基于视觉注意机制深度强化学习的行人检测方法

Deep reinforcement learning with visual attention for pedestrian detection

结合视觉注意机制,并用深度强化学习训练视点选择模型,模拟人类视觉搜索局部关键部位,提出了新的行人检测方法,通过视点选择模型生成聚焦图像,不断地叠加搜索关键区域,由检测网络对关键区域进行行人判别,并通过信息熵度量检测结果的可信度,作为奖赏通过深度强化学习优化视点选择模型。视点选择模型和检测网络协同迭代训练,使该方法具有很强的局部关键区域选择和判别能力,减少了形变和遮挡的影响。与经典的基于部位的行人检测方法在公开的行人检测数据集进行对比实验,结果表明,所提出的行人检测方法可以有效地提高行人检测精度。

The visual attention mechanism is introduced proposes A new pedestrian detection method is proposed,by introducing visual attention mechanism.Meanwhile,the deep reinforcement learning is applied to train the viewpoint selection model to simulate how human visual search local key parts.The method generates focused images by viewpoint selection model to search and overlay them together for key areas.The detecting network is used to identify pedestrian in these key areas.Then,information entropy is computed for measuring the reliability of the result and optimizes the viewpoint selection model as a reward for deep reinforcement learning.The collaborative iterative training with viewpoint selection model and detection network are integrated to improve the ability for searching and detecting local key areas,and reduce the influence for posture changing with body deformation and occlusion.The comparisons with classic part-based pedestrian detection method on public pedestrian detection data sets show that the proposed method can effectively improve the pedestrian detection accuracy.