UnifiedAR 1

Activity Recognition

DataSet

Dataset Format:

Object:

An object is a primitive object, a vector or in the form of a tuple of data components:
Object ={o|     o is Primitive or
                o=[o_1, ... , o_n] such that o_i is Object(Vector of object) or
                o=(Prop_1, ... , Prop_n) forall i in {1...n}, Prop_i(o) is Object}

Time Object:

Time might be a point, in case of an instantaneous event, or an interval during if it is durative. Supported durative time is range.

time | [start_time:end_time]

Event:

Type	Actor	Time

Sensor Events:

(Type, Value)	SensorId	Time

Activity Events:

ActivityId	ActorId	Time

DataInformation:

Sensor Info

Id	Name	Cumulative	OnChange	Nominal	Range	Location	Object	Sensor

Activity Info

Id	Name

File format: CSV

Sensor Info:

Id	Name	Cumulative	OnChange	Nominal	Range	Location	Object	Sensor
int	string	bool	bool	bool	json {min,max}/{items}	string	string	string
in case of nominal sensors, the range contain items and for numeric sensors, the range contain min and max

Sensor events:

Type	Value	SensorId	Time

Activity events:

ActivityId	ActorId	StartTime	EndTime

Approaches

\begin{Example}[Different Segmentation approaches] \end{Example} \begin{lstlisting}[mathescape=true] function Fixed time window(S,X,r,l) {//S=SegmentHistory, X=Events, //r=Shift, l=windowLength p=begin(S[last]) return X.eventsIn([p + r : p + r + l]); } function Fixed siding window(S,X,r,l) { prev_w=S[last]; p=begin(S[last]) be=first({e \in X| p + r $\leq$ time(e)} return X.eventsIn([be : be + l]); } function Significant events(S,X,m) {//m=significant events per segments se=significantEvents(X) $\subseteq$ X begin=time(se[1]);//next significant event end=time(se[1 + m]); return X.eventsIn([begin:end]); } //Probabilistic Approach given:(By analyzing training set) $ws(A_m)$ is average window size of activity $A_m$ $w_1 = min {ws(A_1), ws(A_2), ..., ws(A_M)}$ $w_L = median{ws(A_1), ws(A_2), ..., ws(A_M)}$ $w_l=(w_L-w_1)\times l/L+w_1$ $window_sizes= {w_1, w_2, . . . , w_L}$ $P(w_l /A_m)$//probability of windows length $w_l$ for an activity Am $P(A_m /s_i)$//probability of Activity $A_i$ associated with the sensor $s_i$. function Probabilistic Approach(S,X) { x=nextEvent(X) $w^{\star} =\underset{w_l}{max} {P(w_l /x)}=\underset{w_l}{max}[P(w_l /A_m)\times P(A_m /x)] $ end=time(x);//Next event return X.eventsIn(end-$w^\star$,end]); } function Metric base Approach(S,X) {//S=SegmentHistory, X=Events
indx=len(S[last])+1 //first event not in old segment $m_i=metric({X[indx],...,X[i]})$ find first i which $H({m_{0}....m_i})$ is true// return X.eventsIn([time(X[indx]):time(X[i])]); } function SWAB Approach(S,X,bs) {//bs=Buffer size
indx=len(S[last])+1 //first event not in old segment $m=BottomUp({X[indx],...,X[indx+bs]})$ return m[0]; } \end{lstlisting}

Similar Works

pyActLearn -> documents