In the deployments of cyber-physical systems, specifically predictive maintenance and Internet of Things applications, a staggering amount of data can be harvested, transmitted, and recorded. Although the collection of large data sets can be used for many solutions, its utilization is made difficult by the increased overhead on the transmission and limited processing capabilities of the underlying physical system. For such highly correlated and extensive data, this situation is usually described as data-rich, information-poor. We propose for the first time a novel one-stage method, called Information-Abstraction-Net (IA-Net), for the detection of abnormal events in multi-object anomaly detection scenarios by utilizing highly abstracted sensory information instead of the entire sampled data set to elevate the transmission and analysis needs of the system. We find that the computation complexity of IA-Net is reduced by half compared to competing solutions and the detection accuracy is increased by about 5-47%, as well.