3-Survivor: A Rough Terrain Negotiable Search and Surveillance Mobile Robot with Real-Time Object Detection

摘要

This paper presents the design and integration of 3-Survivor: a rough terrain negotiable search and surveillance robot. In 3-Survivor, a modified double-tracked chained wheel with a caterpillar mechanism incorporates the body design. A passive adjustment is established in the body balance that enables the front and rear body to operate in excellent synchronization. The robot is remotely operated using the web portal, and the entire operation is telecast with a Raspberry Pi Sony IMX477R camera. An object detection module (ODM) is incorporated with live streaming to integrate the surveillance system. A learning-based EfficientDet-D7 network is optimized for precise and versatile target recognition and interaction. The D7 includes a computationally efficient bi-directional feature pyramid network (BiFPN) as the backbone network that allows multi-scale feature fusion. A custom dataset containing 5000 images of indoor-outdoor objects is developed to train and validate the performance of the SAR tasks. A very impressive 56.7 mAP is acquired from this proposed D7 model. The feasibility and efficacy of the proposed system are tested and validated by indoor and outdoor SAR trial operations.