Highquality seeds can increase the germination rate. Directional seeding can make corn blades grow regularly and enhance ventilation and light energy utilization in the field. These two are necessary conditions to achieve directional and precision seeding for corn seeds. This paper provided a device and an image detection algorithm of corn seeds for directional and precision seeding. Those unqualified corn seeds were found from the corn seed samples and the corn embryo direction of the rest qualified seeds were determined using this detection algorithm. The corn seeds were transferred in two lines by conveyors. Two cameras at different locations captured the transferred corn seeds at the rate of 50 frames per second. The same seed in continuous images needed to be detected only once. So the repeated corn seed images were judged and not detected. The seed region and outer contour were detected. The shape characteristic parameters, such as the area of the seed region and the perimeter of the outer contour, were calculated. According to the color of the embryo of the corn seed as close to white and the endosperm was close to yellow, the furthest point of the white part from the yellow area center was determined as the tip point of the corn seed. The axis through the tip point and the centroid point was defined as the major axis. The axis through the centroid and perpendicular to the major axis was defined as the minor axis. The angle α between the major axis and the horizontal direction was calculated. And on this basis, the shape characteristic parameters such as the length of major axis, the length of minor axis, the lengthwidth ratio, the degree of symmetry and the duty ration, were calculated quickly. The 100 qualified corn seed samples were randomly selected as standard seeds. The above shape characteristic parameters were detected successively. A qualified range was determined according to the standard seeds detection result. The unqualified corn seeds with such shortcomings as asymmetric shape, small size, round shape, severe wormeaten and serious damage were found and excluded. The corn seed color image was transformed into saturation binary image. If the target area of this binary image was far below the average area value of the standard corn seeds, the seed was considered with severe mildew. Slight black mildew was judged according to the value of （R+G+B）/3 was small. Slight white mildew or slight damage was judged according to the value of B-R was small. At last, the orientation of embryo, up or down, was detected according to the characteristics which the embryo of corn seed was close to white and it mainly located in the major axis. Of course, the direction of the tip point, left or right, determined the angle of α. Experiments show that this algorithm can detect the qualification and the direction of corn seeds quickly. The time of detection for one seed is about 14ms. The accuracy rate of repeated corn seed detection is 95%. The accuracy rate of qualification detection is 96.1%. The accuracy rate of embryo orientation detection is 97.1%.