| GISdevelopment.net ---> AARS ---> ACRS 2000 ---> Poster Session 1 |
Tracking Automobiles using Air-borne TLS (Three Line Scanner) Images
Ryuichi MURATA, Ryosuke SHIBASAKI
Center for Spatial Information Science (CSIS)
University of Tokyo
4-6-1 komaba, Meguro-ku, Tokyo 153-8505, Japan
Tel&Fax: +81-3-5452-6417
E-mail:rmurata@iis.u-tokyo.ac.jp
Ryuichi MURATA, Ryosuke SHIBASAKI
Center for Spatial Information Science (CSIS)
University of Tokyo
4-6-1 komaba, Meguro-ku, Tokyo 153-8505, Japan
Tel&Fax: +81-3-5452-6417
E-mail:rmurata@iis.u-tokyo.ac.jp
Keywords: TLS (three line scanner), air-borne sensor, ITS, object tracking
Abstract:
Acquiring traffic data such as number of cars, speed distribution, number of illegal parking cars accurately and quickly is needed for ITS (Intelligent Transport System). ITS is expected to mitigate traffic jam and to improve management of limited road resources. Currently, as a common practice, only limited traffic data are collected; vehicles are counted using roadside ultra-sonic sensors. Finding traffic accidents depends on the witness’s notice, patrol of road administrative officers, and monitoring with roadside cameras. TLS (Three Line Scanner) is an air-borne sensor consisting of three parallel one-dimensional CCDs mounted on the imaging plane. It obtains seamless high-resolution images (5-10cm on the ground) with three viewing directions (fore, nadir, aft) simultaneously mainly to generate 3D spatial data accurately. In addition, the high-resolution imagery can be applied to observe running cars, speed and parking cars on the street since TLS scans the same road surface with a time interval with approximately 10 seconds. This paper describes methodologies and the results of applying TLS imagery to the tracking of automobiles.
1. Introduction
ITS (Intelligent Transport System) society will spread in near the future. ITS is expected to operate traffic flows and to provide efficient road administration. But generally, acquiring traffic data depends on roadside ultra-sonic detectors, cameras, witness’s notices or patrols of road administrative officers. Traffic data acquired with such devices are basically point-based and fail to represent spatial distribution. It is quite necessary to develop a method of acquiring traffic data such as number of cars, speed distribution, finding accidental vehicles accurately and quickly over large areas. The high-resolution imagery of TLS (Three Line Scanners) can be applied to the observation of running cars, their speed and parking cars on the street because TLS can scan the same road surface and objects with a time interval with approximately 10 seconds.
2. TLS system
2.1 TLS principle
TLS (Three Line Scanner) is an air-borne sensor consisting of three parallel one-dimensional CCDs mounted on the imaging plane (Fig.1 and Fig.2). It obtains seamless high-resolution images (5-10cm on the ground) with three viewing directions (fore, nadir, aft) simultaneously mainly to generate 3D spatial data accurately with RTK-GPS and INS.
Fig.1: Method of TLS image Acquiring
Fig.2: Plain TLS image
2.2 TLS performance
Table1: system specification of TLS
2.3 TLS characteristics
Characteristics of TLS are summarized as follows.
We examined two automobile tracking methods using TLS gray scale image.
3.1 brightness difference from a pair of images
Since moving objects on the road are only automobiles, it is possible to track automobile objects using differential image from a pair of image covering the same area at the different times (Fig.3). We calculate object's speed by dividing the ground distance of centroids of each object by time difference. Fig.3 shows a result of tracking a white bus. But any automobiles whose color is similar to road surface are not tracked. In addition, we can't apply this method to track standing automobiles due to a traffic jam or an accident.
Fig.3: Differential image and a pair of image covered the same area at the different times
3.2 Template matching
In order to track stopping automobiles, I used template matching method using Hausdorff distance. First operation in the processing is edge detection from TLS image. Secondary is making a proper size rectangle template according to altitude of a platform or image resolution since all automobiles in aerial images have rectangular shape. Last one is affine transformation of TLS image and template matching. We narrowed down the operation image area to roads to improve the accuracy of the matching.
The distance between the each pixel of the template and the nearest TLS edge point defines the Hausdorff distance. We detected template position whose summation of each points Hausdorff distance is under set threshold as automobile object.
Fig.4: Detected objects by template matching
4. Conclusions
This study show that it is possible to track automobiles using TLS gray scale image. Differential image method depends on brightness of TLS image, so automobiles whose colors are similar to those of road surfaces were not tracked. It is difficult to detect edge in template matching method since the test image is experimental with no enough quality. Further studies are required to improve the accuracy of tracking objects using color image and high resolution image from the TLS.
References
Abstract:
Acquiring traffic data such as number of cars, speed distribution, number of illegal parking cars accurately and quickly is needed for ITS (Intelligent Transport System). ITS is expected to mitigate traffic jam and to improve management of limited road resources. Currently, as a common practice, only limited traffic data are collected; vehicles are counted using roadside ultra-sonic sensors. Finding traffic accidents depends on the witness’s notice, patrol of road administrative officers, and monitoring with roadside cameras. TLS (Three Line Scanner) is an air-borne sensor consisting of three parallel one-dimensional CCDs mounted on the imaging plane. It obtains seamless high-resolution images (5-10cm on the ground) with three viewing directions (fore, nadir, aft) simultaneously mainly to generate 3D spatial data accurately. In addition, the high-resolution imagery can be applied to observe running cars, speed and parking cars on the street since TLS scans the same road surface with a time interval with approximately 10 seconds. This paper describes methodologies and the results of applying TLS imagery to the tracking of automobiles.
1. Introduction
ITS (Intelligent Transport System) society will spread in near the future. ITS is expected to operate traffic flows and to provide efficient road administration. But generally, acquiring traffic data depends on roadside ultra-sonic detectors, cameras, witness’s notices or patrols of road administrative officers. Traffic data acquired with such devices are basically point-based and fail to represent spatial distribution. It is quite necessary to develop a method of acquiring traffic data such as number of cars, speed distribution, finding accidental vehicles accurately and quickly over large areas. The high-resolution imagery of TLS (Three Line Scanners) can be applied to the observation of running cars, their speed and parking cars on the street because TLS can scan the same road surface and objects with a time interval with approximately 10 seconds.
2. TLS system
2.1 TLS principle
TLS (Three Line Scanner) is an air-borne sensor consisting of three parallel one-dimensional CCDs mounted on the imaging plane (Fig.1 and Fig.2). It obtains seamless high-resolution images (5-10cm on the ground) with three viewing directions (fore, nadir, aft) simultaneously mainly to generate 3D spatial data accurately with RTK-GPS and INS.
Fig.1: Method of TLS image Acquiring
Fig.2: Plain TLS image
2.2 TLS performance
| CCD | Number of pixel/line | 10200 pixels |
| Pitch of pixel | 7 um | |
| Number of CCD | 3(monochrome), 1(RGB) | |
| Number of shading | 12 bit | |
| Lens | Distance of focus | 60mm |
| Angle of stereo | 21° | |
| Frequency | 500 line/second |
2.3 TLS characteristics
Characteristics of TLS are summarized as follows.
- Seamless high-resolution images (5-10cm on the ground, about aerial photograph class) can be obtained with three different viewing directions (fore, nadir, aft). Easy to create ortho-image from TLS images because TLS images are “line-projection” and less distorted than conventional aerial photo-images, which is point-projection image.
- Much less ground control point is needed since RTK-GPS and INS can estimate the sensor position and attitude accurately.
- TLS system records digital data directly, which enable users to easily process and analyze them on the real-time basis and help minimize processing errors.
- Multi-spectrum images can be acquired by replacing the filters and sensors.
We examined two automobile tracking methods using TLS gray scale image.
3.1 brightness difference from a pair of images
Since moving objects on the road are only automobiles, it is possible to track automobile objects using differential image from a pair of image covering the same area at the different times (Fig.3). We calculate object's speed by dividing the ground distance of centroids of each object by time difference. Fig.3 shows a result of tracking a white bus. But any automobiles whose color is similar to road surface are not tracked. In addition, we can't apply this method to track standing automobiles due to a traffic jam or an accident.
Fig.3: Differential image and a pair of image covered the same area at the different times
3.2 Template matching
In order to track stopping automobiles, I used template matching method using Hausdorff distance. First operation in the processing is edge detection from TLS image. Secondary is making a proper size rectangle template according to altitude of a platform or image resolution since all automobiles in aerial images have rectangular shape. Last one is affine transformation of TLS image and template matching. We narrowed down the operation image area to roads to improve the accuracy of the matching.
The distance between the each pixel of the template and the nearest TLS edge point defines the Hausdorff distance. We detected template position whose summation of each points Hausdorff distance is under set threshold as automobile object.
Fig.4: Detected objects by template matching
4. Conclusions
This study show that it is possible to track automobiles using TLS gray scale image. Differential image method depends on brightness of TLS image, so automobiles whose colors are similar to those of road surfaces were not tracked. It is difficult to detect edge in template matching method since the test image is experimental with no enough quality. Further studies are required to improve the accuracy of tracking objects using color image and high resolution image from the TLS.
References
- Shunji Murai, Yositaka Matsumoto, Li Xun, 1995, stereoscopic imagery with an air borne three-line scanner (TLS), ISPRS commission V, Intercom mission Workshop, pp20-25
- Michihiro Murao, Yasuyuki Matsushita, Katsushi Ikeuchi, Masao Sakauchi, 2000, Visualization of traffic Conditions for Drivers, UM3’2000 session D, p22