Analysis and Synthesis of Traffic Scenes from Road Image Sequences

Abstract

Traffic scene construction and simulation has been a hot topic in the community of intelligent transportation systems. In this paper, we propose a novel framework for the analysis and synthesis of traffic elements from road image sequences. The proposed framework is composed of three stages: traffic elements detection, road scene inpainting, and road scene reconstruction. First, a new bidirectional single shot multi-box detector (BiSSD) method is designed with a global context attention mechanism for traffic elements detection. After the detection of traffic elements, an unsupervised CycleGAN is applied to inpaint the occlusion regions with optical flow. The high-quality inpainting images are then obtained by the proposed image inpainting algorithm. Finally, a traffic scene simulation method is developed by integrating the foreground and background elements of traffic scenes. The extensive experiments and comparisons demonstrate the effectiveness of the proposed framework.

Transformer-Guided Video Inpainting Algorithm Based on Local Spatial-Temporal joint

“…Moreover, in the process of video shooting, uncontrollable factors such as external noise and object occlusion in the video often occur, resulting in missing regions in the video. In addition, it also plays an important role in video surveillance [19,20], remote sensing satellites [21,22] and other fields. In the task of plant protection, the remote sensing satellite image video information can monitor the condition of plants and prevent pests from eating.…”

A Novel Vehicle Detection Framework Based on Parallel Vision

“…By using the stochastic grammar, Jiang et al [24] propose a learning-based pipeline to automatically generate and render indoor scenes with the physics-based rendering. With the foreground and background elements of traffic scenes, Yuan et al [25] construct the traffic scenes by the scene model.…”