The accurate location information of crop and grass provides the basis for intelligent intrarow weeding robot and influences the effects of weeding directly. This work describes the development of hoe-crop-positioning optimization system based on the fusion of machine vision and accurate odometry sensing techniques for intra-row weeding robot. The hoe-crop-positioning optimization system consists of two subsystems: machine vision system and information optimization system. The machine vision system is able to identify a single crop among weeds based on color camera. The information optimization system is able to calculate the accurate distance from crop seedling to weeding actuator. The optimization principle was analyzed. C8051F020 microcomputer was used as the core processor of hoe-crop-positioning system. In order to improve the accuracy of velocity measurement and anti-vibration interference, an accurate odometry collecting logical circuit was designed for frequency multiplication, phase demodulation and the other logic processing of the speed pulses. A double threshold optimization algorithm was used to judge the weights of real-time odometry data and forward distance from vision system, and to predict the discrete forward displacement of the machine in time. System assessment experiment was carried out based on a mobile hydraulic power pack platform. The experiment results show that the hoe-crop-positioning optimization system can effectively filter the error and instability information of machine vision. Under the static experiment the croppositioning error was 6.7mm, 10.3% less than that in earlier time. The weeding simulation experiment showed that the system can reduce the possibility of redundant rotation of hoe and improve the stability of weeding system.