Comparison of Regression Algorithm for Vanishing Point Estimation based on Car Distribution

In order to improve an application from the viewpoint of helping blind people waiting for the bus, classification of the vanishing view is required. The vanishing point from a perspective view is a feature that may improve the performance of viewpoint classification, especially in the case of congested traffic. The idea proposed by this study bases on the vanishing point estimation following car distribution on the road. In essence, the cars are detected using a You Only Look Once (YOLO) technique which extracts all the to the center point of a squared boundary. Normalization of the data points are then computed in percentages. Next is the calculation of the nineteen features of the data points of the car distribution. Finally, the vanishing point is estimated by supervised regression. This experiment compares the estimated results among five different regressions consisting of Linear, k-Nearest Neighbor, Decision Tree, Support Vector, and Multi-layer Per-ceptron regression. The obtain results show that the k-Nearest Neighbor regression computed the lowest distance error of 10.53, compared to the other regression algorithms.