In this paper, we propose a novel technique for generating new objects in a video with a given velocity. The technique is called Abnormal Object Generator (AOG). AOG is used as a data generator to create pseudo-anomalous samples that are then fed into the memory-augmented autoencoder (MNAD) model for video anomaly detection. Specifically, three methods for generating pseudo-anomalous samples are proposed. The first method aims to create a single anomalous object in the original video. The second method generates multiple abnormal objects. The third method attempts to simulate the motion progression of anomalous objects, including uniform acceleration motion, uniform motion, and uniform deceleration motion. We evaluated the proposed methods on two benchmark datasets, including Ped2 and Avenue. Experimental results show that the proposed methods improve the performance of MNAD in video anomaly detection compared to the baseline models. Furthermore, the performance of the proposed approaches is competitive with the best model among the tested models using pseudo-anomalous samples.