Overlapping Shoeprint Detection by Edge Detection and Deep Learning.

Area of Science:

• Computer Vision
• 2-D Image Processing
• Forensic Science

Background:

• Accurate object detection and segmentation in 2-D imaging are challenging, especially with overlapping or obscured objects.
• Forensic shoeprint analysis is particularly difficult due to noisy backgrounds and indistinct prints.
• Traditional Convolutional Neural Networks (CNNs) struggle with delineating overlapping objects in complex, noisy environments.

Purpose of the Study:

• To improve the detection and segmentation of overlapping shoeprints in forensic investigations.
• To address the limitations of traditional CNNs in handling noisy and complex image data.
• To develop a robust methodology for identifying multiple overlapping objects in challenging visual conditions.

Main Methods:

• Employed the YOLO (You Only Look Once) object detection model.
• Integrated edge detection and image segmentation techniques to enhance YOLO's performance.
• Utilized convolution layer heatmaps to visualize network convergence and detection processes.

Main Results:

• Achieved high confidence levels (above 85%) for minimally overlapped shoeprints.
• Maintained significant detection accuracy (above 70%) for extensively overlapped shoeprints.
• Demonstrated improved sensitivity and precision in detecting shoeprints against noisy backgrounds.

Conclusions:

• The enhanced YOLO model with edge detection and segmentation effectively improves overlapping shoeprint detection.
• This approach offers a promising methodology for object detection in noisy, complex environments.
• The findings have implications for advancing forensic image analysis and computer vision techniques.