It is one of the technical difficulties for the inspection robot of high voltage transmission line to identify obstacles effectively under the changes of outdoor lighting conditions. According to the robustness problem of obstacle identification under the low-light conditions, an intelligent method of obstacle recognition based on robot vision is put forward, so that the inspection robot can deal with various degrees of low-light changes. The obstacle images are processed by self-adaptive homomorphic filtering to reduce the influence of illumination partially. A obstacle image is divided into uniform sub-regions. The improved local direction pattern is used to extract the feature histogram vector of each sub-region image, and the sub-block feature histograms are concatenated one by one into the total histogram. The Chi distance is used to perform statistical identification. The experimental results show that this method can make the inspection robots effectively recognize the counterweight, suspension clamp and insulator string on the transmission line. Compared with other algorithms, the improved local directional pattern has a better anti-light interference effect and higher accurate recognition rate, and improves the robustness, adaptability and accuracy of image recognition during robot inspection. It greatly promotes the sustainability of inspection robots in the power industry.