Computer Vision based Accident Detection for Autonomous Vehicles

Abstract

Numerous Deep Learning and sensor-based models have been developed to detect potential accidents with an autonomous vehicle. However, a self-driving car needs to be able to detect accidents between other vehicles in its path and take appropriate actions such as to slow down or stop and inform the concerned authorities. In this paper, we propose a novel support system for self-driving cars that detects vehicular accidents through a dashboard camera. The system leverages the Mask R-CNN framework for vehicle detection and a centroid tracking algorithm to track the detected vehicle. Additionally, the framework calculates various parameters such as speed, acceleration, and trajectory to determine whether an accident has occurred between any of the tracked vehicles. The framework has been tested on a custom dataset of dashcam footage and achieves a high accident detection rate while maintaining a low false alarm rate.

Traffic Accident Detection via Self-Supervised Consistency Learning in Driving Scenarios

“…Compared with frame-level deep Vision-TAD methods, object-centric approaches concentrate on object-level temporal consistency and follow the object detection and tracking stages to generate the trajectories, where various detectors (e.g., mask-RCNN [67], FasterRCNN [68], YOLOv4 [69], YOLOv5 [70], etc.) and many object association approaches (SORT [71], DeepSort [72], Kalman Filter [69], [73], Hungarian algorithm [69], etc.)…”

“…Before the accident determination, trajectory features are encoded by various kinemetric clues (e.g., the velocity and acceleration [67]) or deep features. Then, common formulations learn the dominant trajectory feature clusters and classify the trajectory feature set for finding accident ones, where the accident is determined by the distance between the trajectory feature with the learned cluster centers.…”

Mask R-CNN Approach to Real-Time Lane Detection for Autonomous Vehicles

“…Lane detection is a crucial component of perception systems for autonomous vehicles, as it ensures safe navigation and adherence to traffic rules. Over the years, various approaches have been proposed for different practical tasks and datasets ranging from traditional methods to deep learningbased techniques [9][10][11]. In this section, we review some of the key works in lane detection, considering traditional approaches, CNNs, RNNs, UNet, and other deep learning models (see Table I).…”

Trends and Emerging Technologies for the Development of Electric Vehicles

“…Sensing range: 0-120 m. Multi-cameras can approximate human vision, and stereo cameras can even obtain depth information about the environment. The cameras can detect colors and fonts, which makes it possible to understand the semantic information embedded in traffic signs and stoplights, a capability that other sensors do not have [125]. They can perform as a redundant system in case of failure of other sensors, thus increasing the reliability of perception and system safety.…”