Pedestrian Classification Using K-means and Random Decision Forests

Abstract

In field of autonomous and intelligent vehicles, the goal of pedestrian classification is to reduce amount of accidents. The object classification accuracy depends on the type of classifier and the extracted object features used for classification. Support Vector Machines (SVM), is considered the most effective classifier for this task. However, it depends on a number of factors that require researchers to perform several modifications to obtain a good result with adequate performance. This study presents a promising alternative with fewer parameters, which works on large datasets, and reduced runtime. It also has the advantage of allowing the data analysis between every step of the algorithm. Differently from SVM, which can be considered as a black box approach, our method uses a k-means cluster technique combined with a radial basis function to transform data into a smaller and more relevant set, where the classification is performed using random decision forest. Experimental results show very satisfactory classification, efficient computational performance, simplicity of use, and reduced setup.