Towards Robust Perception Depth Information For Collision Avoidance

Fatal accidents can be avoided in autonomous vehicles by detecting the obstacles, and distance estimation accurately and in real-time is crucial. Still, most of the object detection approaches fail to detect debris and other roadside obstacles. The drawbacks of the object detection approaches are countered using active sensors such as LiDAR, RADAR, and SONAR, but expensive. This study introduces a modified architecture to estimate a depth map based on an unsupervised learning framework. Furthermore, integrating depth map information directly into autonomous vehicles is a slightly critical and challenging task. This paper presents a concept for predicting a potential collision based on the estimated depth map. For this purpose, we designed and developed an efficient and robust algorithm to understand the color encoding used in representing the depth map. Thereby, the possibility of collision regions in real-time identified using predefined threshold values for the color encoding of the predicted depth map. This approach emphasizes integrating the estimated depth map with the level of comprehension of the situation awareness to enhance the ability to recognize and process predicted hazards in the environment. It facilitated the possibility to avoid a dangerous situation in real-time capabilities.