Efficiency gains from driving point cloud format modification

背景：
The original custom point cloud structure is being read、传输、Decoding the entire time period is very time consuming,Therefore, the point cloud structure is upgraded,Improve overall runtime.

protobuf字段类型介绍

修改前：

message PointXYZI {
  optional float x = 1 [default = nan];
  optional float y = 2 [default = nan];
  optional float z = 3 [default = nan];
  optional uint32 intensity = 4 [default = 0];
}

message PointCloud {
  repeated PointXYZIT point = 1;
}

修改后：

message PointXYZI2 {
  repeate float x_array = 1 [default = nan];
  repeate float y_array = 2 [default = nan];
  repeate float z_array = 3 [default = nan];
  optional bytes intensity_array = 4 [default = 0];
}

message PointCloud {
  optional PointXYZI2 point_array = 1;
}

修改点：

• PointXYZIconsists of repeated data structures,Change to repeated points.

意义在于：

• Originally, reading and writing requires multiple reading and writing of this structure,Now instead direct one-time storage points,Store all points in this struct.
• 而且将intensityand other parameters are changedbytes,Analyzed later,Reduced overall point cloud package size,减少传输时间.

protobuf修改字段为bytes需要考虑一些问题：

• Computer major and minor sequence problems;Different systems will be different,read outbytes会不同.If other languages ​​callC++的库,Not parsed through its own language,能解决问题?
• Multilingual compatibility issues,解析问题
• Field alignment issues

Currently taking time to analyze：

经验：

• By reducing the field type,提高传输速率,空间换时间;
• Write-once replaces write-many,解决时间,This part of the time saved accounts for the most part.