| Vision-based forward collision warning (V-FCW) system is one of the typical applications, designed with the aim to prevent incoming dangerous situations by warning the driver. Unlike the existing FCW system that uses radar sensor, V-FCW system uses low-cost cameras as obstacle sensing devices. Besides, it also has the advantages of module recognition and lane detection that other sensors don’t have. However, vision sensor presents great technical challenges in obstacle detection, and particularly the distance measurement. Based on these background and knowledge, a complete V-FCW system is proposed in this thesis, including novel algorithms on object detection and tracking algorithms, lane detection, object of interest (OOI) selection and time to collision (TTC) calculation. Simulation results are presented.The image processing (IP) module is weak at robustness and adaptability to different working conditions. Therefore, large amount of in-vehicle testing and experiments is necessary to test system performance. The testing should be taken under conditions including all the influence factors for V-FCW system, like weather, road condition, light condition, traffic condition, different objects, and many others. This leads to two obvious consequences:on the one hand, the conditions are so complex that many researches only do experiments under a few kinds of conditions; on the other hand, even under the same driving condition, the system performance needs to be tested repeatedly due to the unrepeatability of the road test. This increases the time and economical cost, as well as causing long development cycle.Moogle is data analysis and search engine developed depending on server. It has large data storage containing video data and the relative Ground truth files. Video data is acquired by a camera mounted on testing vehicles working under various driving conditions. Ground truth files contain information about measurements in true values like object coordinates. Making use of the data storage and searching function of Moogle, this paper established a testing and verification platform aimed at supporting V-FCW system development by way of performing testing and verification. The video data is used to replace the in-vehicle test under various conditions, which makes the testing available in laboratory environment on computer. In addition, all the data in the platform has advantages of repeatability, shareability and manageability, so it could be applied repeatedly. A data analysis system is developed in the platform to integrate Moogle and V-FCW system functionally. It consists of accuracy evaluation module and a comparator. Accuracy evaluation is designed to get the accuracies of V-FCW system algorithms for testing under different conditions. The comparator combines Moogle query and search with the accuracy evaluation process and eventually confirms the influence factor(s) that lead to the V-FCW system poor performance. Results of the comparator are utilized to improve V-FCW system algorithms. In the platform, input and output ports are defined among different components to transfer data, and data communication is realized via different ways.At the end of this paper, the V-FCW system algorithms proposed are applied on the large amount of video data in the data storage of the platform. Following the operating flowchart of the platform, evaluated accuracy are transferred back to the data storage, used for data sharing and management by. Moogle. Image processed results, pre-warning calculated results from video data obtained under several typical working conditions are presented, illustrated and analysed reasonably. After achieving accuracies for various conditions, an example to use comparator to query the data storage and sort influence factor(s) out is set up to show how the platform support V-FCW system development. |
PDF Full Text Request