Automatic obstacle avoidance model vehicles based on multi-sensor and deep learning

Abstract

Nowadays, the use of autonomous driving has grown steadily as technology has progressed. This thesis investigates the potential for enhanced autonomous driving accuracy by combining sensors and AI algorithms. All algorithms and other tests for the thesis were performed on a model vehicle. And I mainly verify lane recognition and obstacle detection. The obstacle and lane are recognized by the camera after deep learning, then the car will changing lane to obstacle avoidance. To enhance the efficacy of obstacle avoidance of model vehicles, the advantages and drawbacks of various image recognition and obstacle avoidance algorithms are also discussed in this paper. At the same time, the combination of the use of LIDAR can better fill the shortage of image recognition. Ultimately we propose that the data from the sensors can be processed with different connectivity layers, and that hybrid deep learning can be used to improve the accuracy of autonomous vehicle driving for obstacle avoidance.