The Design And Implementation Of Footprint Recognition System Based On Computer Vision

In the process of public security criminal investigation and solving the case,information collection personnel will collect information such as video surveillance,fingerprints,and footprints available on the spot.The footprint information is usually carried out by experienced police officers on the spot to measure the size and depth of the footprint,and compare it with the footprint of a normal person to roughly judge the approximate gender and height of the person who measured the footprint.Because this traditional method is limited by the scarcity of experienced police officers,the collection method is clumsy,and the manual recognition error is large,it cannot be widely promoted.A footprint recognition system that can automatically recognize the gender and height of the person who owns the footprint is needed.First of all,through the research on footprint recognition,the size and clarity of the footprint in the image are the most important factors affecting the recognition accuracy.According to this feature,the system is divided into two tasks: image correction and footprint recognition.Then the image correction and restoration technology is researched,and the application and advantages and disadvantages of morphological technology and artificial intelligence technology in image correction and restoration are judged.Finally,affine transformation and perspective transformation algorithms in morphology are selected as the application technology of image correction function.Finally,through the analysis of machine learning and neural network classification and regression algorithms,this system chooses the Res Net50 network as the basic model,and combines the identification requirements of the system to adjust the output parameters of the Res Net50 network to support the gender and height of the person whose output footprint belongs to information.Through the Py Torch framework combined with the CUDA calculation library,combined with 17,000 image training sets with real gender and height information,the Res Net50 network model is trained to enable the model to support accurate identification of the gender and height of the person who belongs to the footprint.The system is finally verified on a test set of 2500 test pictures,in which the image correction function realizes that the size error of the printed footprint on A4 paper does not exceed 1 mm.The gender classification accuracy rate of the footprint recognition function is 85%,the average error of the regression model does not exceed 3 cm,and the number of test sets with an error within 3 cm accounted for more than 80%.In terms of processing speed,it realizes the correction and recognition process of processing an image in 1second.