An Feature Fusion Object Detector for Autonomous Driving in Mining Area

An accurate and reliable object detection is an important building block for the most autonomous driving system. Specially, in some scenarios like autonomous driving in the mining area, the detection of small and hard samples are a great challenges for perception system. In this paper, we propose a multi-scales fusion and attention-based model to improve the performance of the object detection for different scales and camouflaged obstacles in the mining area, like trucks, rocks, person and so on. Based on our proposed model, we have developed a new object detector, called FANet, which achieves much better efficiency than prior arts by using some optimization strategies. When we test the proposed model at our datasets, it can achieve 77.76 AP50 in 19 ms on a Titan X, compared to 68.16 AP50 in 22 ms by YOLOv3. The work has successfully apply in the autonomous driving system in mining area.