Anomalous State Recognition of Lane-changing Behavior using a Hybrid Autoencoder Architecture

Fan, Pengcheng; Liu, Yangzexi; Guo, Jingqiu; Wang, Yibing; Qiu, Min

doi:10.1109/itsc45102.2020.9294394

Cited by 2 publications

(1 citation statement)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The calculation of Network2Matrix consists of three stages, as shown in Figure 3, where trajectories indicate the detector through which the vehicle passes; for example, l 52 , l 43 indicates that the vehicle passes through the detector l 52 and then passes through the detector l 43 . Primarily, locations are presented by 2D vectors, through loation embedding composed of word2vector and t-distributed stochastic neighbor embedding (t-SNE) (a nonlinear dimensionality reduction technique) ( 40 , 44 , 45 ). Then, discretization is completed by three steps: splitting, rejection, and shrinkage.…”

Section: Proposed Methods and Algorithmmentioning

confidence: 99%

Traffic Mosaic: Traffic Data Imputation Method Inspired by Image Super Resolution

Shuai

et al. 2022

Transportation Research Record: Journal of the Transportation R

View full text Add to dashboard Cite

Missing data is a major problem in data-driven intelligent transportation systems. This study presents an imputation method, called TrafficMosaic, for effective imputation of lost traffic flow data. If we consider traffic data as a image, then the missing data in it can be regarded as a mosaic. The design of TrafficMosaic is inspired by super resolution techniques. First, a deep neural network (DNN) is employed to recover the lost data by a suite of techniques inspired from image super resolution. This DNN method combines the temporal relevance of historical flow with spatial correlation of current fragmented flow data, and uses the imputation result as input for recursive calculation. Second, a data-driven road network matrixing algorithm is proposed to mine location relations from trajectory data and reconstruct a road network flow into matrixes of road network flow pictures. Convolutional local calculation in a convolutional neural network is introduced to extract local spatial features of the road network, thereby reducing the computational complexity and improving the generalization ability. The final step is to extract the local temporal features of the flow data. Traffic flows at adjacent moments have strong temporal characteristics, and we splice several historical road network traffic pictures in chronological order so that our model can capture temporal information from the sequence. The experimental evaluation used a real automatic number plate recognition dataset for experiments, and the results show the effectiveness of TrafficMosaic for lost data imputation.

show abstract