| Summary: | 碩士 === 國立勤益科技大學 === 電子工程系 === 97 === In many researches, the vision-based preceding vehicle detection is a very important subject in the advanced driver assistance systems (ADAS). The major feature of the intelligent vehicle safety system is collision warning (CW) and it also usually be applied to automatic steering system. Most researches of ADAS used personal computer (PC) to implement such system for the consideration of system efficiency. This thesis implements ADAS by using embedded system for the limitation of the space in a car.
The major function of our research is to find the preceding vehicles in the dynamic background. Our works are choosing the efficient image processing algorithms and implementing on an embedded system with limited resources.
The hardware platform adopted in our work is an INTEL XScale PXA270 SoC based platform which developed by Microtime Co., Ltd. The peripheral devices include: (1) USB web camera: the image capture device, (2) touch screen display: the GUI reveals device, (3) speaker: the voice output device. And the following open source codes are integrated to the vehicle detection system. There are an Embedded Linux 2.6.15.3 operating system, the video for Linux two (v412) open source code for receiving image streams from USB web camera, the MininGUI open source code for revealing plenty information in touch screen display, and madplay open source code for processing voice signal.
The multi-level image processing algorithm of the vehicle detection system we studied in this thesis include: (1) lane detection: for determining the region of vehicle detection, (2)vehicle detection: identify the vehicle footprint in some region of interesting which determined by lane detection procedure, (3)vehicle tracking: this process will be adopted for reducing the amount of operation once the preceding vehicle has been identified, (4)voice output: drivers will be alerted with a warning voice if they would not keep safe distance away from preceding cars, (5) ahead car distance record: the system will capture the picture and record the car distance information in the storage device since the safe distance can not be kept.
In this thesis, we use the pre recorded video to test our system. At present lane, the correct ratio of identification is up to 98%. At the passing car detection, the correct ratio of identification is up to 95%. The detection range is from 5 meters to 80 meters in front of the car. The range finding use static image to testing, and the effective range of range finding is 5 meters to 30 meters. The car distance in the 10 meters, the maximum error is 0.86 meters. The car distance in the 20 meters, the maximum error is 1.5 meters. The car distance in the 30 meters, the maximum error is 2.6 meters. From the results of our experiments, the interferences can be eliminated include: the text on the ground, the seam of bridge, the shadow of the overpass and shadows of trees. Our system can execute the vehicle detection and the vehicle tracking functions well in the daytime.
|
|---|
