Observe locally, Infer Globally: a Space-Time MRF for Detecting Abnormal Activities with Incremental Updates

Jaechul Kim and Kristen Grauman


Problem Statement

Main Challenges

Overview of the Algorithm

Overview of the
      algorithm
                                                        Figure 1. System overview

Algorithm Details

- Motion representation

Histogram of optical flows (HOFs) are used at each local region in the video frame.


                                                                                 Figure 2. Histogram of optical flows

- Learning of motion patter

- Inference of abnormal activities in space-time MRF

- Incremental update

Results

- Examples of the detected abnormal events: [demo videos]

        Examples of the detected abnormal events        
                                              Figure 3. Snapshots of the detected abnormal events in subway station

- Advantages of the space-time MRF

                
     (a) Crowded but otherwise normal scene          (b) A person gets off the train,                         (c) A person abruptly stops walking,
                                                                             and then gets on the train very soon.                and changes his direction.

                     Figure 4. Examples showing advantages of space-time MRF formulation for abnormal events detection

Advantages from localization by dividing the clip into local regions

Advantages from localization by dividing the clip into local regions

- An effect of incremental learning

        
            (a) Frame number 75,381                             (b) Frame number 98,104

Figure 5. An effect of incremental learning. A movement from the right entrance to the gate is at first detected as “abnormal” at (a) because many of the most recent motions were along another path.
Later, the same type of movement is correctly detected as “normal” at (b) since several similar observations are accumulated in between (a) and (b).

- Quntitative results

quantitative_result

Conclusion

Publication

Jaechul Kim and Kristen Grauman, Observe locally, Infer Globally: a Space-Time MRF for Detecting Abnormal Activities with Incremental Updates, In Proc. International Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2009. [pdf] [data] [demo]