An efficient animal detection system for smart cars using cascaded classifiers

Animal-Vehicle Collisions (AVCs) have been a challenging problem since the creation of cars. Consequently, such collisions cause hundreds of human and animal deaths, thousands of injuries, and billions of dollars in property damage every year. To cope with this challenge, vehicles have to be equipped with smart systems able to detect animals (e.g., moose), which cross roadways, and warn drivers about the imminent danger. In this paper, we develop a new animal detection system following two criteria: detection accuracy and detection speed. To achieve these requirements, a two-stage strategy system is investigated. In the first stage, we use the LBP-Adaboost algorithm which supplies the second stage by a set of ROIs containing moose and other similar-objects. Whereas the second stage is based on an adapted version of HOG-SVM classifier. In this stage, the non-moose ROIs are rejected. To train and test our system, we create our own dataset, which is frequently updated by adding new images. Through an extensive set of simulations, we show that our system is able to detect more than 83% of moose.