Object Recognition and Tracking System Using Forward-Looking Sonar for Long-Distance to Near-Field Detection

This paper presents a target recognition and tracking system based on forward-looking sonar for long-distance to near-field detection. The system is designed for autonomous underwater vehicles (AUVs) to detect and continuously track objects of interest over extended ranges, supporting both obstacle avoidance and target following missions. A multibeam imaging sonar from BlueView is used to acquire real-time image streams at maximum detection ranges. A lightweight deep neural network is employed to identify targets within the sonar field of view. An enhanced filtering-based tracking algorithm is developed to maintain target lock. For tracking missions, a distance-tiered strategy based on relative range is adopted to manage target engagement across different distances. For obstacle avoidance, an improved artificial potential field method is implemented to ensure mission reliability. Lake trials involving coordinated AUV formations were conducted to validate the effectiveness of the proposed system.