Focus on the Visible Regions: Semantic-Guided Alignment Model for Occluded Person Re-Identification

Abstract

The occlusion problem is very common in pedestrian retrieval scenarios. When persons are occluded by various obstacles, the noise caused by the occluded area greatly affects the retrieval results. However, many previous pedestrian re-identification (Re-ID) methods ignore this problem. To solve it, we propose a semantic-guided alignment model that uses image semantic information to separate useful information from occlusion noise. In the image preprocessing phase, we use a human semantic parsing network to generate probability maps. These maps show which regions of images are occluded, and the model automatically crops images to preserve the visible parts. In the construction phase, we fuse the probability maps with the global features of the image, and semantic information guides the model to focus on visible human regions and extract local features. During the matching process, we propose a measurement strategy that only calculates the distance of public areas (visible human areas on both images) between images, thereby suppressing the spatial misalignment caused by non-public areas. Experimental results on a series of public datasets confirm that our method outperforms previous occluded Re-ID methods, and it achieves top performance in the holistic Re-ID problem.

Bi‐level deep mutual learning assisted multi‐task network for occluded person re‐identification

“…In [13], a dual attention‐based occluded person ReID method was presented. In [14], an alignment model guided by semantics was proposed, where the semantic information was obtained by an existing human semantic parsing network. In [31], the authors proposed a pose‐guided visible part matching method, which included an attention module and a module for visibility prediction.…”

Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-Identification

“…However, the positioning accuracy of this method is not high and the performance improvement is limited. To make further use of the idea of partitioning, a large number of methods introduce existing pose-estimation models to assist in locating pedestrians and find segmentation containing visible areas of pedestrians to match [8,10,11], thus improving the effective utilization of part features. In addition, HOReID [12] used landmarks generated by the pose estimation model combined with graph convolution to improve performance further.…”

Pedestrian Detection by Novel Axis-Line Representation and Regression Pattern

“…In particular, the idea of ALR can be introduced into both anchor-free and anchorbased methods. 2 We propose a deformable convolution base-offset initialization strategy towards a more aligned receptive field, and further improvement of detection performance by forcing the aspect ratio of the deformable convolution kernel close to the pedestrian aspect ratio. 3 Several experiments are carried out on two benchmark datasets (the Caltech-USA and the CityPersons) to demonstrate the effectiveness and generalization of the proposed ALR pattern in both anchor-free and anchor-based methods.…”

“…Pedestrian detection is a necessary prerequisite and key component of recent research hotspots (e.g., pedestrian reidentification [ 1 , 2 , 3 ], human pose estimation [ 4 ]), for these tasks it is necessary to detect all the existing pedestrians accurately from images or videos before they go to the next step. In engineering fields, pedestrian detection is also an urgent need in the Advanced Driving Assistance System (ADAS) to help to reduce the occurrence of people-vehicle collisions, or in smart buildings for air conditioner control and monitoring systems [ 5 ].…”