19-03-2012, 10:52 AM
System Interface for an Integrated Intelligent Safety System
(ISS) for Vehicle Applications
Abstract: This paper deals with the interface-relevant activity of a vehicle integrated
intelligent safety system (ISS) that includes an airbag deployment decision system (ADDS)
and a tire pressure monitoring system (TPMS). A program is developed in
LabWindows/CVI, using C for prototype implementation. The prototype is primarily
concerned with the interconnection between hardware objects such as a load cell, web
camera, accelerometer, TPM tire module and receiver module, DAQ card, CPU card and a
touch screen. Several safety subsystems, including image processing, weight sensing and
crash detection systems, are integrated, and their outputs are combined to yield intelligent
decisions regarding airbag deployment. The integrated safety system also monitors tire
pressure and temperature. Testing and experimentation with this ISS suggests that the
system is unique, robust, intelligent, and appropriate for in-vehicle applications.
Keywords: system interface; intelligent safety system; ADDS; TPMS; integration
1. Introduction
In any vehicle, the presence of intelligent safety implies an active system that promotes safety,
security and driving comfort [1]. However, to meet high expectations for control and safety, a large
number of individual safety systems are required [2,3]. This has led to concern over safety issues and
has resulted in a need for integrated ISSs that feature effective new technologies, characterize safety
issues and provide solutions for monitoring, detecting, and classifying impending crashes or unsafe
OPEN ACCESS
Sensors 2010, 10
1142
driving conditions. The ISS should warn the driver, improving his or her ability to control the vehicle
and thereby preventing accidents [4,5].
In the past, many researchers have adopted approaches towards individual safety issues such as the
detection, classification and location of occupants, vehicle crash detection and severity analysis,
TPMS, etc. For example, in [3,6,7] occupant detection and characterization parameters were studied to
improve the safety and comfort features for all occupants. However, the challenge of detecting and
distinguishing a particular class of occupant from all others remains daunting. Despite the success of
some of these systems, occupant detection and classification involving human subjects and non-human
objects still poses a number of challenges, and further progress remains necessary for addressing
changes in illumination, image scale, image quality, expression and pose. Sensors for data acquisition,
real time implementations, and operations should also be studied further [8].
Crash detection is a helpful concept in preventative safety, preventing accidents, collisions and
minimizing human injury when an accident occurs [5,9]. In the past, practical crash detection has not
been widely discussed, and researchers have mainly considered the theoretical aspects of crash
analysis using traditional engineering principles [10-12]. Recently, several attempts have been made to
develop an automated system to detect vehicle crashes, vehicle types and crashes under various
conditions such as during and after heavy downpours, driving at dawn or at dusk, sunlight reflections,
vehicles driven at high speeds and out of position. These are considered as high risk problems that
require dedicated solutions. Before now, automated solutions were not feasible or did not perform
sufficiently robustly for everyday use [12]. If these problems are not addressed properly, they will
continue to serve as obstacles to the implementation of intelligent crash detection systems. Therefore,
the national highway traffic safety and administration (NHTSA) and other road related safety
authorities have called for the mandatory consideration of crash detection and analysis as a key safety
issue [13,14].
Similarly, TPMS performance is important for improving both driving experience and vehicle
performance [15]. Vehicles without TPMS features have more safety problems. To date, a number of
TPMS have been widely investigated in order to solve the problems. Major concerns include limited
lithium battery lifetimes, malfunctioning of the electromagnetic RF transceiver unit, echo-based noise
due to broadcasting pulse responses, inadequate sensor capabilities, and low robustness in harsh
environments encountered during vehicle operation [16,17]. In particular, appropriate sensors for
different TPMS applications are still under investigation [18,19]. Accordingly, in the TREAD act the
NHTSA legislated that, after 31 October 2006, all vehicles in the United States must offer TPMS as an
option [13,20-22].
The fields of intelligent vehicles and their applications are rapidly growing worldwide, as is interest
from the automobile, truck, public transportation, industrial, and military sectors. The ISS offers the
potential to significantly enhance both safety and operational efficiency [23,24]. Increasing demand for
quality ISS solutions has driven the design of robust safety technologies, the study of safety issues and
the provision of solutions that involve monitoring, detecting, and classifying impending crashes or
unsafe driving conditions, and by warning the driver, improving his or her ability to control the vehicle
and prevent an accident [3]. In intelligent transportation systems, ISSs use sensing and intelligent
algorithms to understand the vehicle’s immediate environment, either assisting the driver or fully
controlling the vehicle. However, state of the art studies of prototype integrated ISSs suggest that there
Sensors 2010, 10
1143
remains a gap between many of these inventions and actual marketable products [25,26]. For such
products or inventions to be effective, we believe that a robust system is required for interfacing a
given ISS prototype implementation with other vehicle components. Therefore, in this paper, we
highlight the importance of good system interfaces, and demonstrate their use in the development of an
innovative integrated ISS. This ISS can identify major hazards and can assess the associated risks in
various environments where more traditional tools cannot be effectively or efficiently applied. Safety
devices provide data to the ISS that are useful for the development of ADDS and TPMS. This paper
successfully integrates and develops an advanced ISS with such features as occupant detection,
classification and positioning, vehicle crash detection, crash severity analysis, tire pressure monitoring,
and analysis of other hazards.
2. System Integration
The main motivation behind system integration is to reduce the management costs of individual
safety systems, which translates into improved system performance. Further, system integration
reduces the programming resources necessary to meet response time requirements and to maintain a
high service quality. Performance tuning is accomplished by obtaining information about how much
time is spent on each safety measures of a distributed transaction, as well as information about the
delays that might occur in the overall integration process. The integrated ISS aims to provide
heterogeneous workload management concepts and functions to addresses safety issues based on
diagnoses in a developed platform using collected monitoring data. The hardware platform identifies a
set of hardware objects, each associated with a processor. The system interface provides a high level of
interfacing between software running on different processors that control the hardware. The major
tasks of the integrated ISS include performance characterization, problem determination and real
workload data monitoring of distributed safety issues that are incorporated into the system. The
proposed ISS deals with safety and comfort concerns in the modern vehicle, including tire pressure
monitoring, occupant detection, crash detection and vehicle position monitoring. This integrated ISS
gathers environmental data using a set of sensors, collected the data through acquisition processes,
eventually reacts through a CPU, and finally outputs information on safety issues to a LCD
display unit.
3. Algorithm and Methodology
Methods and algorithms for the ISS were developed for ADDS and TPMS, which involved the
individual algorithms for occupant detection, classification and position based on weight sensing and
image processing as well as for vehicle crash detection. For classification purposes, weight
measurement data are used with additional logic elements. For example, when an adult occupant is on
a seat, the adult logical variable is set to true, child and non-human object logical variables are set to
false, the algorithm classifies the occupant as an adult and displays relevant output data on the monitor.
For position detection, we calculated the centroidal distances of Fx and Fy as follows [27]:
( 1 2 3 4)
( 1 2 3 4)
F F F F
Fx x F F F F
+ + +
− + − +
= (1)
Sensors 2010, 10
1144
( 1 2 3 4)
( 1 2 3 4)
F F F F
Fy y F F F F
+ + +
+ − −
= (2)
where F1, F2, F3 and F4 are weights as detected by the four weight sensors, while x and y indicate the
distances from the centre to the sensor in the x and y directions, respectively. The output of the
calculations involving Fx and Fy gives the position of the occupant.
The algorithmic approach based on image processing for the detection and classification of
occupant, non-human object and non-object is shown in Figure 1.
Figure 1. Neural network algorithm for occupancy detection.
The proposed system is a combination of a fast neural network (FNN) and a classical neural
network (CNN). The FNN analyzes any image for which a positive detection has been made, including
false positive identifications. CNN is used to verify the region of detection. Under the proposed system
architecture, the FNN extracts a sub-image from the test image to distinguish between correct object
and false detections. Post-processing strategies are applied to convert normalized outputs back into
consistent units and to eliminate detection overlap. Initially, we assumed that the FNN could be
confounded into false detection by variable lighting conditions. For example, illuminating the side of
an object changes its overall appearance. To solve this problem, an automatic linear function was
initially used to adjust image intensity values using histogram equalization or lighting corrections.
However, neither method was found to be suitable. Rather, an alternative method was used that
Sensors 2010, 10
1145
employed an object verification procedure using the CNN. This CNN object verifier helped reduce
false detection rates. This combined network was capable of higher detection accuracy and exhibited
better computational efficiency compared to a single network, which was unable to fully eliminate the
false detection problem.
The change in vehicle velocity, Δv(t), is an essential parameter for crash detection and is used here
in the development of our crash detection algorithm. Δv(t) is obtained by integrating the acceleration
signal [28] as shown below.
Δ v (t ) = ∫ a (t ) dt = Aω 2 ∫ cos (ω t + δ ) (3)
A suitable vehicle velocity threshold, Vth, is required to facilitate decision making as to whether or
not a crash has effectively occurred. This threshold value Vth can easily be determined from the lowest
effective speed of a crash as defined by NHTSA, which is 22.54 km/h. To detect a crash, the following
algorithmic steps were used:
(i) If Δv(t) ≥ Vth, then output = ‘1’; DECISION: Effective crash is detected.
(ii) If Δv(t) < Vth, then output = ‘0’; DECISION: Effective crash is not detected.
The change of velocity Δv(t) over a period of time T can easily be computed for this decision since
the integral over the noise component is approximately zero. The circuit for computing Δv(t) can be
designed using systolic architecture to determine the real-time speed. The systolic design processes the
output data in the systolic array for required operation of the optimal detection state. The detection
state is fed into a data acquisition card for system development.
For the TPMS, a threshold check algorithm is used to acquire data from the sensors. For the
threshold check, the DAR is preloaded with a threshold value while in standby/reset mode to detect
whether the pressure or temperature has crossed a particular level. The receiver module is capable of
receiving both on-off keying (OOK) and frequency shift keying (FSK) inputs through a UHF receiver
that communicates with the CPU via an SPI. The UHF receiver detects and demodulates the signal
through a Manchester-encoded bit stream, sending the important data out to the CPU. Data is then
monitored in the display unit. The TPM and receiver modules are loaded with a simple software
program to improve the functionality of the hardware. The assembly code for the TPM module is
written using the “WIN IDE” integrated development environment and is programmed into RF2 using
a programmer board that transmits data to the receiver module. The receiver module communicates
with the UHF receiver using a Turbo C compiler under DOS. The “TPMReceiverModule” function is
created in the main interface program UKM.dll to monitors pressure and temperature data transmitted
from the TPM receiver through the SPI connection to the CPU.