Inserting virtual pedestrians into pedestrian groups video with behavior consistency
The Visual Computer: International Journal of Computer Graphics
| Hi-index | 0.00 |
Using a single camera, we capture video of multiplepeople walking in an open area and insert a moving syntheticcharacter into the scene. The people are located in the video usingsimple image processing techniques. Simplifying assumptions andbasic projective geometry are used to locate the people in 3D. Weextract sufficient information to synthetically reconstruct the 3Dscene including the paths of individuals in the crowd. These pathsare also be used to plan the paths of new synthetic charactersavoiding collisions with the existing people. Lighting samplestaken from the real environment and modeled in the syntheticscene are used along with camera calibration specifications. Thenew synthetic characters are rendered in 2D and compositedinto the real video using transparency alpha values extractedby the algorithm. Because we reconstruct the scene in 3D, thenew synthetic character is occluded by, and occludes, the peoplein the original video. This is accomplished with minimal userinteraction.