Traffic Anomaly Prediction System Using Predictive Network

Anomaly anticipation in traffic scenarios is one of the primary challenges action recognition. It believed that greater accuracy can be obtained by use semantic details and motion information along with input frames. Most state-of-the art models extract pre-defined optical flow from RGB frames combine them using deep neural networks. Many previous failed to pre-processed flow. Our study shows provides better detection objects video streaming, which an essential feature further accident prediction. Additional this issue, we propose a model utilizes recurrent network instantaneously propagates predictive coding errors across layers time steps. By assessing over representations pre-trained recognition given video, flows as redundant. Based on final score, show effectiveness our proposed three different types anomaly classes Speeding Vehicle, Vehicle Accident, Close Merging state-of-the-art KITTI, D2City HTA datasets.