This paper proposes a new approach of automated toll gate passing of an automated driving vehicle. This approach enables the vehicleto select an optimal toll gate and to automaticallypass the toll gate by using object detection, 3D environment construction, virtual line generation, path planning and motion control. After designing the concept of the approach, some demonstrationsare conducted to prove it. This data-based scenarioshowsthat the proposed approach can not only perceive the environment well for this purpose but can also plan appropriate trajectories when encountering complex scene near toll plazas.
From functional point of view the proposed approach consists of three parts, which are perception, decision making, and motion control, as illustrated in Figure 1. When approaching a toll gate plaza, the first step is to perceive the environment. Both infrastructure, e.g. gates and traffic signs, and moving objects, e.g. vehicles and pedestrians, are detected by the dedicated detection function modules for diverse object classes. The detected vehicles are further categorized into vehicle queues. Afterwards, the simultaneous localization and mapping (SLAM) algorithm estimates the motion state of the ego vehicle and generatesa 3D sparse map with the localization of the ego vehicle into the environment. https://codeshoppy.com/shop/product/toll-gate-app/
The combination of object detection and SLAM constructs anenvironment in which drivable areas are derived. In the next part, i.e. the decision making phase, traffic signs are further understood and all toll gates are scanned in order to detect passable ETC toll gates out of themwhich are marked as so called candidate gates. Only for these candidate virtual lanes are generated from the
current location of the ego vehicle as starting point and with the gate entry points of candidate gates as the ending points, with the same number of virtual lanes as candidate gates. From all the virtual lanes, the lane with the shortest vehicle queue and minimum lateral distance in total is then selected as the optimal virtual lane. Finally, a collision-free trajectory by considering drivable areas is permanently calculated and transmitted to the motion control module in order to maneuver the automated driving vehicle to the target toll gate.
The output of this step is the best trajectory the vehicle should go within the optimal virtual lane.
A. Gate Scanning After the 3D environment is established, gate scanning is used to detect all the toll gate and filter out all the passable ETC gateswhile discarding unpassable manual gate. Because of the possibility of tollgate occluded by trucks or buses, it could take a while to get all the gates scanned and all the ETC gates determined
B. Virtual lane Generation This module generates the virtual lanes from the positionof ego vehicle to all determined ETC gates, first time carried out when arriving at the toll gate plaza without lane markings.Hereby the road is transformed into a Cartesian coordinate system. Each point on the road has a coordinate. The position of ego vehicle is the coordinate origin, as shown in Figure 4. If the number of ETC gates is N.
C. Lane SelectionThis module is used to select the optimal virtual lane from several virtual lanes to the candidate gates.In order to find the best solution, two factors need to be considered, the lateral distance to each candidate gate and the length of the vehicle queue of each candidate gate. Particle filter  is used to estimate the probability of each factor for the reason that it is based on Monte-Carlo methods, which can be used in any form of state space model. Large probability means a high chance to choose corresponding candidate gate.For 1, 2, 3, …, kdenote each candidate gate. The probability vector N( (1),(2),…,( ))nnnppp kPand Ll( (1), (2),…, ( ))llppp kPdenote the particle filter result of lateral distanceand the particle filter result of vehicle queue length respectively. The finaldecision probability vector Pis combined of PNand PLby Yager formula, which is an information fusion algorithm.In our approach, the maximum probability in Pis chosen to select the final ETC gate
D. Collision AvoidanceThis functional module is used to permanently monitor road users in the surroundings if they could bring danger to the ego vehicle. The determined free space exclusive area of dangers is used for path planning in the next step. Click Here
E. Path PlanningThis module is used to step by step plan a collision-free trajectory for the motion control.The optimal virtual lane is transformed into S-L coordinate system, shown in Figure 5. Sis the direction of the path, while Lis the vertical direction of S. The whole virtual lineis segmented into a series of cells. Each cell has a cost. The cell with small value means it is suitable for driving, while the cell with large value means it is close to the edge of the virtual lane or is closed to the vehicles and pedestrians, which is not appropriate for driving.