Smart car with road tracking and obstacle avoidance based on Resnet18-CBAM

Abstract

While some existing researches in automatic driving demonstrate the ability to perform road tracking and obstacle avoidance tasks, they are not satisfactory in anti-noise ability. It can be attributed to various factors, including latency issues with development boards and sensors and limitations of the chosen model. To accomplish the tasks of road tracking and obstacle avoidance concurrently and improve the model's anti-jamming capability, we propose the use of Resnet18-CBAM in smart cars. More importantly, in order to optimize Resnet18CBAM performance, we filter the hyperparameters and select the group with the highest performance, which is Mish/SmoothL1/Adam. The experimental results demonstrate that our method extracts more features from target objects than existing methods and significantly improves anti-noise performance when performing road tracking and obstacle avoidance tasks. The smart car scored 98% in the training environment and 72% in the environment with lighting noise, significantly higher than the 32% achieved by the existing method.