Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Occluded Person Re-identification (1804.02792v3)

Published 9 Apr 2018 in cs.CV, cs.AI, and cs.MM

Abstract: Person re-identification (re-id) suffers from a serious occlusion problem when applied to crowded public places. In this paper, we propose to retrieve a full-body person image by using a person image with occlusions. This differs significantly from the conventional person re-id problem where it is assumed that person images are detected without any occlusion. We thus call this new problem the occluded person re-identitification. To address this new problem, we propose a novel Attention Framework of Person Body (AFPB) based on deep learning, consisting of 1) an Occlusion Simulator (OS) which automatically generates artificial occlusions for full-body person images, and 2) multi-task losses that force the neural network not only to discriminate a person's identity but also to determine whether a sample is from the occluded data distribution or the full-body data distribution. Experiments on a new occluded person re-id dataset and three existing benchmarks modified to include full-body person images and occluded person images show the superiority of the proposed method.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Jiaxuan Zhuo (2 papers)
  2. Zeyu Chen (48 papers)
  3. Jianhuang Lai (43 papers)
  4. Guangcong Wang (25 papers)
Citations (196)

Summary

Analysis of Occluded Person Re-identification with Deep Learning Techniques

The paper "Occluded Person Re-identification" addresses a critical problem prevalent in security applications: the identification of persons occluded by other entities such as crowds or static objects in video surveillance scenarios. Typical person re-identification systems assume that subjects are fully visible, which limits their applicability in real-world scenarios where occlusion is common. The authors propose a novel approach, which they term Occluded Person Re-identification, designed to effectively handle such occlusions by utilizing partially occluded images to retrieve full-body person images.

Key Contributions

The paper introduces an innovative deep learning framework termed the Attention Framework of Person Body (AFPB). This framework employs two main strategies:

  1. Occlusion Simulator (OS): This component generates synthetic occluded images by introducing artificial occlusions into full-body person images, thereby simulating diverse real-world scenarios where occlusions may occur.
  2. Multi-task Losses: The model leverages a combination of identification loss and occluded/non-occluded binary classification (OBC) loss, which simultaneously focuses on correctly classifying the identity of persons and distinguishing between occluded and non-occluded states.

Empirical Evaluation

Experiments are conducted using a newly developed dataset, Occluded-REID, alongside modified existing benchmarks P-DukeMTMC-reID and P-ETHZ. These datasets incorporate real-world occlusions and aim to provide a comprehensive evaluation ground for the proposed algorithm. The experimental results demonstrate a marked improvement in re-identification accuracy over baseline methods and existing frameworks, confirming the AFPB's efficacy in tackling occlusions.

Variation in performance across different occlusion complexities is also analyzed, with the proposed AFPB framework consistently outperforming alternatives. Notably, the single-shot and multi-shot experimental configurations show increased robustness, underscoring the flexibility and viability of the deep learning approach in dynamic surveillance environments.

Methodological Insights

The authors' approach implicitly integrates a form of attentional mechanism into the learned representations. By simulating different occlusion scenarios through the OS, the neural network is encouraged to focus on discerning the underlying identity of occluded subjects. This attentional focus is further reinforced by a novel loss function combination, which pragmatically balances between person identification and occlusion classification.

The framework's ability to encode prior knowledge of occlusion into the learned model contributes meaningfully to the theoretical advancement of person re-identification tasks under realistic conditions. Additionally, the proposed method exhibits practical applicability given its ease of implementation within existing neural network infrastructures such as Caffe.

Future Directions

The promising results open avenues for enhancing the breadth of neural network applications in surveillance contexts. Considering the success of the AFPB framework’s attention-based mechanism, further exploration into advanced attentional techniques could augment occlusion handling. Moreover, continued refinement of the Occlusion Simulator to better mimic diverse and complex real-world occlusion scenarios will be critical in refining the model's capabilities.

Additionally, integrating temporal dynamics to exploit sequence-based information potentially offers a new direction, enriching spatial and appearance features. Developing collaborative systems that refine person identification through multiple sensing modalities could also yield considerable advancements in enhancing recognition robustness amidst occlusions.

In conclusion, the paper convincingly contributes a proficient framework for handling occlusions in person re-identification tasks, leveraging the strengths of deep learning. The empirical validity demonstrated across multiple datasets corroborates its practical potential, paving the way for robust security solutions in crowded and public environments.