A Multi-Stage Approach for Real Time Speed and Direction Estimations of Pedestrian
Sanun Srisuk · Procedia Computer Science · 2016
In autonomous vehicle system, the direction and speed of pedestrian must be estimated in order to guide the automatic breaking system. It can be used to protect the pedestrian walking on road. This paper presents a new algorithm for speed and direction estimations of pedestrian in real time. In the first step, the HOG features are extracted. The features are then fed to SVM for classification. Given the bounding box, the optical flow for each bounding box is calculated. Therefore, direction and speed are estimated from the optical flow. We achieved a very low time computation by using CPU-GPU co-design which is performed based on parallel processing. Experiments were performed to evaluate our system and compare several methods for time computation. Performance of our algorithm will also be demonstrated.