Domain adaptive semantic segmentation aims to reduce domain shifts / discrepancies between source and target domains, improving the source domain model's generalization ability to the target domain. Recently, prototypical methods, which primarily use single-source or single-target domain prototypes as category centers to aggregate features from both domains, have achieved competitive performance in this task. However, due to large domain shifts, single-source domain prototypes have finite generalization ability and not all source domain knowledge is conducive to model generalization. Single-target domain prototypes are noisy because they are prematurely initialized with all features filtered by pseudo labels, which causes error accumulation in the prototypes. To address these issues, we proposes a covariance-aware cross-domain prototypes method (CACP) to achieve robust domain adaptation. We propose to use both domain prototypes to dynamically rectify pseudo labels in the target domain, effectively reducing the recognition difficulty of hard target domain samples and narrowing the gap between features of the same category in both domains. In addition, to further generalize the model to the target domain, we propose two modules based on covariance correlation, FSPC (Features Selection by Prototypes Covariances) and WSPC (Weighting Source by Prototypes Coefficients), to learn discriminative characteristics. FSPC selects highly correlated features to update target domain prototypes online, denoising and enhancing discriminativeness between categories. WSPC utilizes the correlation coefficients between target domain prototypes and source domain features to weight each point in the source domain, eliminating the information interference from the source domain. In particular, CACP achieves excellent performance on the GTA5 $\to$ Cityscapes and SYNTHIA $\to$ Cityscapes tasks with minimal computational resources and time.

In order to reduce the negative impact of severe occlusion in dense scenes on the performance degradation of the tracker, considering that the head is the highest and least occluded part of the pedestrian’s entire body, we propose a new multiobject tracking method for pedestrians in dense crowds combined with head tracking. For each frame of the video, a head tracker is first used to generate the pedestrians’ head movement tracklets, and the pedestrians’ whole body bounding boxes are detected at the same time. Secondly, the degree of association between the head bounding boxes and the whole body bounding boxes are calculated, and the Hungarian algorithm is used to match the above calculation results. Finally, according to the matching results, the head bounding boxes in the head tracklets are replaced with the whole body bounding boxes, and the whole body motion tracklets of the pedestrians in the dense scene are generated. Our method can be performed online, and experiments suggested that our method effectively reduces the negative effects of false negatives and false positives on the tracker caused by severe occlusion in dense scenes.

Remote photoplethysmography (rPPG) is a non-contact measurement method for estimating heart rate, which has great application potential in medical ward monitoring and remote physiological health monitoring. Existing methods for measuring remote physiological signals are susceptible to different degrees of light changes, head movements, and other conditions. In this paper, we design a multi-scale physiological signal feature extraction module (MPSFE) and a plug-and-play time difference global attention module (TDGAM). The MPSFE module extracts different degrees of 3-dimensional convolution of shallow features in the early stage. It obtains more robust semantic information while ensuring the complete position and detailed information to enhance the accuracy and generalization ability of the model. The TDGAM module combines temporal differential convolution with spatial attention and channel attention to enhance the network’s localization of the region of interest (ROI) in the case of light changes and head movement. We conduct many experiments on the three datasets, UBFC-rPPG, COHFACE, and PURE. The results shows that our modules significantly improves the accuracy of heart rate measurement.