Performance and Analysis of Optical Flow Techniques for Moving Object Segmentation & Detection in Infrared Tracking Method

For the detection of small targets using video infrared (IR), Between video sequences, the spatio-temporal information is crucial. In order to estimate motion and identify tiny targets in IR, optical flow is frequently used correction in order to efficiently utilise the additional temporal information. The standard optical Nevertheless, the broad viewing distance and poor IR imaging speed cause the target's spatial location to vary often between two frames, which reduces the effectiveness of optical flow-based detection techniques. Our solution, an end-to-end video infrared microscopic target identification system is one that is more resistant to significant motion and capable of more precise motion compensation. Flow-based detection techniques are only able to detect very slight motion in video sequences. Nevertheless, the broad viewing distance and poor IR imaging speed cause the target's spatial location to vary often between two frames, which reduces the effectiveness of optical flow-based detection techniques. The end-to-end video infrared microscopic target identification system we propose is more robust to large motion and capable of more precise motion compensation. To do motion estimates in a more accurate manner, we first advise employing an optical flow reconstruction network with many scales. Finally, the neighbouring frames are synchronised to the reference frame using the generated optical flows. The detection network is then given the concatenated neighbourhood frames to obtain the detection results.