Multi-Image Dynamic Fusion Technology for Applications of Visual Blind Spots Elimination

• Main Authors: Chang Cheng-Chuan, 張鉦詮
• Other Authors: Lin Yu-Chen
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/88185302425388830342

碩士 === 逢甲大學 === 自動控制工程學系 === 103 === This thesis presents the multi-image dynamic fusion algorithms and geometry reconstruction technology for eliminating the visual blind spots in forward driver safety assistance and underground parking lot safety applications. For forward vehicle safety assistance application, this type of accident usually occurs when the preceding vehicle decreases speed suddenly and even emergency stop, leaving insufficient time and space for the following vehicle to brake. The main reason why the following vehicle cannot respond quickly is that the blind spot resulting from the preceding vehicle fully or partially blocks the following vehicle’s field of view when the in-between distance is close. This thesis is based on a revolutionary collaborative preceding vehicle visual blind spots elimination and translucentization systems, which is utilizing two cameras respectively in the preceding and following vehicles and two video streams exchange, i.e. inter-vehicle communication, by wireless access in the vehicular environment / dedicated short range communication (WAVE/DSRC) devices. In addition, we utilize the feature matching and epipolar geometry method of front-and-back stereo configuration to find the relative pose between two cameras and then achieve sensor data fusion such that the visual obstructed region from the preceding vehicle becomes transparent notices. Not only does the proposed approach extends the driver’s visibility, but the resulting also improves the driving safety. For parking safety application, crimes and destruction in underground parking lot are common because underground ones have relatively much less surveillance than open ones. Therefore, the security for underground parking lot is always a serious issue that deserves special concern. Although the security cameras nowadays can help deter thieves from breaking into cars, and aid law enforcement in identifying criminals who did rob your car or hurt you. Nevertheless, occluding pillars often lead to the blind spots problem in surveillance systems. Here, we presents a specifically method to create an illusion of occluded pillars see-through visualization for supervisors at blind spots to increase safety in underground parking lot type environments. This illusion is achieved by transferring video information from a camera viewing the occluded area. The goal is to simply eliminate the blind spots resulting from the front occluding pillars in the host image and provide a plausible and appealing visualization aid to the monitoring staff at surveillance systems. This thesis is based on the multi-image dynamic fusion algorithms and geometry reconstruction technology to accomplish an innovative driver safety assistance systems and intelligent surveillance systems for eliminating the blind spots and see-through it. We believe that results of this research not only can effectively be enhance the driving safety and underground parking lot safety, but also can improve overall business opportunity in the global intelligent vehicle and smart city.