Text Content or SSML code:
Self-driving vehicles are capable of sensing the surrounding environment and driving safely with less or no human involvement. visual perception can be regarded as one of the main components in understanding the surrounding environment. Detection of static regulatory and navigational objects which include traffic signs, traffic lights, lane markings and road markings can be identified as an important task. In this project, we detect and identify static objects on the road such as traffic lights, road signs and road markings. Static object detection can be performed using various techniques. Communication methods such as DSRC, Cellular V2X or Bluetooth can be used. Information about static objects can be communicated to the vehicle this way. However, they may require extreme changes to the infrastructure. Pre-mapped static object information along with GPS location can also be used for the purpose. However pre-mapped information needs frequent updates when the environment changes. In realistic scenarios, relying on instant detection information from computer vision-based techniques will be the best alternative. We mainly collect data to create datasets that resembles the real world. We created a Traffic light dataset and a lane dataset covering various scenarios in Sri lankan context. We employ deep learning-based object detectors for detecting static objects. Each static object brings unique set of challenges. We train state-of-the-art neural networks for detection followed by optimization. Finally, algorithms are developed incorporating the deep learning models. Finally, we create an end-to-end system on a real vehicle to demonstrate the detection results in a qualitative fashion. The detection accuracies for static objects are quantified and summarized. Qualitative results are obtained by testing on real-world scenarios. The detection system is deployed in Robot Operating System (ROS) for streamlined integration with future implementations. An end-to-end system is implemented on a vehicle as Proof of Concept.