Accurate positioning is a fundamental prerequisite for intelligent connected vehicles (ICVs). Based on global navigation satellite systems, absolute positioning of ICVs can be augmented by cooperative positioning (CP) which fused the state-related information shared in vehicular networks. Common CP relies on communication signals or special equipment to measure the distance between vehicles. This kind of ranging is suffering from multipath and non-line of sight and hinders the improvement of CP. Using vehicle-to-target relative vectors (V2T-RVs) based on on-board sensors, which is immune to multipath and non-line-of-sight, a distributed fusion framework named multisource-multitarget cooperative positioning is proposed in this paper. Without knowing which target the V2T-RVs are originated from, the positioning problem is converted into a multi-target tracking problem by converting the V2T-RVs into global coordinate. Then, a classic ellipse gate (EG) algorithm is used to pair the ICVs and the converted measurements. Finally, the sequential Kalman filter (KF) is used to complete the state estimation under multiple measurements and obtain the improved absolute position. The above EGKF method is verified in two scenarios generated by microscopic traffic simulator. Performance results show that the EGKF method within multisource-multitarget cooperative positioning can significantly improve the positioning accuracy.