3D Object Detection in Autonomous Driving

3D object detection is an important perception module in autonomous driving systems. It recognizes sensor observations and predicts locations, sizes and orientations of key objects, which provides both semantic and spatial information for high-level decision making. In this chapter, we first introduce and analyze the properties of commonly used perceptual sensors in autonomous vehicles: cameras, LiDARs and RADARs. Then we define the research problem, detail the assumptions and introduce evaluation metrics of 3D object detection in the context of autonomous driving. The main body reviews the state-of-the-art techniques and categorize them into camera-based, LiDAR-based, RADAR-based and multi-sensor fusion methods. For each method, we point out the main problems and their existing solutions. By analyzing the limitations of existing methods, we propose promising directions and open problems for future research.