Violence Detection in Video by Large Scale Multi-Scale Local Binary Pattern Dynamics

We propose a violence detector based on the dynamics of new multi-scale local binary pattern histogram features (MSLBP), that generate high-dimensional space (20 480 dimensions), trained on linear SVM. The results show that MSLBP dynamics can represent violent scenes. Even if the current scores are medium, we think that this simple visual only model can be greatly improved with some futher optimization out of the challenge schedule.