Multiple Human Tracking Using Binary Infrared Sensors.
Bottom Line:
Our algorithm can estimate multiple human movement paths without a priori knowledge of the number of humans in the room.By repeating predictions and estimations of human positions and links from the previous human positions to the estimated ones at each time period, human movement paths can be estimated.Simulation-based evaluation results show that our algorithm can dynamically trace human movement paths.
View Article:
PubMed Central - PubMed
Affiliation: The University of Aizu, Aizuwakamatsu, Fukushima 965-8580, Japan. miyazaki@u-aizu.ac.jp.
ABSTRACT
To create a context-aware environment, human locations and movement paths must be considered. In this paper, we propose an algorithm that tracks human movement paths using only binary sensed data obtained by infrared (IR) sensors attached to the ceiling of a room. Our algorithm can estimate multiple human movement paths without a priori knowledge of the number of humans in the room. By repeating predictions and estimations of human positions and links from the previous human positions to the estimated ones at each time period, human movement paths can be estimated. Simulation-based evaluation results show that our algorithm can dynamically trace human movement paths. No MeSH data available. |
Related In:
Results -
Collection
License getmorefigures.php?uid=PMC4507662&req=5
Mentions: Note that each element can belong to more than one cluster in our new clustering method, contrary to that used in our previous algorithm [24]. Thus, we can manage the case that multiple humans exist in the detection range of the same IR sensor, which cannot be managed by our previous algorithm introduced in [24]:(1)min∀ci,∀cj∈C;i≠jΔ(ci,cj)=E(ci∪cj)−E(cj)−E(cj)(2)E(c)=∑∀k∈c{(xk−X¯)2+(yk−Y¯)2}X¯=1/c/∑∀k∈cxk, Y¯=1/c/∑∀k∈cyk where is the coordinate of element , and is the number of elements in cluster c:(3)r2<1/ci∪cj/E(ci∪cj)(4)xi(t)=∑∀s∈ci(t)ws(t)⋅xs(t)∑∀s∈ci(t)ws(t), yi(t)=∑∀s∈ci(t)ws(t)⋅ys(t)∑∀s∈ci(t)ws(t) where “s” is a sensor ID. The weighted centers of all clusters are set to CC(t). They become candidates for human positions at time t. Figure 4 shows an example. In the example, cluster c1(t) consists of sensors s = 1, s = 2, and s = 3. If the position coordinates of the sensors are (10, 10), (9, 5), and (14, 8), and their weights are w1(t) = 4, w2(t) = 1, and w3(t) = 2, the center coordinate cc1(t) will be (11.0, 8.7), i.e., x1(t) = (4 × 10 + 1 × 9 + 2 × 14)/(4 + 1 + 2) and y1(t) = (4 × 10 + 1 × 5 + 2 × 8)/(4 + 1 + 2). Furthermore, sensor s = 2 belongs to another cluster c3(t). Thus, the coordinate of s = 2 is also used to calculate the weighted center cc3(t) of cluster c3(t). If a cluster contains only one sensor, like cluster c2(t) in Figure 4, the center position coordinate is simply the position of the sensor. After all weighted center coordinates of the clusters are calculated and set to CC(t), the path estimation step will be initiated. |
View Article: PubMed Central - PubMed
Affiliation: The University of Aizu, Aizuwakamatsu, Fukushima 965-8580, Japan. miyazaki@u-aizu.ac.jp.
No MeSH data available.