基於車輛偵測器重建車流影片

Abstract

車輛偵測器通常設置於道路的地底下，用來檢測該路段的交通狀況。這些偵測器都是提供整體的信息，如平均流量，平均速度和車道佔用率。由於收集的資料都是以統計數據的方式呈現，使得一般民眾很難能夠馬上理解，因此我們藉由提供逼真的車流影片來將車輛偵測器的統計數據視覺化。也就是說，我們的系統收集道路監視器的畫面，以及相同位置的車輛偵測器數據，之後建立這兩種類型的數據之間的連接。由於車輛偵測器的分佈比道路監視器密得多，因此對於一些沒有道路監視器的路段，可以利用我們的系統，合成出相對應的車流影片，以在視覺上描繪出此路段的交通情況。為了要達到這個目的，我們建立車流影片中車輛的SIFT流動以及車輛偵測器的數據之間的回歸模型。最後，由車輛偵測器給定的數據，我們的系統將檢索出數個符合車輛偵測器數據的車流影片片段，並將他們無縫地縫合在一起，合成車流影片。這些評估和實驗結果證明了我們系統的可行性。

Vehicle detectors (VDs) are usually distributed on/under a road network to detect the traffic situation. These detectors provide global information such as flow, speed, and occupancy of vehicles. Given that the collected statistical data are difficult for citizens to interpret, we visualize the data by providing users with realistic traffic videos. That is, our system collects the surveillance videos and the VD data that represent the traffic situation of the same position. It then builds the connection between these two types of data. Considering the distribution of VDs is much denser than that of surveillance cameras, for those road segments with a VD but without a surveillance camera, one can utilize our system to synthesize videos to visually depict the traffic situations over there. To achieve this aim, we estimate vehicular SIFT flows from the video and apply a regression model to build the mapping between the SIFT flows and the VD data. After that, given by a VD dataset, our system retrieves the videos, which match the VD data, and seamlessly stitches them together to synthesize the traffic video. The evaluations and the experimental results demonstrate the feasibility of our system.