4.4 Experimental Results and Analysis
4.4.5 Vehicle Detection Comparison with Inductive Loop Detectors (Dataset D5)
Magnetic Sensor VS Inductive Loop
As discussed in section 3.2.1, the magnetic sensor HMC1051Z [3.14] is a passive device that measures the strength and direction of the Earth’s magnetic field. By contrast, the inductive loop is an active device: a 6’ by 6’ copper loop is excited by a 20 kHz voltage in order to creating a magnetic field [4.13]. Conducting material passing over the loop lowers the inductance and the change in inductance is measured by an electrical detector card used with the loops. Special high scan-rate detector cards used for vehicle classification sample the inductance at 140Hz.
Another difference stems from the fact that the magnetic sensor node measures a highly localized change. As the vehicle travels over the sensor, it records the changes in the fields caused by different parts of the vehicle. By contrast, the 6’ by 6’ standard loop geometry results in the “integration of the inductive signature over the traversal distance”, which can remove distinctive features from the inductive signature [4.14]. So the standard loop is not ideal for vehicle classification. Fig. 4.4.5.1 reproduces the inductive loop signatures of a
magnetic sensor node measurement provide much more detail than an inductive loop signature.
Fig. 4.4.5.1 Inductive loop signature from a pickup truck (left) and a passenger car (right):
Source [4.15]
Detection Performance Comparison
Fig. 4.4.5.2 Picture of the experimental setup for dataset D5
An experiment was conducted on Oct 6, 2004 on a local traffic lane of Martin Luther King Way (MLK) in downtown Berkeley from 1:20pm to 3:20pm (119min). A Sensys sensor node [3.18] was placed on the pavement in the middle of an inductive loop located at a section before a traffic light-controlled T-intersection. With help from the City of Berkeley, real time data were collected from both detection systems for detailed analysis. Moreover, video of the traffic was captured to be used as the source for ground truth.
# Detections Correct counts [%]
Video 791 100
Inductive loop 904 114.2857
Sensor node 791+7-7 = 791 791-14 = 777 98.23009
over-counting 7
Adj. lane 4
Double counted 2
Packet loss 1
under-counting 7
Level of measurements not high enough 3 Changing lane / Not along middle of lane 2
Motorcycle 1
Packet loss 1
Table 4.4.5.1 Summary of vehicle detection results of dataset D5
The vehicle detection results are summarized in Table 4.4.5.1. A total of 791 vehicles were observed in the recorded video. 904 detection events were generated by the inductive loop detector, which over-counted by 14%. This can be explained by the fact that this loop detector was used as a presence detector instead of a traffic counter, so it may not be well calibrated for counting vehicles. Since the traffic is quite heavy at MLK, a large number of stop-and-go cases were observed. This could be the main source of error for double
counting by the loop detector.
Fig. 4.4.5.3 Correlation of occupancy time for individual detections between the sensor node and loop detector in dataset D5
On the other hand, the sensor node shows a virtually 100% correct detection of the overall
7 under-counting instances were identified. The causes of these errors are also summarized in the Table 4.4.5.1. Even after subtracting these 14 cases, a very high successful detection rate of 98.2% was achieved. A plot of the correlation of occupancy time between the sensor node and loop detector is shown in Fig. 4.4.5.3. An overall correlation coefficient of 0.67 was obtained. This is caused by the difference in zone of detection as discussed at the beginning of this section. With this promising result from an urban traffic intersection with heavy traffic flow, one may confidently predict that such a robust wireless sensor network can detect vehicles and estimate speeds as well as a highly calibrated inductive loop detector.
Ch. 5 Vehicle Classification by Wireless Sensor Networks
Vehicle classification refers to the process and methodology to classify a vehicle signature in a specific format into a pre-defined vehicle class (e.g. passenger vehicle or truck). It is an important source of information for transportation design and management that can be used for many purposes. In pavement design and management, pavement life is estimated according to the distribution of vehicle types running over it, and this distribution may be used to schedule re-surfacing. In traffic safety research and implementation, the distribution of trucks in traffic is a critical design factor, because of its significantly lower speed and large size. In traffic control, signal priority can be given to vehicles classified as bus or an emergency vehicle. The distribution of vehicle types also provides valuable data analysis input to the prediction of highways capacity, assessment of the effectiveness of traffic legislation, automatic toll collection, weight enforcement strategies and environmental impact studies.
As with vehicle detection, a number of technologies were developed for classification.
Vision-based, inductive loop, microwave, piezo-electric and acoustic-based classification technologies are the common ones in use nowadays [section 2.1]. Vision-based
classification can achieve a correct classification rate higher than 90% [5.1]. The major limitation of vision-based classification is that the system’s performance is greatly affected by the environmental and lighting conditions. Classification stations with highly calibrated inductive loops are also in use [5.7] [5.8]. However, the infrastructure and maintenance costs of such a vehicle classification station are high. This makes deployment of such a system economical only at particular sites of interest, such as a toll plaza. On the other hand, vehicle classification by wireless sensor networks provides a much more flexible deployment configuration, making the system portable and, once again, scalable for large scale deployment.
In this chapter, the current classification technologies are first reviewed in section 5.1. The characteristics of magnetic vehicle signatures are studied in section 5.2. The data
processing and classification schemes for a platform with limited computation resources are discussed in section 5.3. And the experimental results and analysis are presented in section 5.4.