Adapt2detect: Cooperative learning framework for low-lit object detection

In this work, we propose an adaptive multisensory pipeline to detect an object in the low light scenario and call it Adapt2detect. In this pipeline, we integrate LiDAR and Camera to enhance the performance of the object detection framework. We observe that in well-lit conditions, RGB sensors are the preferred choice as they provide texture and shape information, whereas, in low light conditions, we seek light-sensitive sensors like LiDAR or RADAR. In this work, we address the tradeoff of latency and accuracy. LiDAR data in well-lit conditions are expensive as RGB data suffices the objective. However, to address the challenges associated with RGB data in low-light conditions, we integrate LiDAR into the pipeline. We use scene illumination as a threshold to decide the modality object detection and the corresponding object detection framework in action. We used a variant of the YOLO and Complex-YOLO network object detection strategy in line with RGB and LiDAR data, respectively. The efficacy of integrating the LiDAR with the camera lies in reducing latency and the demand for the high compute platform.