Written on Modified on
Mitsubishi Electric News
Supports detection and tracking of suspicious persons and lost children using multiple cameras.
Example of new AI technology’s repeated identification and tracking of human subject.
Mitsubishi Electric Corporation (TOKYO: 6503) announced today that it has developed an artificial intelligence (AI) technology that identifies individuals captured with multiple surveillance cameras in one-tenth time of conventional AI, using special compression technology to eliminate extraneous data and thereby achieve highly efficient and accurate reidentification, tracking and searching that is relatively unaffected by changes in camera angles or poses of human subjects. Mitsubishi Electric expects to provide the new technology in its Maisart®* lineup, including in new security systems for the detection and tracking of suspicious persons and lost children, and in the customer-service field for customer-flow analysis.
Proprietary compression technology reduces the time to reidentify individuals to one-tenth
- By analyzing the correlation of multiple features extracted from an image and removing information that is less important for reidentification, the AI compresses data volume to as little as one-tenth conventional data sets, thereby reducing the computational time for reidentification to one-tenth.
- To identify a specific person, the AI first extracts essential features from the person’s image, then compares the features using different camera angles to calculate similarities and determine if other images are those of the same person. In general, AI can extract several thousand features from a single image, so if the number of people increases then the calculation can become highly complex. With Mitsubishi Electric’s new technology, however, the AI analyzes correlations between multiple features and then removes unnecessary data while retaining a person’s salient features.
Fast, accurate tracking and searching unaffected by different camera angles and poses
- By compressing the data size, the AI speeds up processing of images from different angles and poses.
- Tracking and searching accuracy is improved by reducing the influence of different poses and angles.
- Tracking and search accuracy for single persons is improved by about 10% compared to conventional methods.
- The AI still works even if the person is wearing a mask or clothing similar to those of other persons nearby.