SBIR Phase I: Auto-Tracking Using Trailing Templates and Skeletal Guides

来自Mostert Group的这个小型企业创新研究第一阶段项目提出开发一种方法来跟踪运动中的动物和人类，特别是在体育赛事中，运动的特点是在相对较短的时间内重复运动。 提出了一种不需要放置视觉目标的生物运动跟踪算法。 虽然所提出的算法将在许多其他领域中应用，但Mostert Group最初专注于在实际体育赛事中在固定距离内实时跟踪跑步对象。 重复运动将被开发机制，学习后续的搜索区域的约束模板匹配。 跟踪位置和速度的小型数据库（从先前的运动序列导出）将允许算法推广到在同一路线上移动的新对象。 主要研究者Paul Mostert创建了一种尾随模板方法，该方法将与Zhong、Jain和Dubuisson-Jolly的可变形模板方法相结合，并与骨骼模型结合使用，该骨骼模型将通过潜在的混淆关系（例如，腿的交叉运动）。 第一阶段研究的主要目标是初步开发图形用户界面，然后开发两种关键的跟踪技术;（1）预测算法，用于有效地引导对视频帧内的对象特征的下一位置的搜索，及（2）一名模特儿─的方法来变形的图像模板，使用骨架指南，以提高跟踪生物运动的鲁棒性。该软件的应用Mostert集团提供的产品有现成的市场需求。 目前商业化的生物运动跟踪技术需要在主体的关键位置放置侵入式控制目标。 将开发跟踪和表征一般生物运动的商业需求，包括动物行为分析工具，以及预测和提高运动员的运动效率。 一个最初的垂直市场，用于获得已知对赛马未来表现有重要意义的统计指标，具有相当大的潜力。

This Small Business Innovative Research Phase I project from the Mostert Group proposes to develop a method to track animals and humans in motion, particularly in athletic events, where movement can be characterized by repetitive motions over a relatively short period of time. An efficient algorithm for tracking of biological motion through cluttered backgrounds and significant self-occlusion that does not require the placement of visual targets, is proposed. While the proposed algorithms will have application in many other domains, the Mostert Group initially focuses on real-time tracking of running subjects over a fixed distance in actual athletic events. Repetitive motions will be exploited by developing mechanisms to learn subsequent search regions for constrained template matching. A small database of tracked positions and velocities (derived from previous motion sequences) will allow the algorithm to generalize to new subjects moving over the same course. The Principal Investigator, Paul Mostert, has created a trailing template method that will be combined with the method of deformable templates of Zhong, Jain, and Dubuisson-Jolly, and used in conjunction with skeletal models that will guide the deformation of the ZJDJ templates through potentially confusing relationships (e.g., the crossover motion of the legs). The primary objectives of Phase I research will be a preliminary development of a graphical user interface followed by that of two key tracking technologies; (1) a predictive algorithm to efficiently guide the search for the next position of the object features within video frames, and (2) a model-based approach to deformation of image templates using skeletal guides to improve tracking robustness of biological motion.Applications for this software proffered by the Mostert Group have a ready market demand. Present commercial tracking technology of biological motion requires the placement of intrusive control targets at critical positions on the subject. The commercial need for tracking and characterizing general biological motion will be exploited, including tools for animal behavior analysis, and predicting and improving motion efficiency in athletes. An initial vertical market for obtaining statistical measures known to be significant to the future performance of a racehorse has considerable potential.