracy in the case
0:11:15of the precision it evaluates the uh overlapping between the ground truth and the track
0:11:20itself and we see in its around plus or minus one percent but this is
0:11:27not significant if since uh we have um already detected the person but there's a
0:11:33slight displacement with respect to the position of the tracks with respect to the ground
0:11:40uh the one which is a really significantly fees the accuracy in this case with
0:11:45our proposal we gain about seven percent and also the false negative rate on the
0:11:50false positives pretty much which are also decreased in our proposal and finally the identity
0:11:56switches this is when a truck has to be really se reinitialised and given a
0:12:02different id from the one but it had in our case uh the number channel
0:12:08and then i show you some examples
0:12:12and this case we have attract a pedestrian which is isolated and it works quite
0:12:18well both with the original proposal and on the with the original approach and the
0:12:24state based approach we see uh the potential track which is initialized then it uh
0:12:31divorced what tracked track then uh you can see she has several detection so now
0:12:38false positives will which will disappear but the track is currently uh track i don't
0:12:43time so there's no problem this case
0:12:46in the case in which we really see that the differences between the traditional approach
0:12:50in our proposal is in this case for example have an occluded uh case we
0:12:56have to target which overlap each other and in the first case the um person
0:13:02on the front high yets the person on the back so the model uh drifts
0:13:08and it gets a stark on the background and finally there's a new uh track
0:13:13for the new detection and then new id for these same person so he would
0:13:18have a switching in the idea
0:13:20in our case uh the same happens the two targets okay with each other but
0:13:25the system is able to reinitialize the track with the same id by detecting that
0:13:31in the third frame in this case and there's an occlusion
0:13:36and finally the case in which a track has been lost in this case the
0:13:40track is computed by the background but the scene um uh by the person but
0:13:46the our system is able to detect that it hasn't lost and it finally uh
0:13:52detect that in matches the track that have we lost with a new detection in
0:13:57the system that colour model matches and it we initialize the tracker and with the
0:14:03same i
0:14:05so to sum up the conclusions are that we have presented a and the state
0:14:10based uh tracking approach which is able to deal with it got tracking problems that
0:14:15are their creations that hijacking of the target one and drifting at sea trial it
0:14:20uh gives a performance improvements with respect to the traditional on the state based approach
0:14:26it's applicable to different existing approaches such as uh or occlusion classifiers specifically some classifiers
0:14:35are used we saw that kind of filter in the test and one with a
0:14:40different state of the art
0:14:42uh_huh lost target detections the could be actually included in our proposal and especially in
0:14:48the future work um since this is a on uh mutual to be framework this
0:14:54is we have make use of very simple ingredients so for example the um the
0:14:59colour model for the for the person is quite simple socially going to more advanced
0:15:06that um
0:15:08components for instance appearance models for the for the for the tracks or an increasing
0:15:13the number of particles depending on the state of the person uh we think that
0:15:18it would uh provide a better results
0:15:23thank you
0:15:42uh experimentally there are uh
0:15:47as many as the ones you would use for a non state based approach but
0:15:54multiplied by the number of states
0:16:03and then they are related with the transitions for the for example for the isolation
0:16:09uh we use the syllable five which is typically used uh
0:16:14for the overlapping and for the classifier itself it's also i just the bike and
0:16:20by looking experimentally
0:16:34no not at the moment we uh have one frame per second
0:16:45the case is my
0:16:48ten people more or less yeah can people yeah
0:17:02to become an attempt works
0:17:04yeah in fact it would because uh the particle filtering uh allows you to process
0:17:10independently the different uh tracks and seems uh each one of the states allows you
0:17:16to um to make use of different parameters depending on this date you could spend
0:17:22less particle so less processing on the trucks that for instance are isolated so by
0:17:27accelerating these things ending in these things we can we think that it would be
0:17:31to reach we have then
0:18:00it is based on the chance writing
0:18:03or tracking on particle filtering
0:18:14yeah i don't know which one but it's based on that
0:18:20okay fine
0:18:28a speaker