Distance Estimation using Tensorflow Object Detection

• Published in: 2019 International Conference on Electrical, Electronics and Computer Engineering (UPCON) pp. 1 - 6
• Main Authors: Amin, Faizan, Mehdi, Syed Ali Mohd Murtaza, Khan, Ahmad Danish
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2019
• Subjects: bounding box; depth; disparity; stereo images; Tensorflow object detector
• ISSN: 2687-7767
• DOI: 10.1109/UPCON47278.2019.8980216

Today, as a civilization we produce an unprecedented amount of data, in the form of audio, images, video and so forth. One application of this data-driven approach is autonomous vehicles. The current technologies have made it possible to happen. The top car companies like Hyundai, Kia, Ford motors, Tesla motors have been working on the self-driving cars projects and they have achieved it to some extent. But self-autonomous vehicles are not only limited to the self-driving cars, but the UAVs (Unmanned Aerial Vehicles) are also part of it. While the applications of the self-driving cars are somewhat limited to the usage of normal public, the UAVs have applications that vary from surveillance to patrol to enemy reconnaissance, in short, the UAVs have more applications for the military than the normal public. These autonomous vehicles require an understanding of the environment they operate in. As these vehicles are used to travel in cities (in case of self-driving cars) and also might be used in forests or mountains (in case of UAV use by for reconnaissance), they require to detect obstacles in order to avoid them. This is often achieved through scene depth estimation, by various means. We propose an approach which not only requires a minimum amount of space but also consumes far less power. Our approach is based on Obstacle Detection and calculating distance using the disparity estimated. These represent highly desirable features, especially for micro aerial vehicles.