[Retracted] A Way to Automatically Generate Lane Level Traffic Data from Video in the Intersections

Table 1

Video-based vehicle trajectory extraction methods.

Work	Key method	Characteristic
Vehicle detection	Motion-based [13]	They are restricted to the constant brightness assumption, resulting in the incomplete detection of moving objects
	Model-based (R-CNN [14], Faster-CNN [5], SSD [6], YOLO [7]	They need large training datasets, complicated designs of deep convolutional network structures, and long training times to achieve optimal performance for targeted scenes
Vehicle tracking	Kalman filter (KF) [15]	Tracked objects can still drift over time due to the error accumulation from the first to the last frame
	CNN-based trackers [16]	It needs a large amount of training data, the significant manual-processing efforts in preparing the training datasets