Secret room in the train

Abstract

Passenger trains in the Netherlands and in other countries suffer from crowdedness. Crowdedness in public transport can be defined as the extent to which a vehicle is crowded or filled to excess. Crowdedness can cause lower customer satisfaction, longer perceived waiting times and stress. A problem that amplifies the crowdedness in the train is that passengers are often not equally distributed over the train. Some compartments tend to be fuller than others. This problem may be countered by informing passengers about the current crowdedness in each separate location compartment in the train, so they can anticipate and enter the train in a relatively less crowded compartment. This can for example be done by integrating this information in an existing mobile application about travel information or by using dynamic signs at the platforms of stations. To be able to provide this information to the train passenger it is needed to measure the occupancy of locations in the train in real time. Therefore the following main research question has been drafted in this research: Which localization method is most suitable to monitor occupancy in the train in real time? To determine the most suitable method(s) it is first relevant to define which characteristics of a train environment are relevant with regards to indoor localization. The most important characteristics are described. Some trains already have Wi-Fi routers and/or security cameras installed which can be used for some methods of indoor localization. The location of the interior of a train compartment is known and static and the train compartment is thus a good fit for a local reference system that only functions for a small region. Due to this static infrastructure some assumption can also be made with regards to the locations of passengers, passengers are for example often located on chairs, and this can be used by indoor localization methods that are able to focus on a specific location. To evaluate the indoor localization methods they have been categorized per technology. Since this research is about monitoring train passengers technologies that require subjects to carry additional devices, such as tags, are disregarded as this is deemed too unpractical. An exception is made for Wi-Fi since many Dutch people carry a Wi-Fi device and have their Wi-Fi turned on most of the time. The technologies (that do not require tags) evaluated are: Wi-Fi/ WLAN, infrared, sound localization, ultra-wideband, camera and pressure sensors. The advantages and disadvantages of these methods are further assessed using a qualitative analysis of the literature. Based on this analysis and the characteristics of the train a combination of Wi-Fi and camera-based localization seems the most suitable. This is mainly because the infrastructure for these technologies is already available in the train and the costs of implementing these methods are therefore relatively low. The other technologies do not seem to have an edge in performance that can outweigh their relatively higher cost. The advantage of using two technologies seems that they can be used to verify and amplify each other and to mitigate each other’s disadvantages. These two technologies are further researched and tested in environments similar to a train. The focus of the tests lies on one train to narrow down the scope of this research. The train chosen for this is the FLIRT, which is employed by the railway operator NS, because the FLIRT already has cameras and Wi-Fi. A camera-based localization method is employed in this research by designing an algorithm that detects the difference between a frame of an empty train to the real-time frames of the footage from security cameras in a train by detecting differences in the color model of the pixels. In this algorithm the HSV color model is used and the focus lies on hue to avoid noise from differences in light. To determine the number 2 of taken seats this algorithm detects whether a relevant contour of pixels changed significantly on the headrest areas of the train. To determine the occupancy in the hallway of the train a ratio of changed pixels in the hallway is identified. This data of the hallway and the seat can be combined to estimate the occupancy in the whole compartment. To determine occupancy in the train compartment using Wi-Fi localization the number of unique Wi-Fi devices is measured. This is done by sniffing Wi-Fi probe requests, which are signals send out by Wi-Fi devices used to actively seek a Wi-Fi access point. Some probe requests contain a unique Media Access Control (MAC) address that belongs to the corresponding mobile devices. These MAC addresses are used to detect unique Wi-Fi devices. The camera-based localization method has been tested in an office environment by using old train chairs and both of the methods have been tested in an old train in a railway museum. When interpreting the results of the tests it is important to take into account that testing in a real train may lead to different results, as the test setting are not a perfect description of reality. From the tests derived that the camerabased localization has an average false negative error of 5-9% and an average false positive error of 1-3% during a train journey when used to estimate the number of taken seats in relation to the total number of available seats. During a train stop a false negative error of 4-5% and a false positive error of 8-9% have been found. For the hallway it can be stated that it seems like camera-based localization can used to estimate its occupancy to some extent, but this estimation is most likely not as accurate as it is for the occupancy of the taken seats. The reason for this is that in the hallway the number of people per area can vary, whilst for the seats the number people is less variable (usually one person per chair). The occupancy of the seats is therefore easier to detect. In the railway museum test setting the number of Wi-Fi devices in a train compartment can be estimated using Wi-Fi localization with a false negative error of 10-15% and without a false positive error during a train journey of about three minutes. In the test setting the test subject were all instructed to bring one Wi-Fi device with its Wi-Fi enabled. Therefore the relation between the number of Wi-Fi devices and the number of Wi-Fi probes can be researched and not the relation to the number of passengers. It is therefore also hard to statistically determine the extent to which these two approaches can supplement each other based on the tests of this research. It seems however that camera-based localization is more accurate as it detects people instead of Wi-Fi devices. Even though the camera-based localization appears more accurate, it seems likely that the systems can complement each other during a train journey by mitigating each other’s disadvantages. The Wi-Fi localization is probably more accurate the more passengers there are in compartment. This is because it seems likely an expected ratio between Wi-Fi devices and passengers becomes more reliable the more passengers there are in a train. This is in contrast to the camera-based localization which may become less accurate when there are more passengers in a train compartment than the number of available seats. This because the number of passengers standing in the hallways is hard to detect using camerabased localization and because the camera view of train chairs may be blocked by passengers standing in the hallway. It can thus be stated that the Wi-Fi and camera-based localization may complement each other when used to measure occupancy in the train because they can be used to verify each other and they both thrive during different amounts of occupancy. Based on the test results is seems that a combination of Wi-Fi and camera-based localization is suitable to measure occupancy in the train, but the test results should be verified by testing the proposed methods in a real train environment.