# AUTOMATION OF PEDESTRIAN TRACKING IN A CROWDED SITUATION USING ANGLED UNCALIBRATED CAMERA Saman Saadat

Sovannarith Heng Kardi Teknomo

Ateneo de Manila University

Overview
•Introduction •Problem •Method •Results •Conclusion

Introduction
• The importance of measurement of individual pedestrian trajectories E.g. In Business, Evacuation, and Traffic … etc. • Pedestrians’ movement influencing safety concerns in crowded locations. E.g. such as sports venues, public transport station, and religious events … etc

Pedestrians in crowded areas

Literature
• Kelly (2007) states that most of the person detection technique works better in well controlled environments such as laboratory setting with a small amount of people.

Literature
• Kelly, O’Connor, and Smeaton (2009) shows that other techniques require more than one camera with overlapping fields of view in order to cope with the pedestrian segmentation problem in the a dense crowd. • We need to use extra cameras which cost much.

How can we track all pedestrians in the scene using a single camera and save the data of tracking as a NTXY table?

Research Problem

Fig. 1 Walking experiment at Ateneo de Manila

Method in

Automation

Object Detection
1. Split the video sequence into three channels of Red, Green and Blue (RGB) 2. Subtracting green channel from the red channel 3. Threshold of 70 (out of 255)

i(x, y) = r(x, y) – g(x, y)

1 if i ( x, y ) > φ f ( x, y ) =  0 otherwise

Object Detection (cont.)

Object Detection (cont.)

Object Analysis
• Run the particle analysis in ImageJ to: – Detect all objects in a frame by using connected component algorithm – Analyze each detected object for their area and centroid coordinate • The result will be a table with ID, coordinate and slices number of each object.

Object Analysis (cont.)

Fig. 4 Results of particles analysis of the binary image

Fig. 5 Tabulated result of particle analysis

Object Analysis in ImageJ

Flowchart of Tracking Algorithm

Object Tracking

Searching for the neighborhood

Object Tracking (cont.)

Comparison with Manual Tracking

Summary
Research Question:
How can we track all pedestrians in the scene using a single camera and save the data of tracking as a NTXY table?

1. Red hat detection 2. Blob identification 3. Frame by frame tracking of trajectories 4. Conversion to NTXY

Conclusion
• We were able to create algorithms to detect, to analyze and to track pedestrian from uncalibrated angle camera for outdoor crowded situation • The system was able to recognize pedestrians at reasonable rate of misclassification • 86.7% correctly detected (red hat) • 13.3% maximum error (any other red object) • Accuracy can further be improved by taking into account the size and shape of the blob as it moves (distinguishing red hat from other red objects)

Thank you…