Locust is the main pest endangering Chinese agriculture and even global agriculture. Therefore, the development of an efficient and real-time locust monitoring technology is of great significance to the prevention and management of locust disasters. Video monitoring technology has been widely used in locust monitoring, but the different postures of locusts in the field and the problem of plant shelter pose great challenges to accurate locust monitoring. In this paper, taking a 5-year-old migratory locust as the research object, video target detection based on the MOG2-YOLOv4 network is proposed to address the problem of video frame recognition failure caused by different postures and partial occlusion during locust movement. First, the frame compensation algorithm is used to identify locusts with different postures; then the background separation algorithm (MOG2) is used to extract the spatio-temporal features of the next and previous frames of the video, and the combination of the MOG2 algorithm and YOLOv4 algorithm to extract the static features of locusts in the video can improve the detection accuracy. Experimental results show that the average detection accuracy of the MOG2-YOLOv4 model is 82.33% (8.17% higher than that of YOLOv4), and the F-mean is 89.37% (4.76% higher). Therefore, this method has a good effect on the identification of East Asian migratory locusts, particularly under occlusion and motion blur conditions; it achieves accurate monitoring of the occurrence of locust plagues, which is of great significance to the prevention and control of locust plagues.