The technology described herein includes a system and/or a method for global hypothesis tracking. In some examples, a method generates one or more paired segments based on track data representing kinematic data of target objects. Each paired segment includes a list of tracks incompatible with the paired segment, which are tracks sharing common track data, and a likelihood score. The method generates a transition probability between each pair of the paired segments based on the list of tracks incompatible with the paired segment and the likelihood score associated with each paired segment. The method further generates one or more multi-segment tracks based on the one or more paired segments also based on the transition probability between each pair of the paired segments.