Public transport occupancy estimation using WLAN probing

Abstract

Abstract-Prediction of availability of physical services can be a valuable addition to transportation systems operation. In this paper we are focusing on estimation of public transport occupancy (PTO), or more specifically, on estimating bus passenger load, i.e., the number of people on the bus. This information can be used by bus operators as input to the analysis of bus routes' efficiency, or to provide an app indicating passenger load. PTO estimation based on collecting WiFi probes emitted by WiFi enabled devices is cheap and easy to install. This paper presents a prototype implementation of this method, analysis of the collected data and of the estimation algorithm accuracy. Analysis of passenger load in a bus has indicated that there are two main challenges of the estimation using WiFi probes. The algorithm provides overestimation due to inclusion of WiFi devices that are outside the bus and underestimation due to exclusion of people without an active WiFi enabled device or by missing out probes in the detection algorithm from devices carried on board. We have shown how by fine-tuning parameters of the algorithm the probes received from people outside the bus can be filtered out thereby reducing the severity of the underestimation problem. The typical approach to combat the overestimation problem is to make the adjustments based on a statistical ratio of people possessing a WiFi enabled smart device over the whole population.

Classifying Passenger and Non-passenger Signals in Public Transportation by Analysing Mobile Device Wi-Fi Activity

“…Comparison with other methods is difficult because of differing goals, assumptions or technical limitations. Mikkelsen et al [11] use a simpler filtering mechanism with only 2 parameters, RSSI and device visibility duration. They report an accuracy of around 50%, using a minimum duration of 6 minutes which is too large to provide real-time congestion information.…”

“…Mikkelsen et al [11] use a mechanism very similar to ours but using only two parameters, RSSI and duration of device visibility, to obtain 50% accuracy. Compared to their mechanism, ours uses five filtering parameters in a two-tier approach that accounts for devices likely located inside or outside the bus and performs significantly better.…”

“…Number of Pax Remark 1 9 2 12 Departured 3 12 4 12 5 12 6 13 get on 1 7 13 8 13 9 13 get on 1, get off 1 10 13 11 13 Arrived cumulative duration of the bus trip in minutes, the second column is the number of passengers aboard during each of those minutelong intervals, the third column lists significant events such as bus departure/arrival or passengers getting on/off the bus. We use this information to visualize passengers' behaviour on the bus to help validate a control data set for evaluation of our real-time classification mechanism.…”

A Case Study of WiFi Sniffing Performance Evaluation

“…In this section, we present the comparison results of different channel configurations using the in-built WiFi module on board. We chose 4 different channel hopping strategies that can be found in the literature; these are: 1).ch1: hopping across the 11 802.11b/g/n channels sequentially; 2).ch2: hopping across the three non-overlapping channels (1, 6 and 11); 3).ch3: hopping across the channels from 1 to 13 by jumping to the next non-overlapping channel (1,7,13,2,8,3,9,4,10,5,11,6,12) [37]; 4).ch4: hopping across the specific sequence of channels (1,6,11,2,7,3,8,4,9,5,10), also refereed to as the default Kismet hopping schedule [38].…”

Crowd data analytics as seen from WiFi

“…and left the bus. Similar measurements were conducted in Aalborg [84]. As well as for the shuttle bus for Thammasat University, Thailand [85].…”