Complementary and situation sensitive object detection, review performance, and performance dependencies of common approaches

Object detection can be performed on different modalities like camera or lidar systems. Image-based approaches are highly sensitive to variations of illumination. On the other hand, lidar-based approaches fail on detecting small objects or objects at higher distances. The research in this field leads to a variety of different approaches with different advantages and drawbacks. Fusion-based approaches aim to utilize the advantages of the available modalities and specialized approaches. Usually, the performance of object detection approaches is measured over a set of situations. The performance measures are typically provided as average values over a test dataset. Local variations are not considered for comparison and not utilized. For a better understanding, this contribution analyzes performance variations based on different situation parameters to understand the complementary potential and improve the application of redundant and situated object detection systems.