A method for detecting key points of shoe samples and footprints based on deep learning

Abstract

The invention discloses a method for detecting key points of shoe samples and footprints based on deep learning, including: S1. Obtaining a shoe sample / footprint database: using crawler technology to crawl shoe sample data pictures, and after obtaining the shoe sample pictures, use manual calibration Mark the key points in a way to generate a shoe sample data set; the footprint data set includes on-site footprint data and suspect footprint sample data; S2. Set the network model; S3. Calculate the loss function and propose a loss function based on the sole / footprint outline; S4. Train the network model, and use the transfer learning mode of partial network structure adjustment for training; S5. Normalize the size of the image and input it into the trained network model, and the coordinates of the output result are marked on the original image. Using the deep learning network to extract key point information makes it possible for the computer to mark footprints or shoe-like images, which greatly reduces labor costs.

Description

technical field [0001] The invention relates to a method for detecting key points based on deep learning, in particular to a method for detecting key points of shoe samples and footprints based on deep learning. Background technique [0002] Footprint information plays a pivotal role in the field of modern criminal investigation and investigation, and is one of the important physical evidence of on-site inspection. Through the trace characteristics reflected by the sole on the bearing body, not only the natural information such as the approximate height, weight, and age of the person can be initially analyzed and described, but also the walking posture, center of gravity and other characteristics of the person can be reflected through the wear information of the sole. The sole pattern information of the shoe sample can complement the incomplete footprint information, and through comparison, the incomplete pattern and worn areas of the footprint can be restored, which has pla...

AI Technical Summary

Problems solved by technology

The outline of the detected target is better, especially in complex scenes with multiple targets, the semantic understanding is more prominent, but it also cannot solve the detailed position of the target

Images