João G. P. Rodrigues

A mobile sensing architecture for massive urban scanning

Mobile Sensor Networks based on connected vehicles and smart phones are poised to become key enablers in the development of sustainable and intelligent transportation systems in urban environments. By gathering and processing massive amounts of data in real-time, this form of information and communication infrastructure can be instrumental towards improving traffic flow, reducing carbon emissions and promoting multi-modal mobility and enhanced coordination among public transit systems. We propose a system architecture for a Massive Multi-Sensor Urban Scanner capable of acquiring large quantities of real-time information from a vast variety of sources and sending the data to a back-end data processing cloud using multiple communication interfaces. Requirements, technical challenges, design choices and first results are explained in detail based on a prototype that is currently being deployed in Porto, Portugal.

Mining geographic data for fuel consumption estimation

Mobility is one of the greatest contributors to the personal carbon footprint and to pollution and noise in urban areas. Still, these factors are not yet easily quantifiable in personal or urban scale, e.g. impact of each car trip or areas most exposed to CO2 emissions. In this article, we propose an innovative solution for estimating fuel consumption and emissions leveraging the opportunities generated by the ubiquitous availability of mobile devices. We collect a large data set of GPS and fuel consumption data crowd-sourced by volunteer participants with an Android mobile application that logs the smartphones embedded GPS data and gathers vehicle data using an external On-Board Diagnostics (OBD) device. This data is used to develop a model that estimates the instantaneous fuel consumption from the smartphones GPS data alone, using the OBD data as ground truth. We use speed, acceleration and steepness as predictor variables to train polynomial models with and without cross-product terms. With the best general model (trained and tested on all participant vehicles), we obtain an average residual standard deviation of 1.58 l/100km for average consumption on 1min intervals. For individual models (trained and tested on each participant vehicle), we obtain an average residual standard deviation of 1.43 l/100km. The average fuel consumption for the used data set was 6.7 l/100km.

A non-intrusive multi-sensor system for characterizing driver behavior

Understanding driver behavior is critical towards ensuring superior levels of safety and environmental sustainability in intelligent transportation systems. Existing solutions for vital sign extraction are generally intrusive in that they affect the comfort of the driver and may consequently lead to biased observations. Moreover, low-complexity devices such as GPS receivers and the multitude of sensors present in the vehicle are yet to be exploited to the full extent of their capabilities. We present a real-life system that combines wearable non-intrusive heart wave monitors with a wireless enabled computing platform capable of gathering and processing the data streams of multiple in-vehicle sources. Observed variables include electrocardiogram, vehicle location, speed, acceleration, fuel consumption, and pedal position, among others. Preliminary results show that the proposed system is well suited not only for characterizing driver behavior but also for identifying and mapping potentially dangerous road segments and intersections.

FPGA-based rectification of stereo images

In order to obtain depth information about a scene in computer vision, one needs to process pairs of stereo images. The calculation of dense depth maps in real-time is computationally challenging as it requires searching for matches between objects in both images. The task is significantly simplified if the images are rectified, a process which horizontally aligns the objects in both images.

Opportunistic mobile crowdsensing for gathering mobility information: Lessons learned

SenseMyCity is an opportunistic mobile crowdsensing tool available for researchers to design and implement data collection campaigns for studying large-scale processes. We discuss how our tool addresses two critical aspects of large scale data gathering campaigns: compliance with data protection guidelines and low intrusiveness. The latter is achieved by low battery consumption and low interaction with the participants, which lead us to implement an algorithm to automatically start gathering data when participants move (and stopping when they stop). Then, we report on three interdisciplinary data collection case studies on the wild in the city of Porto involving overall 641 participants, reflecting on the participant engagement mechanisms and the results of the data collection campaign. Finally, we report on the most common timestamping and position mismatches observed in the data collected in such uncontrolled scenarios from a wide variety of hardware and software versions, and that can impact data analysis.

Impact of Crowdsourced Data Quality on Travel Pattern Estimation

Mobile crowdsensing can provide mobility researchers with fine grained spatio-temporal location data. But crowdsourcing impacts data quality both due to device and OS heterogeneity, and to annotation errors. Additionally, it is often necessary to deal with multimodality, i.e. participants using different travel modes often in the same trip. In this paper, we address how to draw value from a crowdsensed dataset for characterising mobility demand through origin-destination (OD) matrices, highlighting challenges and providing some solutions. First, we identify typical errors in heterogeneous location data, propose and compare methods to automatically improve data quality. Then, we devise a method to detect among 5 transport modes (walk, car, bus, metro, bike) offline a posteriori. We segment trips on stopped periods and propose a random forest model to detect transportation mode per segment using only location data. Our results show that with adequate pre-processing and robust features, an RF classifier is able to achieve accuracy and precision of 85% in trip segments. This is similar to the literature, but our work uses a very heterogeneous crowdsourced trajectory dataset when compared to the others. Finally, we quantify the impact of the model on mulit-modal OD matrices and whole trip characterisation. We can correctly identify used transportation modes accurately, but the precision is impaired by the high likelihood of at least one false positive in the whole trip.

Map-Aided Dead-Reckoning Using Only Measurements of Speed

We present a particle-based framework for estimating the position of a vehicle using map information and measurements of speed. The filter propagates the particles’ position estimates by means of dead-reckoning, and then updates the particle weights using two measurement functions. The first measurement function is based on the assumption that the lateral force on the vehicle does not exceed critical limits derived from physical constraints. The second is based on the assumption that the driver approaches a target speed derived from the speed limits along the upcoming trajectory. Assuming some prior knowledge of the initial position, performance evaluations of the proposed method indicate that end destinations often can be estimated with an accuracy in the order of

SenseMyCity: Crowdsourcing an Urban Sensor.

100\,[\mathrm{m}]

A Mobile Sensing Approach to Stress Detection and Memory Activation for Public Bus Drivers

. These results expose the sensitivity and commercial value of speed data collected in many of todays insurance telematics programs, where the data is used to adjust premiums and provide driver feedback. We end by discussing the strengths and weaknesses of different methods for anonymization and privacy preservation in telematics programs.