Comparative Analysis of Two Stage and Single Stage Detectors for Anomaly Detection
Linu Shine, C. V. Jiji · 2021
Detection of traffic anomalies involves detecting stalled or crashed vehicles. This vehicle becomes stationary and hence becomes a part of the background. The performance of anomaly detectors depends on the performance of object detectors on background images. Usually, the object detectors are compared for the performance on original frames. In this paper, we compare the performance of the two-stage detector and single-stage detector on background images. Faster Region-based Convolutional Neural Network(FRCNN) with Inception and ResNet network are compared with single-stage detector YOLO V3. The performance analysis shows that YOLO has a higher F1 score compared to FRCNN.