Human pose estimation is a key technology in the field of human action recognition. It aims at recognizing human poses by extracting features in images or videos. At present, human body occlusion in multi-camera views is a major challenge for human body pose estimation. To tackle the problem, we proposed the MaxHeatmap Pose model. The model obtains high-quality heatmap estimation through the MH structure module, in order to locate the target person more accurately. Then the HumanPose regression network is proposed to estimate the detailed 3D human pose. The experimental results on the data set show that the performance of the model is better than previous methods.

Siamese network is widely used in object tracking. However, it suffers the problem of poor generalization. The existing tracker based on template update mechanism uses complex calculation strategies and time-consuming optimization to achieve good tracking performance, but it does not meet the requirements of real-time tracking. In this paper, we first propose a tracking framework based on space-time history frames, and use feature enhancement to enrich the features of history frames. Then, we propose EnhanceNet which is an offline trained network for performing online data augmentation. It can enhance the tracking accuracy while preserving high speeds of the state-of-the-art online learning. In challenging large-scale datasets, such as LaSOT, VOT2018 and OTB-2015, our framework is better than state-of-the-art real-time trackers and has achieved excellent results while running at 32 FPS.