复杂监控背景下基于边缘感知学习网络的行为识别算法

BEHAVIOR RECOGNITION ALGORITHM BASED ON EDGE-AWARE LEARNING NETWORK IN COMPLEX SURVEILLANCE BACKGROUND

由于复杂背景、多视角变化等因素的影响,准确识别、分析现实场景中人体的行为仍然是一个具有挑战性的问题。为了提升行人检测与行为识别的精度,提出一种新颖的边缘感知深度网络。通过边缘感知融合模块提升行人轮廓精度,利用多尺度金字塔池化层捕获视频序列的空时特征。边缘相关特征的互补特征能够有效地保留行人目标的清晰边界,而辅助旁侧输出与金字塔池化层输出的组合可以提取丰富的全局空时上下文信息。大量定性定量的实验结果表明,该模型可以有效地提高现有行人检测与行为识别网络的性能,在UCF101数据集上取得了90.55%的行人行为识别准确率。

Due to the influence of complex background and multi-angle changes,it is still a challenging problem to accurately identify and analyze human behaviors in real scenes.In order to improve the accuracy of pedestrian detection and behavior recognition,this paper proposes a novel edge-aware deep network method.It used the edge-aware fusion module to improve the accuracy of pedestrian contours,and used the multi-scale pyramid pooling layer to capture the space-time features of the video sequence.The complementary features of the edge-aware features could effectively preserve the clear boundary,while the combination of the auxiliary side output and the pyramid pooling layer output could extract rich global context information.A large number of qualitative and quantitative experimental results show that the proposed model can effectively improve the performance of existing pedestrian detection and behavior recognition networks.On the UCF101 data set,it achieves 90.55% accuracy rate of pedestrian behavior recognition.