Objectives  Previous research on intelligent video analysis methods were unable to detect the violation of vehicles and pedestrians in the video timely and efficiently.

  Methods  We propose an improved trajectory model (Tra-Model) from the geospatial view, which takes into account the attributes, spatiotemporal relationship and semantic information of dynamic objects in the geographical scene. In particular, we design a traffic violation behavior analysis method based on the rules of trajectory and geometric constraints. Our proposed method can detect three kinds of traffic violations in real time, including retrograde, violation of prohibited marking instructions and no entry. Taking a university as experimental site, we analyze the three types of violations of targets in different traffic scenarios.

  Results  The experimental results show that: (1) Compared with the existing dynamic target tracking algorithm, the accuracy is improved by 15.6%, and the extracted trajectory contains rich semantic information such as target type and trajectory sequence. (2) The accuracy of our proposed method for the violations analysis from multiple camera sequences is above 70%, which is better than other methods.

  Conclusions  The proposed method achieves global comprehensiveness, high precision and diversity of detection types for the analysis of various traffic violations in geographical scenarios.