Autonomous anomaly detection is a fundamental step in visual surveillance systems, and so we have witnessed great progress in the form of various promising algorithms. Nonetheless, majority of prior algorithms assume static surveillance cameras that severely restricts the coverage of the system unless the number of cameras is exponentially increased, consequently increasing both the installation and the monitoring costs. In the current work we propose an anomaly detection system based on mobile surveillance cameras, i.e., moving robots which continuously navigate a target area. We compare the newly acquired test images with a database of normal images using geo-tags. For anomaly detection, a Siamese network is trained which analyses two input images for anomalies while ignoring the viewpoint differences. Further, our system is capable of updating the normal images database with human collaboration. Finally, we propose a new tester dataset that is captured by repeated visits of the robot over a constrained outdoor industrial target area. Our experiments demonstrate the effectiveness of the proposed system for anomaly detection using mobile surveillance robots.