Detection and Tracking of Moving Objects Using 2.5D Motion Grids

Alireza Asvadi, Paulo Peixoto and Urbano Nunes
Institute of Systems and Robotics, Department of Electrical and Computer Engineering, University of Coimbra

Abstract:

The interest in intelligent vehicles which are able to perceive the environment and take appropriate actions has increased significantly in recent years. However, to increase their reliability and capability to operate in real-life environments they need to be empowered with a stronger representation and understanding of their surroundings. Awareness of moving objects is one of the key components for the perception of the environment. By detection, tracking and analyzing the moving objects, an intelligent vehicle can make a prediction about objects’ locations and behaviors and plan for next actions.
Here, a method is presented for detection and tracking of moving objects (DATMO) in dynamic environments surrounding a moving vehicle equipped with a Velodyne laser scanner and GPS/IMU localization system. The proposed method comprises two main parts: 1)- detection of moving objects; 2)- data association and tracking of moving objects. First, at every time step, a local 2.5D grid with excluded ground cells is employed for the representation of the environment. Each cell value is computed by averaging the heights of all measured points mapped in the cell. Along time, the generated grids combined with localization data are integrated into an environment model called local 2.5D map. In every frame, the last generated 2.5D grid was compared with the updated static model of the environment to detect the 2.5D motion grid. A mechanism based on spatial properties is presented to suppress false detections that are due to small localization errors. The object level representation is achieved by grouping the motion grids. The detected moving objects are tracked using Kalman filter. Gating and nearest neighbor strategies are used to determine which detected object goes with which track. Track management unit is used for the initialization of a new track for every newly detected object and for removing exited tracks when their location falls outside the local grid. The proposed method outputs a list of objects’ 3D bounding boxes and tracks. It extracts moving objects from 2.5D motion grids in the absence of a priori assumption on the shape of the objects which makes it suitable for a wide range of targets.

The architecture of the proposed algorithm for 2.5D grid-based detection and tracking of moving objects (DATMO).

sys

Result:

Click [Results] to download the video result. [Youtube link] 

Screenshots

Datasets:

The algorithm has been tested on KITTI dataset.

Paper:

1. A. Asvadi, P. Peixoto, and U. Nunes, “Detection and Tracking of Moving Objects Using 2.5D Motion Grids,” accepted in IEEE 18th International Conference on Intelligent Transportation Systems (ITSC 2015), Canary Islands, Spain, 2015.
[PDF] [presentation] [code] [result]

Code:

The code runs on Linux/Windows with MATLAB R2013a-R2015b 32-bit.
[code]