DAGMaR: A DAG-based Robust Road Membership Estimation Framework for Scenario Mining

Scenario Mining in the Urban Domain: Exploiting the Topology of a Road Network for Maneuver Annotation and Scenario Extraction

“…However, the approaches typically differ in that they look at traffic and the road network from either a microscopic or macroscopic point of view. For the latter, the road network's level-of-detail is typically lower than the former one or not relevant for the use-case at hand so that intersections are typically left out, and the network only contains interconnecting roads [9], [6].…”

“…This work uses the deviation between the measured object pose x and all of its mapped poses P x to estimate the road mapping likelihood. In particular, the euclidean distance d(x, p) and the orientation deviation ∆ω(x, p) ∈ (−π, π] between the pose and its mapped siblings is used as in a previous work [6]. Hence L(w | p) is defined as…”

“…A data-driven approach using clustering methods is proposed to annotate roads with primitive routing maneuvers w.r.t their shape. Furthermore, a route backtracking approach is proposed based on map-matching to estimate a road user's traveled route on intersections by extending a previously [6] proposed method. The semantically enriched road network and the backtracked route enable inferring a road user's maneuver robustly even in complex mixed traffic environments with various routes and thus maneuver candidates.…”

Scenario Mining in the Urban Domain: Exploiting the Topology of a Road Network for Maneuver Annotation and Scenario Extraction

“…with |X| denoting the number of poses in the trajectory and r as the most likely route. To solve the stated problem in (2), i.e., finding the most likely route a traffic participant follows on an intersection, this work employs a Directed Acyclic Graph of Mapped Roads (DAGMaR) presented in a previous work [6]. That is, by iteratively building up a Directed Acyclic Graph (DAG) with roads R I of the intersection I according to the current state of the traffic participant x ∈ X, the likelihood being mapped on the roads L(w 1 | x), L(w 2 | x), .…”

Identification of Lane-Change Maneuvers in Real-World Drivings With Hidden Markov Model and Dynamic Time Warping

“…To qualitatively evaluate the performance of the presented approach, the lane-change classification is treated as a binary classification problem. For that purpose, the approach used in a previous work [20] is adopted to state whether a lanechange was correctly classified or not. Let t c,i = [t s c , t e c ] denote the ith interval represented as start and end time of the lane-change identified by a classifier c with c ∈ {ANN, SVM, HMM+DTW, HMM+DTW-Ex} and t g = [t s g , t e g ] the manually identified start and end times.…”