In the INRIX 2018 Global Traffic Scorecard, Ottawa was ranked the 68th most congested city in the world. In addition to congestion’s negative impacts on the environment, it also impacts businesses that rely on roads for the transportation of goods and services. With the main contributor of congestion being traceable to the vast (and growing) number of individual cars on the road, the question remains on the best way to reduce personal vehicles and encourage transit usage.
It is widely understood that public transit is the most important and affordable mode of travel that will reduce traffic congestion and make efficient use of existing roads. But the negative stigma surrounding public transportation being crowded, unreliable, and unpredictable still stands. To encourage people to use public transportation and to switch from private vehicles, this mode of transportation must be well-designed, reliable, and comfortable.
Even with the arguments against public transit, the benefits are clear and growing in importance. Per passenger, a bus carrying 40+ people takes up much less room on the road than a person driving them self to work, as the majority of society does. Also, personal cars are expensive, and cutting transit investment in favor of car infrastructure leaves low-income citizens to suffer crowded commutes and infrequent service. Lastly, the climate argument that buses are more emission-friendly than single-occupancy cars should be enticing enough, with region climate goals becoming more regulated and with urgency increasing.
Overall, public transit agencies need to make optimal operational decisions to increase the efficiency of their services. Such decisions rely on estimating the number of passengers, identifying their origins and destinations, and optimizing the travel cost.
With all of this in mind, our objective with this study was to capture the travel route and origin-destination information along the OC Transpo Bus in Ottawa and use the information to better manage and optimize Ottawa transit .
SMATS innovative approach for solving the traffic congestion problem is based on using Wi-Fi and Bluetooth sensors for route travel times, as well as passengers’ origins and destinations. For this case, specifically, SMATS TrafficBox™ portable sensor was placed in the bus for capturing and storing signals from mobile devices. The TrafficBox™ is also equipped with a GPS module, a battery, and a processing unit.
The SMATS TrafficBoxTM captures passengers’ devices’ Wi-Fi and Bluetooth MAC addresses passively and anonymously. It also stores the detecting signals’ timestamps, their Received Signal Strength Indicator(RSSI), and the sensor location. TrafficBoxTM can detect 3 types of Bluetooth signals: Bluetooth Connected, Bluetooth Discovery, and Bluetooth LE as well as WiFi signals.
The primary challenge in using traffic sensors is that they collect not only signals of passengers’ device but also all the signals which are within the range of their antennas. Therefore, the sensors can receive signals which could be transmitted from mobile devices of other vehicles, pedestrians, or near buildings. Luckily SMATS iNode can distinguish passengers’ signals from non-passengers and estimate their Origin-Destination by using various parameters and advanced filtering algorithms.
Traditional methods of transit passengers’ origin-destination data gathering stem from field-based studies such as observational surveys and passenger interviews. Observational surveys are conducted to understand the behavior of subjects without any specific response from the subjects, they are merely observed as they perform normal activities such as getting on or off a bus at a certain stop. Observational surveys measure the system as it currently exists, but most times, it is necessary to understand the changes in travel behavior due to changes in the operating systems.
Another commonly used method is passenger interviews/surveys, which need interviewers physically present to choose the individual(s) and pose the question(s). Unfortunately, in many situations, the application of in-person interviews is not practical, and surveyors are unconsciously subject to ‘convenience-based sampling’. With in-person sampling, there is a high probability that the surveyors will be attracted by people who are friendly and/or are similar to the surveyor, leading to unwanted bias in the data collection.
Field-Based transit studies should ideally be conducted on as many stops along the route as possible. However, manpower and cost limitations for the surveys restrict the number of stops that can be covered by the survey. With this in mind, traditional methods of traffic studies are insufficient for efficient and cost-effective transit planning and optimization.
A key factor in this study was user ability to choose specific sensor monitoring day(s) and time(s) for relevant data capture. In structuring the study, it was apparent that people are using public transit more during rush hours, so we could collect data during these peak times to obtain the largest amount and most relevant (to congestion) data. Therefore, the study determined the peak time was during evening rush hour from 4:00 PM to 5:30 PM and that period was monitored for six days. Among different urban routes along the Ottawa Transpo line, route 87-Baseline was selected because it serves downtown Ottawa and has plenty of bus stops.
The TrafficBoxTM sensor was placed at the middle of the bus for each day to capture Wi-Fi and Bluetooth signals and GPS data with a 1-second resolution. The sensor captured device signals, and the raw data was automatically synced to the cloud in real-time. The raw data contain a considerable amount of noise, outliers, and other inconsistency because it captures all the signals which are within its antennas’ ranges. After cleaning the data set from outliers, we employed clustering techniques to separate passengers from non-passengers’ signals automatically. Therefore, the following features were extracted from the raw data set:
- The average of RSSI: Summation of all the RSSI values for each MAC address divided by its counts.
- The variance of RSSI: Variance of all the RSSI values for each MAC address.
- The number of detection: Number of all the records for each MAC address.
- Travel time: Difference between the last and first detection time.
- Travel Distance: The route distance between entry and exit coordinates.
With the collected data and applying filtering algorithms, we were able to distinguish the signals of passengers from non-passengers and determine their Origin-Destination information by using GPS and stop location data. Comparing the passenger manual counts, an average sample rate of 20% was calculated. Also, by comparing ground-truth OD data and passengers’ cluster, passengers OD was estimated with around 90% accuracy.