Abstract:
Visual analytics can bridge the gap between computational and human approaches for detecting traffic anomalies near the round-about, making the data analysis process more transparent. The main problem with anomaly detection is the unavailability of anomalous data as they do not occur frequently. Moreover, there is a variety in the characteristics of the same real-time scenario. A solution is proposed to handle these issues using unsupervised learning which mainly encloses the detection of vehicles, road, roundabout violations, and traffic jams near the round-about. Initially, vehicles, roads, and round-about are detected based on Histogram of Gradient (HoG) feature descriptor and color information. Then, optical flow is applied to the detected vehicles for obtaining their trajectory information. Finally, the round-about violation is detected based on the position and angle between the centre of the round-about and the vehicle trajectory. Additionally, the number of vehicles near the round-about is computed for detecting the traffic jam. The effectiveness of the proposed algorithm is demonstrated over 30 real-time traffic videos. Some additional comparative studies are done over the benchmark data YouTube8M. Through an extensive study, it can be concluded that our proposed algorithm is superior to some state-of-the-arts.