Cyber Mobility Mirror: A Deep Learning-Based Real-World Object Perception Platform Using Roadside LiDAR

Infrastructure-Based Vehicle Localization through Camera Calibration for I2V Communication Warning

“…As shown in Table 1 , the false negative rate is higher than the false positive one, similar to what is obtained in [ 19 , 21 ], meaning that there is a higher tendency to miss the detection rather than creating false detections. Moreover, when the error is there it may last for a relatively long time.…”

“…From the table, it can be seen how the false positive rate is larger than the one obtained in the multiple cameras approach, while it is lower than the LiDAR-based system. On the other hand, the percentage of false negatives is much lower than [ 21 ] and still comparable to [ 19 ].…”

“…As mentioned in Section 2 , in the literature there is a relevant amount of studies regarding roadside vehicle localization. In Table 2 , a comparison in terms of detection reliability for the proposed method with respect to different and more expensive technologies such as multiple fish-eye and thermal cameras [ 19 ] and single LiDAR [ 21 ] installed at the intersection is reported.…”

“…The work in [ 15 ], similar to the present study, adopts a single RGB camera to reconstruct the vehicle position in the 2D world coordinate system from the 2D camera coordinates by optimizing the transformation matrix based on the acquired data. In [ 19 ], a detection system is proposed fusing fish-eye and thermal cameras to localize vehicles at the intersection, while the authors in [ 21 ] present a localization method based on a single LiDAR installed on a pole at an urban intersection. A multiple LiDARs method is investigated in [ 22 ], showing the capabilities of a system combining IMM and JPDA for vehicle position and velocity estimation.…”

“…As far as cooperative LiDAR-based localization algorithms, Ref. [ 21 ] proposes a real-world object perception platform using a single LiDAR installed on a pole in correspondence of an urban intersection. The detection model used is trained on already available datasets collected based on vehicle-equipped LiDAR.…”