Task Order 5327
Traffic Operations Research


Automated Collection, Comparison and Fusion of Data From
All Vehicle Presence Detectors in I-405 Detector Test Bed

Art MacCarley
Electrical Engineering Department
California Polytechnic State University, San Luis Obispo

Summary

The Caltrans I-405 Test Bed integrates and allows side-by-side testing of various types of vehicle detectors at two locations (plus one offramp). Duplex inductive loops are located below each overcrossing upon which are mounted video cameras above each lane. Mounting structures are provided for almost every type of non-intrusive vehicle presence detector. This project addresses the needs of the Caltrans I-405 Detector Test Bed by providing a fully automated means for the integration and synchronization of detection data from all current and anticipated vehicle presence sensors deployed and evaluated in the Test Bed. A digital image will be acquired for every car in every lane at the moment of detection by any sensor under test, by accepting trigger signals from all sensors, either in real or near-real time. The resultant JPEG image file for each vehicle detection event will be coded with a globally-accurate date/time/location code. These coded images will be available to all collaborating researcher in real time via direct access to a distributed network file system. In addition, we will provide a user-friendly computer application that will import and compare data generated by all sensors/detectors, and provide detailed raw and aggregated performance results, as well as a composite absolute ground truth reference set, and support for causal error analysis to reveal relative advantages and limitations of detection technology. Provision will be made for compatibility with a wide range of possible sensors, by accommodation of a number of communications and signaling methods, including hardwired, serial and network signaling options. This work will include full consultation with all Test Bed researchers and sensor vendors, directed toward the development of at least a local standard for signaling and communication interfaces and protocols for the Test Bed.

Methodology

The integrated data fusion system will be implemented as expanded capabilities of the existing V2SAT (Video Vehicle Signature Analysis and Tracking) system already deployed in the I-405 test bed. In it's current configuration, this system performs two primary functions:

  1. Presence Detection: A digital (Jpeg) image is acquired of every vehicle in every lane, as detected by the computer vision system. These images are both locally stored, and transmitted to a server in the UCI/ITS laboratory, where they are recorded and displayed in near-real-time (all 14 lanes from three sites). Recent data has verified that under acceptable lighting conditions, 99+% of all vehicles are detected, with fewer than 0.5% false detections.
  2. Tracking and Correlation Between Sites: The video image sequence for each vehicle is analyzed by the detector, and a small (120 byte) vector is immediately generated which uniquely characterizes each vehicle. These vectors are transmitted asynchronously to the correlation server (or multiple servers) where vehicle signatures from the two sites are matched in real time to re-identify and track the progression of every vehicle. The intended application of this capability is for tracking individual vehicles through a large multi-branch network, for measurement of travel times, original destination records, and microscopic flow model validation.
Aside from the required software-based capabilities for acquiring and analyzing sensor data, the central issue in the synchronization and fusion of a multiple vehicle detection sources is the means for real-time or pseudo-real time signaling by the sensor/detectors under test. We are aware that all sensors currently deployed in the test bed are network connected, or could easily be network connected by a local PC interface. This facilitates a fairly direct common signaling solution that requires no hardware modifications to the sensors/detectors: network-based signaling.

The signaling method we recommend utilizes the local area network (LAN) that couples all detectors at each site. IP socket-based communications is used. The signal from an external detector opens a socket connection on the V2Sat detection computer, which allows it to compare the client's IP address with the list of allowable addresses specified in the setup. V2SAT immediately captures and stores an image, and assigns the correct suffix letter to the jpeg image file. V2Sat only needs the IP address information of the external trigger device to correctly catalog an image. The external trigger device acts as a client and sends its IP address during socket connection initialization. V2Sat closes the socket connection immediately without transferring data. This short connection allows for very low latency and multiple triggering devices.

Trigger latency is an important consideration for this approach. Network delays are minimal when external trigger devices are located on the same local area network at the same site. Tests have verified that the latency associated with this signaling mechanism is typically 0.1 to 2 milliseconds, and at most 10 milliseconds depending on the local traffic load. This is well within the RS-170 field-to-field imaging period of 17ms, so that, in the worst case, a delay of one video field (half-frame or 1/60 second) is realized. In the setup of the V2SAT system, the user is allowed to specify a unique filename suffix character for unique identification of the JPEG images acquired for each detector.

Although we believe that signaling over the common LAN is the most flexible and extensible standardized communications method, alternative signaling methods can be implemented for those detectors that cannot provide a real-time signal over the LAN. Options include:
  • TTL-level binary signaling - V2SAT will be modified to accept accepts binary (0-5V) signaling using a hardware interface to bi-directional parallel port. Disadvantage is that this approach requires physical routing of wires between adjacent cabinets, and possibly provision for hardware interface by some sensors/detectors.
  • RS-232 serial real-time signaling - V2SAT accepts serial connections from detectors in adjacent cabinets. Requires physical routing of wires between adjacent cabinets, and some software modifications to V2SAT and the possible the external detector.
  • Time-stamp-based delayed signaling via the LAN. Accurate time synchronization is maintained between V2SAT and all detectors via reference to a common NTP time standard (either the V2SAT local time server or direct reference to the NIST Boulder time standard). A time and lane code is transmitted by the detector over the LAN to V2SAT, somewhat after the actual detection event. V2SAT will be modified to retain a FIFO queue of all images for a period of up to 5 seconds (300 images). The closest corresponding time coded image would be copied from the volatile circular queue and stored with the appropriate detector-designating filename suffix.
With images acquired and time/source coded for all sensors, the V2SAT system is easily modified to provide, in real time, cumulative statistics on sensor counts and potential detection errors. However, for a fully-capable data fusion application, the addition of user-friendly features which specifically identify discrepancies between detectors, and synthesize composite ground truth, would be useful. This is not necessarily a trivial computational task, since the exact time of detection for the same vehicle differs between different detectors, due to different positioning of detection zones and different mechanisms and thresholds of detection, e.g., change in inductive mass for analog loop detector, optical flow front edge for computer vision or laser-based detectors, and microwave-reflective target density for RADAR-based detectors. Temporal proximity with automatic adjustment of detector clock time discrepancies and/or latencies will be implemented to facilitate reliable comparison of records from all detectors, specifically identifying differences. An application and graphical user interface (GUI) will be implemented, building upon the existing V2SAT server software in the ITS laboratory, which would display comparative verification images for each detector side-by-side, to aid in human analysis of discrepancies between detectors. This GUI will facilitate rapid generation of absolute ground truth by requiring the human operator to consider only detection cases for which all or some subset of the detectors do not agree.