Widespread adoption of the Internet of Things (IoT) in many sectors, including homes, industries, healthcare, and finance, attracts cyber attackers to conduct malicious activity throughout the network. The IoT devices are resource-constrained in nature, and is not possible to install security mechanisms like in the conventional systems. Additionally, the advent of sophisticated attacking tools and the resource limitations of the IoT device allow attackers to perform attacks more easily. Hence, there is a crucial need to bar IoT devices from any form of attack. To address this, we present an intrusion detection system, which is a hybrid method known as CV-AE, incorporating two Autoencoders, namely Convolutional Autoencoder (CAE) and Variational Autoencoder (VAE), and a classifier for extracting a compressed spatial feature to identify the IoT network attacks. The beauty of the CV-AE hybrid method is the efficient way of extracting network traffic features, which are represented in the latent space of the CAE and VAE architectures. The effectiveness of the proposed hybrid method is evaluated on three high-dimensional IoT intrusion datasets, and the performance is compared with some recent methods. The method shows more than 98.96% accuracy on all the datasets as well as outperforms other competing methods.