Detection of Road Objects Based on Camera Sensors for Autonomous Driving in Various Traffic Situations

Successfully detecting road objects from camera sensors in various traffic environments would significantly facilitate the development of driving safety applications for autonomous driving. However, object occlusion, bad weather, and small objects in images would restrict the performance of object detection approaches based on camera sensors. To resolve these issues, a lightweight neural network modified convolutional block attention mechanism (MCBAM)-CenterNet was proposed. In this approach, convolutional block attention module (CBAM) was used to find the focus in an image to enhance the detection capability. The lite-hourglass module was then adopted specifically for the detection of cars and pedestrians. A large-scale realistic road object detection dataset BDD100K that includes various traffic images from camera sensors was adopted in this article to evaluate the effectiveness of our proposed MCBAM-CenterNet. The evaluation results show that MCBAM-CenterNet can effectively detect road objects under different scenarios.