<p>Obstacle detection and real time planning of collision free pathways is a key design requirement for an autonomous ground vehicle. In this paper we propose a complete system for obstacle detection and avoidance using computer vision in combination with machine learning methods. Using the vision system, the autonomous vehicle is able to detect and classify objects in its environment with high accuracy and determine its navigation. Furthermore, the necessary image files are transported into the robot operating system (ROS) to test in the simulation. Google’s TensorFlow software library and ImageAI computer vision Python library, are two readily available open-source softwares used to create a custom neural network models to detect and classify specific objects, i.e. lane lines to drive within or caution cones to avoid hazards. Initial development focused on processing and annotating two hundred images of a random object in various angles, distances, and light conditions recorded from the cameras used on the autonomous vehicle. These images are processed through Google’s CoLaboratory, a cloud-based development server providing free GPU hardware, for faster processing. The goal is to take the customized detection model and provide a desired output from the camera’s images. The test procedure included coding annotated pictures of different objects and creating datasets. The experiment followed individually collecting and annotating two hundred photos for each of the following objects: orange cones, flags, and ramps. The procedure for training the datasets were developed and as a result, several models for each of the objects are obtained. The object detection model with the best capability to detect and distinct the designated object was evaluated. The intention of generating different models is to develop a database to reinforce recognition of the said obstacle, and with the obstacle detection implemented in the autonomous vehicle software architecture, the navigation was determined. The featured paper reflects how machine learning is utilized for the execution of obstacle detection and navigation, demonstrating where the execution grants the autonomous vehicle the ability to maneuver around designated impedances while complying with the user’s assigned constraints. It also discusses the troubleshooting that comes along with the generation of the detection models and what it takes to implement different models in ROS where consequently, the autonomous vehicle can then proceed with its objective of obstacle detection and navigation. The process was evaluated and after successful tests in the simulation, the proposed approach is effective for obstacle detection and avoidance in the outdoor environment.</p>

Vision Based Obstacle Detection and Navigation of an Autonomous Vehicle

Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2021/volume/85543

ASME 2021 International Mechanical Engineering Congress and Exposition (IMECE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.