RETRACTED: Super-Resolution pedestrian re-identification method based on bidirectional generative adversarial...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Pedestrian re-identification (ReID) is vital for intelligent security but suffers from low-resolution images due to shooting distance, leading to detail loss and reduced performance.
• Traditional super-resolution methods often introduce artifacts and excessive sharpening, hindering effective pedestrian image reconstruction.

Purpose of the Study:

• To develop an advanced super-resolution pedestrian re-identification method that overcomes the limitations of existing techniques.
• To enhance the accuracy and robustness of pedestrian recognition in low-resolution surveillance scenarios.

Main Methods:

• Proposed a novel bidirectional adversarial network architecture integrating forward super-resolution reconstruction and backward downsampling simulation.
• Incorporated residual residual dense blocks and optimized the ESRGAN loss function for improved image realism and naturalness.
• Evaluated the BSRGAN ReiD method on public datasets (Urban100, DukeMTMC-ReID, CUHK03) and simulated monitoring scenarios.

Main Results:

• Achieved leading performance on public datasets with PSNR of 34.23 and SSIM of 0.78 on Urban100.
• Reached high average precision (mAP) of 91.4% on DukeMTMC-ReID and 82.7% on CUHK03.
• Demonstrated a 90.2% correct recognition rate in simulated monitoring, with false positive/negative rates below 7%, alongside reduced computational costs.

Conclusions:

• The BSRGAN ReiD method offers an efficient and robust solution for low-resolution pedestrian re-identification.
• The bidirectional adversarial network architecture significantly improves image reconstruction quality and recognition accuracy.
• This research holds strong theoretical value and practical application potential for intelligent security systems.