Impact of Adaboost and Support Vector Machine Classifier in Automotive Sector

Abstract

The automobile industry is moving towards complete automation of vehicles. While the automation of vehicles can reduce the workload to human beings, we also need to make sure that safety is not compromised in the process. One of the integral components of the automation of vehicles is the detection of several obstacles. Obstacles may include lamp post, pedestrians, animals, vehicles, etc. This project involves detection of commercial vehicles that are present in front of the host vehicle and is limited only to far and middle region. The classification is done in two steps: Adaboost Training and Support Vector Machine Training. Adaboost being a weak classifier is used to reduce the reaction time of the entire system. Even though the classifier does not produce more accurate results it will reduce the time taken for classification considerably. Support Vector Machine algorithm will produce better classification result as it makes use of a hyperplane to separate between the classes. Support Vector Machine being a strong classifier gives better result but increases the reaction time taken for classification. In order to counter this problem Adaboost Training is done. Adaboost is a set of weak classifiers which will classify at a better speed but the results may not be accurate. Therefore, this Adaboost classification is done at first and only the positive classification from the Adaboost classifier are sent for Support Vector Machine classification. This will make sure that more time will not be used for the Support Vector Machine classification. Based on the classification if the result specifies that there is an object, the distance will be computed. Based on the distance and the speed of the host vehicle, the time to avoid collision can be calculated. The needs to be a lookup table to match different time frames with reduction of gears and the speed respectively. The ECU will manage the speed variation and the gear shift required through the control system that will give the time to avoid collision as the output. Therefore, the collision can be avoided by obstacle detection and computation of distance of the object.