Sequential Reliable-Inference for Rapid Detection of Human Actions

We present a probabilistic reliable-inference framework to address the issue of rapid-and-reliable detection of human actions. The approach determines the shortest video exposure needed for low-latency recognition by sequentially evaluating a series of posterior class ratios to find the earliest reliable decision point. Results are presented for a set of people walking, running, and standing at different styles and multiple viewpoints, and compared to an alternative ML approach.