Comparative analysis of features extraction methods for detection traffic rule violation on Raspberry Pi hardware

Abstract

In India, People losing their lives  in road accidents due to traffic rule violations, compared to naxal violence or natural disasters. People can follow traffic rule but by disobeying the rule they put their life in danger. Government organizing  lots of awareness program to  educate people to follow the traffic rules and save lives. Recent days traffic rule violation detection is by manual monitoring, time consuming, hectic and change of corruption . Sensor based approach has been used to track these violations, but at the cost of infrastructure requirement. Here, authors have presented how feature extraction technique can play role in smart traffic rule violation detection on roads and highways. Machine vision based approach to identify the traffic rule(s) violators on highways at toll tax unit by extracting significant features/descriptors of the images and use of classification/matching algorithms using low cost Raspberry Pi hardware. Author has performed several rule violation detection using SIFT ,SURF, ROOTSIFT ,Grab cut, HAAR cascade and Hough transform on test image dataset to identify traffic rule violations.  Brute-Force and FLANN-Index matcher to detect and identify traffic rule violator without wearing helmets and seatbelt and haar cascade for mobile usage detection in test image data set using low processing capacity hardware. Amongst the feature extraction techniques, RootSIFT algorithm with FLANN-Index matcher gives computation time of 0.50s and having 91.5% accuracy. Grab cut for Lane detection and haar cascade for mobile usage detection has produced superior results on Raspberry pi 2(B).

Keywords: Features,  Lane, helmet, SIFT,SURF.