Semantic road lane segmentation and recognition of pedestrian behavior for autonomous driving vehicles | |
| Author | Ravulapati, Kavya Samanvitha |
| Call Number | AIT RSPR no.CS-18-05 |
| Subject(s) | Structural analysis (Engineering) Image processing--Digital techniques |
| Note | A research study submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Computer Science, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Research studies project report ; no. CS-18-05 |
| Abstract | The computer vision and autonomous control communities are working towards fully automating the driving process in vehicles with the help of various tools such as object detection and traffic sign board recognition. Such tools play a major role in ensuring safety and preventing accidents. When designed with vigilance, we may create an automated, driverless society. One of the main obstacles to building such a system is to develop a good contextual model that can correctly understand what is going around the vehicle while it is driving. This may include detecting drivable areas and pedestrians. Even more challenging is to precisely interpret pedestrian actions and their latent meanings. Some possible pedestrian actions include running, waving hands, and looking. These are challenging actions for machine learning models to correctly interpret. The main reason behind this is that though their actions may be semantically the same, each pedestrian has his or her own way of performing the action. For example, the meaning of the action of waving hands may be to ask the driver to stop or to continue moving. Though such a task is instinctive to humans, it can be very ambiguous for the model to differentiate. The model should thus be trained well in order to interpret actions. Developing a system that accurately segments roads, detects pedestrians, and classifies their direction of motion in video streams during driving is the main idea of my research study. The system semantically segments the road lanes, detects bounding boxes for pedestrians on the road, and classifies the direction of motion of pedestrians into moving towards, away, left or right from the vehicle. Based on the decision taken after the classification of pedestrian behaviour, the system sends an alert message whether to stop, or to proceed with caution. Though the system performs well in the laboratory testing, it may not be ready for deployment in actual autonomous vehicles |
| Year | 2018 |
| Corresponding Series Added Entry | Asian Institute of Technology. Research studies project report ; no. CS-18-05 |
| Type | Research Study Project Report (RSPR) |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Computer Science (CS) |
| Chairperson(s) | Dailey, Matthew N.; |
| Examination Committee(s) | Mongkol Ekpanyapong;Phan Minh Dung; |
| Scholarship Donor(s) | Asian Institute of Technology Fellowship; |
| Degree | Research Studies Project Report (M. Eng.) - Asian Institute of Technology, 2018 |