Simultaneous Localization and Mapping with
Detection, Tracking and Classification of Moving Objects

Core Description
     Both simultaneous localization and mapping (SLAM) and detection, tracking and classification of moving objects (DTCMO) play key roles in robotics and automation. For certain constrained environments, SLAM and DTCMO are becoming solved problems. But for robots working outdoors, and at high speeds, SLAM and DTCMO are still incomplete. In earlier works, SLAM and DTCMO are treated as two separate problems. In fact, they can be complementary to one another. In this project, we present a new method to integrate SLAM and DTCMO to solve both problems simultaneously for both indoor and outdoor applications.
Sensors and Test Vehicles

                  Navlab8 with SICK PLS 100 and an Omni-Camera

               Navlab11 with SICK LMS 220
SICK Laser Scanner
PLS 100
LMS 220
Scanning angle
180 degree
180 degree
Angular Resolution
0.5 degree
0.5 degree
Range Accuracy
Less than 5 cm 
Maximum range
 50 m
 50 m
Collection Rate

Simultaneous Localization and Mapping for both indoor and outdoor applications
     The results of SLAM algorithm are shown below. The small black circles are the current scan data. The light blue dots belonging to the previous scans are stored in Stationary Object Map (SO-Map). The magenta line is the path of the test vehicle and the small magenta circles are the estimated positions of the vehicle in previous scans.


SLAM with Detection of Moving Objects
     The light blue dots and light orange dots respectively belong to SO-Map and Moving Object Map (MO-Map). The current scan contains black, red and green circles. The black circles mean stationary features. The green circles are new features since we don't have enough information to tell if they are moving or stationary. The red circles are moving objects.

Vehicle Detection

Pedestrian Detection
Our algorithms found both moving pedestrians and cars successfully.

SLAM with Detection and Tracking of Moving Objects



     As it can be seen from figures, SLAM with DTCMO can get rid of moving objects and get a more consist mapping result.

Global Localization and Intersection Detector

Digital Map

The results of SLAM with DTCMO
[We used ONLY range images to get this result.]

    Since localization errors are cumulative, we need to develop algorithms that can register the local localization with the global. We did this by recognizing landmarks in the environment and correcting the pose of the test vehicle to the known global positions of the landmarks. Since intersections are a large part of the urban road infrastructure, an intersection detector would provide many opportunities for eliminating the cumulative error. The circles in figure are intersections found by the current intersection detector.

Odometry vs.  SLAM with DTCMO
     Compared to the ground truth in the aerial map, the result of SLAM with DTCMO is not perfect yet. But it is still much better than odometry. Also by intersection detection, we can use some global consistent range scan alignment methods to improve SLAM with DTCMO.
C.-C. Wang and C. Thorpe, "Simultaneous Localization and Mapping with Detection and Tracking of Moving Objects", IEEE ICRA '02, 2002. [PDF 421 KB]
Jump to:

(C) 2001-2002 Chieh-Chih Wang
Last Updated: Jan. 29th, 2002 by Chieh-Chih Wang