Crash to Not Crash: Learn to Identify Dangerous Vehicles Using a Simulator

Abstract

Developing a computer vision-based algorithm for identifying dangerous vehicles requires a large amount of labeled accident data, which is difficult to collect in the real world. To tackle this challenge, we first develop a synthetic data generator built on top of a driving simulator. We then observe that the synthetic labels that are generated based on simulation results are very noisy, resulting in poor classification performance. In order to improve the quality of synthetic labels, we propose a new label adaptation technique that first extracts internal states of vehicles from the underlying driving simulator, and then refines labels by predicting future paths of vehicles based on a well-studied motion model. Via real-data experiments, we show that our dangerous vehicle classifier can reduce the missed detection rate by at least 18.5% compared with those trained with real data when time-to-collision is between 1.6s and 1.8s.

Classifying Near-Miss Traffic Incidents through Video, Sensor, and Object Features

“…The following sections use the grid size of (G h , G w ) = (8, 10). However, we confirmed that the classification performance did not increase even if the grid resolution was increased from (8,10) to (16,20). We consider that there are two reasons for this.…”

“…The first is that high resolution makes training difficult by increasing the number of attention parameters. To be specific, the number of attention parameters α g i, j is quadrupled if grid resolution (8,10) is increased to (16,20). This suggests that high resolution grids need more complex training than low resolution ones.…”

“…Ke et al [9] detect near-miss scenes using pedestrian detection from the front video captured by a vehicle-mounted camera; the distance between the car and the pedestrian is used to calculate the risk level of each frame. Kim et al [10] analyze front video of car crashes captured during car simulator trials on variety of roads for clarifying traffic characteristics of dangerous scene. While their model detect near-miss scenes using front video, they do not consider the classification of the near-miss incidents.…”

Risk Ranked Recall: Collision Safety Metric for Object Detection Systems in Autonomous Vehicles

“…However no objects are identified to pose a risk of an Imminent collision. This is an expected limitation of real world datasets and motivates the need for synthetic datasets that can include imminently unsafe situations [17], [18]. Fig.…”

Spatio-Temporal Action Graph Networks

“…Collision vision data is difficult to collect in the real world since these are unexpected rare events. [22] tackles the collision data scarcity by simulation. However synthetic data is still very different from real data, and hence training on simulation is not always sufficient.…”