3D Models Recognition Using Overlap Histograms and Machine Learning

Abstract

We propose three different methods to conduct 3D model recognition through Euclidean distances between pairs of randomly selected vertices within the object. The first method involves comparison through a bin array and is used as a baseline solution. Data from the input object must be within 40% of the data it is being compared to. At least 80% of the compared data must be matched in order to be classified as a possible object. The second method uses a direct comparison of distance vs. probability histograms. By using histograms, we can contrast the input to the comparative data visually. Lastly, we use Support Vector Machine to classify the object’s class by computing a hyperplane based on input data. Our results show that by using our histogram overlap comparison, we are able to classify 85% of tested objects correctly.