FFAVOD: Feature fusion architecture for video object detection

• We designed a novel architecture for video object detection that capitalizes on temporal information. fusion module to merge feature maps coming from several temporally close frames. proposed an improvement the SpotNet attention module. trained and evaluated our with three different base detectors two traffic surveillance datasets. demonstrated consistent significant of model over baselines. A amount redundancy exists between consecutive frames video. Object typically produce detections one image at time, without any capabilities taking advantage this redundancy. Meanwhile, many applications work videos, including intelligent transportation systems, advanced driver assistance systems surveillance. Our aims similarity better detections. propose FFAVOD, standing detection. first introduce allows network share nearby Second, we learns enhance them. show using can improve performance benchmarks containing sequences moving road users. Additionally, further increase performance, Using improved detector, obtain state-of-the-art UA-DETRAC public benchmark as well UAVDT dataset. Code is available https://github.com/hu64/FFAVOD .