Object Distance Estimation with Machine Learning Algorithms for Stereo Vision

This paper presents a novel distance estimation to calculate distances from the stereo camera to the object accurately. This study collected stereo camera images as a dataset, each object determined at two different lighting environments and five different distances between the stereo camera and the object. To estimate the distance, researchers applied supervised learning methods to approach this task. There were performed with two machine learning algorithms: Linear Regression, and Artificial Neuron Network Regression. In the experimental results, the efficiency of the proposed method was examined by using the evaluation metrics to calculate the distance estimation errors. The results showed the model of convolutional neuron networks operated with densely connected neuron networks has the lowest errors rate in comparison with other models. The model eliminates the error rate of distance estimation at 0.000531, 0.014490, and 0.000048 meters, measured by mean square error, mean absolute error and mean logarithmic error respectively.