Classification of human poses for the purposes of automating time studies as well as object counting.

The project will three stages:Human Pose Estimation Baseline. Benchmark state-of-the-art human pose estimation algorithms, such as OpenPose and Mask R-CNN, on RIOS video data. Repetitive action detection and counting. Develop algorithms for using estimated and tracked human-pose to segment repetitive actions. Object counting and throughput analysis. Develop algorithms for detecting and counting multiple object instances of a single category/type in video data to determine the number of unique instances processed per unit time. and must deal with partial occlusions, viewpoint variation and lighting variation.At the completion of each stage a short report and software source code will be delivered to RIOS. The source code may include open-source packages. The code should be considered research quality not commercial quality.