Cooperative lidar object detection via feature sharing in deep networks

Abstract

Situational awareness as a necessity in connected and autonomous vehicles domain is a subject of significant research in recent years. The driver's safety is directly dependent on robustness, reliability and scalability of such systems. Cooperative mechanisms have provided a solution to improve situational awareness by utilizing communication networks. These mechanisms mitigate problems such as occlusion and sensor range limitation. However, the network capacity is a factor determining the maximum amount of information being shared among cooperative entities. A focus of this work is to reduce the network capacity requirements while maintaining the desirable object detection performance by utilizing and modifying the concept of feature sharing. Described here is a mechanism to further improve object detection performance by utilizing novel decentralized parallel frameworks and a new shared data alignment method to allow parallel and cooperative processing of sensed data in multiple locations yielding significant improvements.