基于强化学习的公共空间行人轨迹跟踪

对公共空间中的多目标行人轨迹跟踪问题，提出一种基于强化学习的多目标行人轨迹跟踪算法。首先采用高精确度的目标检测器检测公共空间视频中的行人目标，并为每个目标分配一个独立的单目标跟踪器进行轨迹跟踪；将每个目标作为独立智能体，通过深度强化学习方式进行训练；接下来结合跟踪轨迹与检测目标之间的表观和位置特征构建相似度代价矩阵；最终通过匈牙利算法实现数据关联。实验表明，在常用公开数据集上本文算法跟踪精确度达76.1%，表明算法对多目标轨迹跟踪的可行性与有效性。

Pedestrian trajectory tracking in public space based on reinforcement learning

Aiming at the multi-objective pedestrian trajectory tracking problem in public space, a multi-objective pedestrian trajectory tracking algorithm based on reinforcement learning is proposed. Firstly, a high-precision target detector is utilized to detect pedestrian targets in space, and an independent single-target tracker is assigned for each tracking target trajectory. Each target is trained as an agent by means of deep reinforcement learning, and combined with the appearance and position characteristics between tracking trajectory and detecting target, the tracking target is constructed. Similarity cost matrix is built to realize data association through Hungarian algorithm. Experiments show that the tracking accuracy of this algorithm is 76.1% on common open data sets. Good results have been achieved in multi-objective pedestrian trajectory tracking in public space.