Camera-Based Object Detection, Identification and Distance Estimation

A system combining deep learning and stereovision for detection, tagging and distance estimation of objects ahead, is presented in this study. A Convolutional Neural Network (CNN) is used to detect and identify objects in the field of vision of the stereo camera. Once the objects are detected and identified, their distance from the camera is estimated by first constructing 3D point cloud using triangulation method. The system is implemented on Nvidia® Jetson TX1 with a Zed® stereo camera, tested and found to be capable of a detection accuracy of around 84% with the average error in distance estimation less than 4.7% up to a distance of five meters.