The Study On The Vision-based Detection Method For Automotive Safety Glass Fragmentation

The safety glass is one of the most important parts of the car. Its quality directly relates to the safety of users. In order to protect the consumers, China government practices some compulsory detection regulations for the safety glass. Fragmentation test is one important part of these regulations. By detecting the glass fragments of broken safety glass, the possibility of injury to the drivers and passengers caused by the broken glass can be evaluated. Up to now, the result of fragmentation test is determined completely by workers, which is inefficiently, unstable and easily affected by the subjective faults. In order to improve the test efficiency and accuracy, a practical automotive safety glass detecting system is developed in this thesis by applying the digital image processing technology.This thesis systematically studies the detection method of automotive safety glass fragmentation, including studying the edge extraction of the safety glass, giving the method of eliminating glass stripes caused by the heating wires or the antenna stripes on the car windscreen, and developing a practical automotive safety glass detecting system in the last.First, a new method of extracting edge of the safety glass fragments is proposed based on watershed algorithm and dynamic threshold segmentation algorithm. And accurate extraction result is realized.Secondly, an algorithm of extracting and eliminating the stripe is designed to eliminate the effect caused by the heating wire or the antenna stripe on the car windscreen,.Thirdly, by analyzing the needs and performances, an automotive safety glass detecting system which has both manual and automatic mode is developed. The manual mode simulates the detecting process by workers and identifies the number of glass fragments automatically. The automatic mode, as we know, is to realize the whole detecting process automatically.The edge extraction algorithm and the method of eliminating glass stripes proposed by the thesis behave well in the experiments of the safety glass fragmentation detection. The software system initially achieves the use request.