Research On The Toe-off Feature Of Planar Shoeprint Based On Machine Learning

In Chinese forensic science,a planar footprint can provide police office lots of information.The toe-off feature is an important feature of planar shoeprint,which can indicate the gait pattern of the walkers.However,due to the vague description of footprint characteristics,complex causes and unclear mechanism,the toe-off features of planar shoeprints are still analyzed artificially by criminal investigators,which is inefficient and subjective.In order to suit the demand of the criminal investigation and improve the efficiency,and inherit the development of the traditional footprint examination,a novel algorithm for the automatic detection of the toe-off feature is developed in this research.About 320 volunteers were recruited to collect more than 6,000 planar footprint images.We took photo of planar shoeprints by the way of criminal scene photography,and performed some pre-processing steps on these pictures.We defined the crescent feature in the toe-off feature of planar footprint as a positive sample,and defined no such feature as a negative sample.The first step is to detect if there is any "crescent" trace.A self-built 6-layer small convolutional neural network was used to detect it,and the test accuracy rate could reach about 97.0%;The VGG-16 classical convolutional neural network was used to detect it,and the test accuracy rate was about 84.8%;The Haar-like features was extracted and classified with Adaboost,and the test accuracy was about 84.3%.It is available that the 6-layer small convolutional neural network worked best in this step.The second step is to judge the direction of this trace based on the relative positional relationship between the crescent trace and the footprint,and use it to judge the direction of the footsteps.The GIST features was extracted and classified with the SVM classifier,and the test accuracy rates of the left and right feet were about 85.4% and 81.1%.The HOG features was extracted and classified with the SVM classifier,and the accuracy rates of the left and right feet were about 80.7% and 76.3%.The Haar-like features was extracted and classified with Adaboost,and the test accuracy rates of the left and right feet were 72.7% and 80.4%.It could be obtained that extracting the GIST features and being classified with SVM is the best method.Finally,the relationship between the direction of the Toe-off feature of planar shoeprint and the posture of the people was explored.The camera behind the volunteer recorded the whole process of walking.The image of the moment they raised their feet was intercepted and then we measured the angle between their calf and their foot.It is found that the angle of volunteers who is judged to be a measured gait with the toes pointing outwards in the second step was smaller than others’.Through the research,the possibility of automatically extracting and classifying the impression pattern of planar footprints was proven.And this provided a new horizon of identification recognition with impression pattern in planar footprints.