Neural Networks (NNs) have gained enormous research interest in the recent years. However, several studies proofed that NNs are not reliable. In order to solve this dilemma, uncertainty estimation methods are introduced, which capture the confidence of the NN itself. Nevertheless, state-of-the-art uncertainty estimation techniques are very challenging to apply in safety-critical systems due to resource constraints. To overcome these disadvantages, this thesis targets to integrate resource-efficient uncertainty estimations in Convolutional Neural Networks (CNNs) for semantic segmentation in the automotive domain. Hence, this work makes use of a monitoring NN, called Feature Activation Consistency Checker (FACER), which is originally used for anomaly detection. In order to integrate uncertainty predictions into FACER, a novel approach is presented and integrated into a proposed pipeline. This proposed pipeline allows us a semantic segmentation as well as a corresponding uncertainty estimation. For a later comparison of the novel FACER with state-of-the-art uncertainty approaches, the suggested pipeline includes also two current uncertainty approaches based on Mutual Information (MI) and Entropy. In more detail, the proposed pipeline consists of two NNs: on the one hand, U-Net, a CNN applied for semantic segmentation and on the other hand the proposed FACER for uncertainty. Additionally, uncertainty estimations are integrated into U-Net, based on Monte Carlo Dropout and Entropy/MI. Furthermore, the second, proposed network, FACER for uncertainty, outputs uncertainty by supervising the U-Net. The output represents an uncertainty estimation for non-overlapping regions of the image, so-called patches. To obtain this uncertainty output, intermediate feature outputs of U-Net are extracted. The evaluation of the outputted uncertainty of FACER and of the reference methods is done based on the Patch Accuracy vs. Patch Uncertainty (PAvPU) score. Several experiments are carried out to evaluate the usability of the novel approach as well as finding strengths and weaknesses. These experiments cover simulations to investigate the influence of imbalanced training data and out-of-distribution (OOD) samples on uncertainty. In addition, tests to gain a visual understanding of uncertainty by detecting highly uncertain features are provided. Besides, the relation between the patch size and the PAvPU-score is investigated. Relying on the results, this work shows that the proposed FACER for uncertainty is a novel approach increasing the reliability of CNNs in the field of semantic segmentation. This novel approach has high potential since it not only provides a meaningful uncertainty estimation but also has a resource-efficient character which is essential for safety-critical systems in the automotive domain.