Adaptive Correction of Landmark for Visual Homing in Mobile Vehicles

This paper presents an optimized visual homing method based on Average Landmark Vector (ALV), termed Adaptive Correction of Landmark (ACoL), which exhibits good homing performance and robustness in both static and dynamic indoor environments. The salient features of this paper are: (1) Inherent constraints of visual homing are analyzed in detail to reveal the limitations of existing ALV-based methods. (2) Instead of removing mismatched landmarks or artificially assigning fixed weights to landmarks, a novel solution is proposed by adaptively evaluating a reasonable weight for each landmark to account for contribution to the output. (3) Computation time of the proposed method is particularly concerned to enable the vehicle to perform real-time homing tasks in actual scenarios. (4) Simulations and experiments demonstrate the superiority of the proposed method under both static and dynamic conditions. Our method focuses on improving a vehicle's homing ability in terms of landmark matching, landmark distribution and homing direction. Comparative studies with other ALV-based methods demonstrate that our method generalizes well to various complex and challenging indoor scenarios.