MoCap Trajectory-Based Animation Synthesis and Perplexity Driven Compression

Abstract

We design a 3D Motion Capture Animation Synthesis and Compression pipeline that allows reproducing people’s movement in a 3D environment and compressing it at ultra-low bitrates. The method deploys a stage-wise strategy. A surveillance video is used as the input to obtain the movement trajectory by adopting an object detection algorithm. The acquired trajectory is lifted to 3D space by spatial transformations. Based on the reference MoCap animation, an adaptive animation synthesis algorithm processes the input through position and rotation correction; frame interpolation and deletion; trajectory smoothing. Finally, a perceptually adaptive compression algorithm driven by perplexity is applied for MoCap animation compression. Experimental results demonstrate that our animation synthesis method eliminates artifacts, such as foot-sliding, while ensuring accurate position and smooth transition. Our proposed compression algorithm can achieve results that are perceptually similar compared to the PCA-based benchmark, while requiring only about one-tenth of the storage space.