Enhancing Ability of Autonomous Vehicles to Detect Objects/Obstacles in Diverse Environments and Weather Conditions

Abstract

The ability for autonomous vehicles to detect objects and obstacles in various weather conditions, such as rain and fog have become a challenge. The weather conditions make it even harder for the accurate and precise detection of objects. Convolutional neural networks, CNN models work as a multiple deep learning model utilizing data to increase the performance of the model. The technique for data augmentation is useful for limited training for the applications. The framework is then applied to the algorithm at two layers, single model and multi model. Machine learning models such as You Only Look Once, or YOLO contains 80 built-in object classes that is able to detect. Object detection algorithms provide accurate and precise results, allowing vehicles to drive without sensors, enable assistive devices to convey real-time scene information to human users. The results for the model accuracy are as follows for inclement weather conditions: 52.56% average precision in detecting cars and other objects.