Traffic scene analysis based on combination of superpixel-based and multiple object detection
Pei Liu, Xue Yuan, Liping Zhang, Hansong Li · 2018
With the development of auto-driving and vehicle-assisted driving, traffic image parsing is becoming more and more important. Generally, superpixel-based approach is used in image parsing. But the parsing accuracy of this method is not high enough. To solve this problem, a new mathed is proposed, which combines the superpixel-based approach and multiple object detection. Superpixel-based approach is used to label background scene (e.g., roads, sky, trees, buildings) and multiple object detection mainly determines target class scene (e.g., pedestrians, vehicles, non-motor vehicles). The parsed image is generated by combining the two algorithms results. With this method, the accuracy increases 3%.