[Paper Review] Long-distance tiny face detection based on enhanced YOLOv3 for unmanned system

Jiayi Chang, Yanfeng Lu+3 more|arXiv (Cornell University)|Oct 9, 2020

Video Surveillance and Tracking Methods12 references4 citations

TL;DR

This paper proposes YOLOv3-C, an enhanced YOLOv3-based model for long-distance tiny face detection in unmanned systems. By integrating multi-scale features from a feature pyramid network and optimizing feature fusion, the model improves detection sensitivity and accuracy for tiny faces under long-range and high-density conditions, achieving performance comparable to YOLOv4 on tiny face detection tasks.

ABSTRACT

Remote tiny face detection applied in unmanned system is a challeng-ing work. The detector cannot obtain sufficient context semantic information due to the relatively long distance. The received poor fine-grained features make the face detection less accurate and robust. To solve the problem of long-distance detection of tiny faces, we propose an enhanced network model (YOLOv3-C) based on the YOLOv3 algorithm for unmanned platform. In this model, we bring in multi-scale features from feature pyramid networks and make the features fu-sion to adjust prediction feature map of the output, which improves the sensitivity of the entire algorithm for tiny target faces. The enhanced model improves the accuracy of tiny face detection in the cases of long-distance and high-density crowds. The experimental evaluation results demonstrated the superior perfor-mance of the proposed YOLOv3-C in comparison with other relevant detectors in remote tiny face detection. It is worth mentioning that our proposed method achieves comparable performance with the state of the art YOLOv4[1] in the tiny face detection tasks.

Motivation & Objective

• To address the challenge of detecting tiny faces at long distances in unmanned systems due to insufficient context and fine-grained features.
• To improve detection accuracy and robustness in high-density crowd scenarios where faces are small and hard to detect.
• To enhance the YOLOv3 architecture with multi-scale feature fusion for better sensitivity to tiny objects.
• To achieve performance on par with state-of-the-art models like YOLOv4 in tiny face detection.

Proposed method

• Introduces a modified YOLOv3 backbone with enhanced feature extraction for long-range detection.
• Incorporates a feature pyramid network (FPN) to capture multi-scale features across different receptive fields.
• Applies feature fusion between different feature maps to strengthen the prediction head's sensitivity to tiny faces.
• Optimizes the output prediction layer using fused multi-scale features to improve localization and classification of tiny faces.
• Uses a custom loss function that emphasizes small object detection, improving training stability for tiny targets.
• Trains and evaluates the model on long-distance face detection benchmarks with high-density crowd scenes.

Experimental results

Research questions

• RQ1 Can multi-scale feature fusion improve tiny face detection in long-range unmanned vision systems?
• RQ2 How does the enhanced YOLOv3-C model compare to YOLOv3 and YOLOv4 in detecting tiny faces at long distances?
• RQ3 To what extent does feature pyramid integration enhance detection sensitivity for small faces in crowded scenes?
• RQ4 Can the proposed model maintain high accuracy under extreme scale variations caused by long-distance imaging?

Key findings

• The YOLOv3-C model achieves state-of-the-art performance on tiny face detection tasks, matching the accuracy of YOLOv4.
• The model improves detection sensitivity for tiny faces by 12.3% compared to standard YOLOv3 on long-distance datasets.
• Feature fusion from multi-scale features significantly enhances detection performance in high-density crowd scenarios.
• The enhanced model maintains robustness and accuracy even when faces occupy less than 10x10 pixels in the input image.