Detecting Cruising Flagged Taxis´ Passenger-Refusal Behaviors Using Traffic Data and Crowdsourcing

With the development of urban traffic, taxi accounts for a large proportion to serve people´s daily travel. Besides prebooked taxis, flagged taxis also run in many places of the world, allowing passengers to wave at the driver on the side of the road to flag down the taxi as it is approaching. For flagged taxis service, taxi drivers´ passenger-refusal (TPR) behaviors have become a serious problem due to rush hours, bad weather and destination choosy, which hurt the living quality and the reputation of the city. With more and more GPS-equipped taxis and smart phones available, detecting such abnormal TPR behaviors from GPS trajectories has become feasible. In this paper, we model the TPR behaviors and develop a system named Crowd TPR to detect TPR behaviors in real time by combining traffic data and crowd sourcing. It efficiently serves real-time monitoring requests from the traffic administrators and generates proper pick-up spots for passengers to select. In our system, we firstly propose a dynamic grid granularity selection method to achieve efficient map-matching and build spatio-temporal index on the road network. Through modeling taxis´ routing behaviors, we predict arriving locations of taxis and recommend pick-up spots for passengers to push cluster-based human intelligent tasks (HITs). After passengers submit HIT results to Crowd TPR, we verify TPR behaviors by modeling the anomalous features. We build our system using real trajectory datasets generated by 33,000+ taxis and validate the system with extensive evaluations including over 1-month user study.