Autonomous vehicles and advanced driver assistance systems are among the hot research topics in the automotive industry. Lidar, camera and automotive radar sensors are the backbones of the autonomous vehicle [1], [2]. Among these sensors, the ability of the radar to detect obstacles in poor visibility and weather conditions makes it most suitable to handle the challenges in autonomous driving.

For instance, consider a complex urban environment with several buildings and guardrails, the radar transmitted signal might take a different propagation path called direct path and multipath on the transmitting or receiving way [3]. The direct path is when the transmitted signal directly hits the target and reflects towards the radar. On the other hand, the multipath signals are due to scattering and reflection of the transmitted signal from other reflectors such as guardrails and buildings. Fig. 1 depict this effect based on a scenario, where an automotive vehicle is tracking another vehicle travelling ahead. In this instance, the guardrail of the road act as the reflection surface and the radar generates measurements from direct and multipath signals. As a result, multiple tracks could be generated for a target, which are known as ghost targets, if multipath measurements are treated as direct path measurements. Therefore, at times in the receiver, the multipath signals are distinguished and discarded as clutter. However, these multipath signals also consist of some valuable information about the target state. By utilizing these signals, the performance of the tracking could be improved  [3]-[5].

In several applications, the multipath signals have been utilized to improve the performance of interest. For example, a multi-frame assignment technique is proposed to incorporate the multipath information from distinct propagation modes to track multiple targets by an Over The Horizon Radar (OTHR) with known reflection points in [4]. This algorithm outperforms the performance RMSE of the conventional method.

In a practical situation, it is not possible to know about the reflection surface entirely. Hence, in our approach, we have considered a problem with an unknown reflection surface geometry. As a result of the relaxation about the assumption, the observability of the system can be affected. Therefore, it is vital to find the observability of the problem. Thus, in our approach the observability of the problem is analysed using the invertibility of the Fisher Information Matrix (FIM) [6]. In addition to the observability analysis, the enhancement in performance by incorporating the multipath measurement with direct path measurement for target tracking is analyzed via simulation. Therefore, the contributions of this paper are summarized as follows:

1. The observability of the multipath assisted target tracking is analyzed with the unknown reflection surface. The FIM is derived, the observability is determined by checking the invertibility of the matrix.
2. The Extended Kalman Filter algorithm is derived based on the observability analysis. In addition, the simulation results on three different cases show the effectiveness of fusing the multipath with a direct path.