Study On Optimal Sensor Configuration Approach For Self-driving Cars

Environmental awareness is the core to realize self-driving cars.Its importance lies in making self-driving cars better simulate the perception ability of human drivers so as to understand themselves and surrounding environment.Environmental sensing system collects road surface and environmental information through various on-board sensors so as to provide decision-making basis for self-driving cars.In order to ensure that self-driving cars can drive smoothly and safely on the road,on-board sensors are placed at different locations on the car which need to provide comprehensive and accurate sensory information.It has become an urgent problem to be solved by the self-driving technology that how to arrange the sensor and how to set the external parameters of the sensor to make the environment perception system provide high-quality decision-making basis.The thesis focuses on the problem of optimal multi-type sensor configuration for self-driving cars.It including some problems,such as normalizing model of multi-type sensors into discrete unit for perception objects which are based on 2D evaluation indicators and 3D evaluation indicators,so as to calculate the evaluation indicators and optimize the sensor configuration.The problems such as design of model and solution method,numerical simulation verification,design of sensor configuration about simulation experimental platform were all integrally studied.The specific research work is as follows.On the analysis of the different types about sensors by considering the basis of the principle and main parameters,we extracted the common features of different types of sensors’ parameters,which are based on Gauss equation for constructing multi-type sensors’ area normalization model.In the context of 2D and 3D evaluation indicators,TIN and Octree theory are respectively introduced to construct the surface model and volume model of the perception object which are based on the point cloud data of the perception object.We described the surface and volume element characteristics of the perception object,and realized discrete surface and discrete volume.Based on the normalized model of sensor perception area and the discrete results,we have analyzed the coupling relationship between sensor perception area and the discrete unit of perception object,and have designed the perceived judgment method of discrete unit and the calculation method of sensor configuration evaluation indicators.Based on the above research results,the sensor configuration optimal model was established by taking the 2d evaluation index and 3d evaluation indicators as the objective function.The sensor installation position and angle have been used as the decision variables,and the intelligent algorithm which are based on particle swarm optimization was designed to solve the model.Numerical simulation results show that the proposed optimal method correctly and effectively.The influence of decision componentson 2D and 3D evaluation indicators is analyzed by thr convergence process of particle swarm optimization(PSO),and then we determined the influence weight of external parameters on sensor configuration evaluation indicators.Combined with two-dimensional evaluation indicators and three-dimensional simulation results under the evaluation indicators,the thesis reveals the same evaluation indicators about the optimal results under different distance perception consistency and optimal results for differences between two kinds of evaluation indicators.The differences between cloud under the orders of magnitude of perceived object model have an effect on the result of optimization.The experimental simulation platform for sensor configuration have been designed and implemented which are based on the approach of sensor optimal configuration by using two-dimensional evaluation indicators.The platform uses PreScan and SketchUp.Matlab,or Meshlab or other third-party software has also been used to design the sensor coverage area of the perceived object in the virtual scene.The perceived method which are quantized verifies the effectiveness of the approach of sensor optimal configuration.