Heterogeneous Sensor Fusion with GMPHD for Environmentally Adaptable Obstacle Detection in Mobility Systems

Obstacle detection is an essential process in consumer's autonomous mobility systems such as autonomous vehicles inside the dedicated lane to acquire the location of obstacles, and it has become a popular topic in this decade with the blooming of various object detection algorithms and the enhancement of sensor quality. To maintain high accuracy of obstacles' detection in mobility systems outdoor, a sensor fusion system is required to essentially support environmental influence such as lousy weather as well as high moving speeds and adaptably deal with clutter and miss detection based on the incoming measurements from heterogenous sensors with Camera, LiDAR and Radar. Since no current literature about Gaussian mixture probability hypothesis density (GMPHD) handles the above low accuracy fusion problem due to environmental influence for heterogeneous sensors, we propose the concept of integrating GMPHD to heterogeneous sensor fusion with three architectures, Track-to-Track-Fusion (T2TF), Measurement-to-Track-Fusion (M2TF) and Track-to-Association-Fusion (T2AF) and further evaluate their performances respectively in terms of their fusion improvement abilities to determine their practicalities for mobility systems by using the simulation datasets which reproduce ordinary and poorer conditions with the degradation of sensors' performance in the assumption of environmental influences.