In the field of mobile robotics and intelligent vehicles, a key requirement for basic functionalities, such as collision awareness systems, is the ability to perceive and model a relevant representation of the environment where objects are moving. This crucial task is a challenging step in terms of accuracy, complexity and uncertainty management. The objective of this paper is to describe the multi-sensor Bayesian perception approach which has been developed by Inria 1 in the scope IRT Nanoelec 2 , implemented and tested on the IRT experimental platform including an equipped Renault Zoe vehicle. These systems can be adapted to off road environments such as agriculture fields or construction area where workers lives are jeopardized by heavy moving machineries. Embedded multi-sensor Bayesian perception algorithms have been developed on several experimental hardware in cooperation with CEA 3. For the purpose of computing accurate representation of the dynamic environment, these systems have been implemented on our intelligent vehicle, as well as on portable connected perception units. Shared perception have been implemented using Vehicle to Vehicle (V2V) and Vehicle to Infrastructure (V2I) communication protocols. Experimental results are presented and discussed in the paper.