An Autonomous RSSI Filtering Method for Dealing with Human Movement Effects in an RSSI-Based Indoor Localization System

Abstract

In this paper, an experimental evaluation of received signal strength indicator (RSSI-based) localization methods in an indoor wireless network is studied. The major contributions of this work are twofold. First, the well-known and widely used min–max and trilateration methods are tested in the cases of without and with human movement effects. By this purpose, how RSSI data during human movements affect the accuracy of such methods and which method shows the best position estimation result, have been investigated. Second, we also design and develop a new RSSI filter to automatically reduce RSSI variation and the position estimation error caused by human movements. Experiments are carried out in a parking building. An LPC2103F microcontroller interfaced with a CC2500 RF transceiver as a low-cost, low power, 2.4 GHz radio module is used as a wireless node. Results demonstrate that without human movement effects, the performances by both localization methods are not much different. However, with human movement effects, the min–max method shows better accuracy than the trilateration method in handling the RSSI variation problem. The results also indicate that by applying the proposed RSSI filter, it can directly cope with the RSSI variation problem caused by humans. The localization error decreases by 69.87% for the case of the min–max method, and it decreases by 72.74% for the case of the trilateration method (the best case). Compared with the case of employing the moving average filter as the commonly used filter, the localization error only decreases by 18.67% and 12.99%, respectively.

Appendix A: Span thresholding filter [27] and its results

The span thresholding (ST) filter was introduced in [27]. It was proposed as a preliminary/guideline solution for handling the RSSI variation problem caused by human movements. The ST filter are shown in (29) to (32). To detect human movements, a number of measured RSSI samples in each predefined window size ($W$) are first checked. If the absolute difference level between the maximum RSSI (${RSSI\_max}_{j}$) and the minimum RSSI (${RSSI\_min}_{j}$) is lower than the threshold ($Threshold$), it estimates that there is no human in the wireless network. Here, the RSSI input at the sample number $j$ (${RSSI}_{j}$) is used as the RSSI output (${RSSI\_ST}_{j}$) for the position estimation. On the other hand, if the absolute difference exceeds the threshold, indicating human movements, the RSSI output from the previous sample (${RSSI\_ST}_{j-1}$) is used on estimating the position instead.

$${RSSI\_max}_{j}=max\left({RSSI}_{j},{RSSI}_{j-1}, {RSSI}_{j-2},\dots {RSSI}_{j-W+1}\right)$$

(29)

$${RSSI\_min}_{j}=min({RSSI}_{j},{RSSI}_{j-1}, {RSSI}_{j-2},\dots {RSSI}_{j-W+1})$$

(30)

$${RSSI\_ST}_{j}=\left\{\begin{array}{c}{RSSI}_{j}, \left|{RSSI\_max}_{j}-{RSSI\_min}_{j}\right|\le Threshold\\ {RSSI\_ST}_{j-1}, \left|{RSSI\_max}_{j}-{RSSI\_min}_{j}\right|>Threshold\end{array}\right.$$

(31)

$$Threshold=max\left(\left|{RSSI\_max}_{k}-{RSSI\_min}_{k}\right|, \left|{RSSI\_max}_{k+1}-{RSSI\_min}_{k+1}\right|,\left|{RSSI\_max}_{k+2}-{RSSI\_min}_{k+2}\right|,\dots \left|{RSSI\_max}_{n}-{RSSI\_min}_{n}\right|\right)$$

(32)

The optimal values of $W$ and $Threshold$ are manually determined by the experiments. Here, $Threshold$ refers to the maximum of the RSSI variation level during no human presence in the wireless network. It is determined in an offline phase by finding the maximum value of the absolute subtraction between the maximum RSSI and the minimum RSSI (in each window) of the measured RSSI, where $k$ is the order of the window, and $n$ is the number of windows.

From our experiments, the RSSI signals from Fig. 8 (i.e., without human movements) are used to calculate the $Threshold$ value. Here, when $W$ is 110 (the same setting for the proposed method in Sect. 3 and the comparative filter in Sect. 4.3), the maximum value of the absolute subtraction or $Threshold$ is equal to 3.5, as shown in Fig. 21a. We also show the absolute subtraction results in the case of using the measured RSSI data from Fig. 10 (i.e., with human movements), to see the difference, as shown in Fig. 21b.

Fig. 21

The absolute subtraction between the maximum RSSI and the minimum RSSI with $W=110$; (a) and (b) are the results from the RSSI signals in Figs. 8 and 10, respectively

Finally, the experimental results show that by applying the ST filter with $W$ of 110 and $Threshold$ of 3.5, as shown in Table 4, the average error distances decrease from 0.35 m to 0.21 m for the case of the min–max method, and 1.10 m to 0.66 m for the case of the trilateration method. Thus, the average error distances decrease by 41.70% and 40.21%, respectively. The error distances are also illustrated in Fig. 22. Here, we prove that the ST filter outperforms the moving average filter in terms of the estimation accuracy, but it is worse than the proposed method as presented in this paper. In addition, the ST filter still has the major limitation that it requires the off-line phase to run its algorithm.

Table 4 The average error distances determined by the ST filter with $W$ of 110 and $Threshold$ of 3.5

Fig. 22

The error distances for the target position (${x}_{t}$ = 1.80 m, ${y}_{t}$ = 3.10 m) after applying the ST filter with $W$ of 110 at $Threshold$ of 3.5