Embedded Files
Mohammad Shokrolah shirazi and Brendan Morris
Proceeding of the IEEE Intelligent Vehicle Symposium, (IV 2015), pp. 1264-1269, June, 2015, Seoul, Korea.
This work presents a framework to analyze traffic intersections by counting vehicles and pedestrians and assessing their behavior and safety. The major reason of developing this framework is to facilitate manual analyses from the video recordings by only providing detection files, typical paths, distance and conflict points. After tracking and recognizing paths, pedestrian and vehicle trajectories are extracted and their counting, behavior and safety information are estimated. Experimental results include estimated speed profile, turning movement count, waiting time, Time to Intersection (TTI) and Time To Collision (TTC) for two highly cluttered videos of Las Vegas intersections. The accuracy of 90%, 99% and 90% were obtained for vehicles waiting time, turning movement count and pedestrians crossing count. The semi-automatic system is a comprehensive solution for video based behavior, safety and counting analyses at intersections with high accuracy.
Goal
Developing a tool that takes detection files of vehicles and pedestrian for sequence of frames and provides tracks. In addition. distance to intersection and conflict points can be defined to provide safety analysis. The system facilitate manual observation of videos for behavior and safety analysis of intersections.
System overview
Abstract
Input
Typical paths, distance and conflict points along with detection files are provided for the tool.
Tracking and path recognition
A detection-track mapping matrix uses nearest global matching by greedy algorithm. Trajectories are recognized and counted using
Longest common subsequence (LCSS).
Behavior and safety results
The speed profile is estimated for different turning movements by accumulating speed of each category including Stop, Slow, Normal, Speeding. The probability density function of speed for turning right, left and going. The experiments show some interesting results. The going straight path usually gives drivers an opportunity to increase their speed through the intersection. As it is shown with green bar, all vehicles are considered speeding when going straight. This particular path travels from East to West (EW) and does not have any signal in the camera field of view. In contrast, vehicles in the North approach have a stop in view reducing average speed.
TTI and DTI plots have linear decreasing trends and this characteristic is shown below for 5 tracks. Since estimated velocity might
be noisy, the trend line by polynomial regression is utilized to better understand the approaching behavior (black curve). High TTI values including noisy or infinity value due to zero velocity are thresholded and replaced by100 to get the observable continuous trend line. One turning right vehicle, which stops for a long time before turning, shows different behavior (Track 58). First, its TTI value decreases during its approach to the intersection. However, since it encounters a red light, its speed decreases and the TTI value caps out at 100 when it stops. After it starts moving again, the typical trend line is reestablished.
Video
Page updated
Google Sites
Report abuse