In response to the requirements for small size, low-power, and high-precision target recognition and ranging systems for platforms such as unmanned vehicles and drones, research has been conducted on a multi-camera array bionic compound eye system and its accompanying target recognition ranging algorithm. In this paper, we have designed a multi-camera array compound eye by combining multiple focal length lenses with a CMOS array. This compound eye system boasts a compact size and low power consumption. Integrated with an RV1126 core processing board for image processing, it is capable of independent target recognition and ranging within the system without the need for an external computer. Additionally, it provides real-time streaming of video and target distance information. Furthermore, this paper introduces a multi-focal length ranging algorithm based on the improved YOLOv5. By enhancing the small object recognition capability of the YOLOv5 and effectively utilizing the multi-scale target information obtained from multiple focal length lenses, the system achieves improved ranging accuracy, especially for distant small targets. The proposed multi-camera array compound eye system has dimensions of 80mm×80mm×80mm, weight of 460g, power consumption of 6.5W, and a single-frame image ranging speed of 60ms. Experimental results demonstrate that it successfully measures the distance to human targets within the range of 30m-120m, with an average ranging accuracy error of less than 1%. This bionic compound eye system holds significant potential for a wide range of applications, including emergency obstacle avoidance for drones, munition-borne reconnaissance, and navigation for underwater unmanned submersibles.
|