遮挡和光照变化下的单鱼目标跟踪

Single Fish Tracking under Occlusion and Light Change

鱼类跟踪是对鱼类行为分析的基础,在对水中的鱼类跟踪时,目标鱼会由于姿势的变化、受到周围鱼或者物体的遮挡或者光照的影响变得难以追踪。针对这种情况,提出一种利用MobileNet-SSD(SSD,single shot multibox detector)与Dlib关联跟踪器相结合对鱼体进行跟踪的方法,通过SSD算法精准地检测到视频中的鱼体,再将信息输入到Dlib关联跟踪器中,使跟踪对象定位更加准确,提高了鱼体在水中运动发生遮挡和光照变化时跟踪的鲁棒性。实验结果表明,该方法在鱼体跟踪视频中的表现明显优于其它算法,在不同环境下的跟踪成功率达到90%以上。

Fish tracking is the basis of fish behavior analysis,when tracking fish in water,the target fish will be difficult to track because of the change of posture and the influence of the occlusion or light of the surrounding fish or object.In order to solve the above problem,a method of tracking fish body by combining Mobilenet-SSD(SSD,single shot multibox detector)with Dlib association tracker was proposed.The fish body in video is accurately detected by SSD algorithm,and then the information is input into Dlib association tracker,which makes the tracking object location more accurate and improves the robustness of fish body tracking when the movement of fish in water occured and the light changes.The experimental results show that the performance of the mentioned method in fish tracking video is obviously better than that of other algorithms,the tracking success rate in different environments is more than 90%.