Comparison of feature extraction techniques to recognize traffic rule violations using low processing embedded system

Abstract

In India, it is observed that the number of people losing their lives in road accidents especially on highways is more than the death resulting due to naxalite, terrorism activity or epidemic. Government is investing plenty of money to educate people regarding road safety and curb death due to accidents, but people used to avoid it and entering themselves into danger zone. Several lives could be saved if the person(s) make use of helmet and wear seat belts while driving vehicles. Further, it is next to impossible for traffic police to catch each rider violating traffic rules, thus there is a need of the system to identify people disobeying road safety guideline which involves use of helmet and seat belt. The idea is to impose appropriate fine on such people to force them follow the road safety guidelines. Bike-riders without helmet and driving four wheeler without wearing seatbelt should be caught. Authors have performed four feature extraction techniques namely Scale-Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), Template Matching and Oriented FAST and Rotated BRIEF(ORB) to detect objects like vehicles, helmets, number plates, seatbelts for traffic data sets on Raspberry Pi 2 (B) using OpenCV3.0 and Python 3.4.2. These feature extraction techniques have been evaluated on collected dataset and simulation results performed on raspberry pi on valid dataset. The observation suggests that SIFT algorithm can be used to get higher accuracy compared to SURF and ORB for rule violators at toll system on highways or traffic cross road in city.